diff --git a/-9E1T4oBgHgl3EQfUwOX/content/2301.03094v1.pdf b/-9E1T4oBgHgl3EQfUwOX/content/2301.03094v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6a73425d4e867b1ea2a6a1443368edb8909a45cd --- /dev/null +++ b/-9E1T4oBgHgl3EQfUwOX/content/2301.03094v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc620aa0d0e4c82ab825655d8b0cf8a401e1410c6f54723e6eb37bc07f98cfdc +size 1585378 diff --git a/-9E1T4oBgHgl3EQfUwOX/vector_store/index.faiss b/-9E1T4oBgHgl3EQfUwOX/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..bf18c2e882fed70401ea12259c86a0bc91c8b568 --- /dev/null +++ b/-9E1T4oBgHgl3EQfUwOX/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:464d060f6719d7e6066a613e6c8c47746291292953d47ce35e61e01ec22d534a +size 1835053 diff --git a/-9E1T4oBgHgl3EQfUwOX/vector_store/index.pkl b/-9E1T4oBgHgl3EQfUwOX/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..0302b16fef7479f9390ceccaf38f66595867acd6 --- /dev/null +++ b/-9E1T4oBgHgl3EQfUwOX/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:470679b09eadb82da321ba38e08d63573b9c9138152caa4df71121439b357c6c +size 82952 diff --git a/.gitattributes b/.gitattributes index 056154baed41d80b854d142e2d118de899be321f..5b8eaa98a9661e319ce034b176f1d3b15868b2b1 100644 --- a/.gitattributes +++ b/.gitattributes @@ -436,3 +436,41 @@ _tAzT4oBgHgl3EQfS_vY/content/2301.01243v1.pdf filter=lfs diff=lfs merge=lfs -tex JtAzT4oBgHgl3EQfyP50/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text UNE0T4oBgHgl3EQf2gIi/content/2301.02712v1.pdf filter=lfs diff=lfs merge=lfs -text JtAzT4oBgHgl3EQfyP50/content/2301.01750v1.pdf filter=lfs diff=lfs merge=lfs -text +ktAzT4oBgHgl3EQfpv1w/content/2301.01617v1.pdf filter=lfs diff=lfs merge=lfs -text +DNE5T4oBgHgl3EQfUA_Z/content/2301.05541v1.pdf filter=lfs diff=lfs merge=lfs -text +GdFKT4oBgHgl3EQfby5L/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +UNE4T4oBgHgl3EQfmg2h/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +49E4T4oBgHgl3EQfbgzb/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +nNE5T4oBgHgl3EQfHw7p/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +DNE5T4oBgHgl3EQfUA_Z/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +-9E1T4oBgHgl3EQfUwOX/content/2301.03094v1.pdf filter=lfs diff=lfs merge=lfs -text +46dWaIoBww_ee65IHl0v/content/2301.00001v1.pdf filter=lfs diff=lfs merge=lfs -text +UNE0T4oBgHgl3EQf2gIi/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +9NFLT4oBgHgl3EQfBy4O/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +-9E1T4oBgHgl3EQfUwOX/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +9NE5T4oBgHgl3EQfRA5T/content/2301.05517v1.pdf filter=lfs diff=lfs merge=lfs -text +GdFKT4oBgHgl3EQfby5L/content/2301.11813v1.pdf filter=lfs diff=lfs merge=lfs -text +49E4T4oBgHgl3EQfbgzb/content/2301.05075v1.pdf filter=lfs diff=lfs merge=lfs -text +9NE5T4oBgHgl3EQfRA5T/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +UNE4T4oBgHgl3EQfmg2h/content/2301.05169v1.pdf filter=lfs diff=lfs merge=lfs -text +7tFJT4oBgHgl3EQfmyyq/content/2301.11589v1.pdf filter=lfs diff=lfs merge=lfs -text +gdE1T4oBgHgl3EQfzAU4/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +3NE4T4oBgHgl3EQfbAwV/content/2301.05068v1.pdf filter=lfs diff=lfs merge=lfs -text +t9E1T4oBgHgl3EQfQgOR/content/2301.03041v1.pdf filter=lfs diff=lfs merge=lfs -text +t9E1T4oBgHgl3EQfQgOR/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +46dWaIoBww_ee65IHl0v/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +UNAzT4oBgHgl3EQfXvzs/content/2301.01325v1.pdf filter=lfs diff=lfs merge=lfs -text +_tAzT4oBgHgl3EQfS_vY/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +6tE1T4oBgHgl3EQfTQPr/content/2301.03077v1.pdf filter=lfs diff=lfs merge=lfs -text +7dAzT4oBgHgl3EQf-f5K/content/2301.01934v1.pdf filter=lfs diff=lfs merge=lfs -text +4NAyT4oBgHgl3EQfo_jf/content/2301.00519v1.pdf filter=lfs diff=lfs merge=lfs -text +7tFJT4oBgHgl3EQfmyyq/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +19FAT4oBgHgl3EQfDBzX/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +gdE1T4oBgHgl3EQfzAU4/content/2301.03439v1.pdf filter=lfs diff=lfs merge=lfs -text +WdFJT4oBgHgl3EQf4i1g/content/2301.11666v1.pdf filter=lfs diff=lfs merge=lfs -text +ANE1T4oBgHgl3EQfowXh/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +3NE4T4oBgHgl3EQfbAwV/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +UNAzT4oBgHgl3EQfXvzs/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text +EtFLT4oBgHgl3EQfFS_K/content/2301.11987v1.pdf filter=lfs diff=lfs merge=lfs -text +X9FRT4oBgHgl3EQf_DhA/content/2301.13693v1.pdf filter=lfs diff=lfs merge=lfs -text +ItAzT4oBgHgl3EQfx_5k/content/2301.01746v1.pdf filter=lfs diff=lfs merge=lfs -text diff --git a/19FAT4oBgHgl3EQfDBzX/vector_store/index.faiss b/19FAT4oBgHgl3EQfDBzX/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..29262a1ce1cced1f5461323d6f11b4fe2d1df4ea --- /dev/null +++ b/19FAT4oBgHgl3EQfDBzX/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7861d5c7b90481b3a51340144988a633372395bbe7121c7a9cb3853f1c41e8bd +size 3473453 diff --git a/19FAT4oBgHgl3EQfDBzX/vector_store/index.pkl b/19FAT4oBgHgl3EQfDBzX/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..43b800717c3bf8255dca3b4a3baa7a72a81c5784 --- /dev/null +++ b/19FAT4oBgHgl3EQfDBzX/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:748b2e5704c9e4ffaec169234e4acdbea475f4332e752cd92eb52b756d13f860 +size 115476 diff --git a/3NE4T4oBgHgl3EQfbAwV/content/2301.05068v1.pdf b/3NE4T4oBgHgl3EQfbAwV/content/2301.05068v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9c3d62e9da77b7e5fef12b973c0dcbcf85dd1bda --- /dev/null +++ b/3NE4T4oBgHgl3EQfbAwV/content/2301.05068v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf6e9d1d654597287053fdbe3fcb4d92d6b35ddab429695855a7a273738162bc +size 828051 diff --git a/3NE4T4oBgHgl3EQfbAwV/vector_store/index.faiss b/3NE4T4oBgHgl3EQfbAwV/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..7047b00f83c57e18c8266db36f42e8ee9aced6b7 --- /dev/null +++ b/3NE4T4oBgHgl3EQfbAwV/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8719708cfeef62a619e27d3d1aa8076bb788f450e9bff3747db7a445c1bfa7aa +size 6094893 diff --git a/3NE4T4oBgHgl3EQfbAwV/vector_store/index.pkl b/3NE4T4oBgHgl3EQfbAwV/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..c501a3170977e944db72576359e03d9b85b5a72b --- /dev/null +++ b/3NE4T4oBgHgl3EQfbAwV/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7977f0d2e4964ada7ce904e0972f08a2bcf6b80298f97e387d5be52e47424714 +size 192324 diff --git a/46dWaIoBww_ee65IHl0v/content/2301.00001v1.pdf b/46dWaIoBww_ee65IHl0v/content/2301.00001v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3280ca8ed902a4a4c55980dccb88bc6646448721 --- /dev/null +++ b/46dWaIoBww_ee65IHl0v/content/2301.00001v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a288e5a34c387ec36329091d6359470e84e098cb8dc91efbfcd320adf286f66a +size 765699 diff --git a/46dWaIoBww_ee65IHl0v/vector_store/index.faiss b/46dWaIoBww_ee65IHl0v/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..e9a198c8b59a5cea819591b8bd17e7067144d3d0 --- /dev/null +++ b/46dWaIoBww_ee65IHl0v/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64013327466851fe04dc9cd27343c2f45002404e6139cba3736b0a3a3eed8208 +size 2555949 diff --git a/46dWaIoBww_ee65IHl0v/vector_store/index.pkl b/46dWaIoBww_ee65IHl0v/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..82339e5a15ae8d535d13806f5d494b7a064b28f2 --- /dev/null +++ b/46dWaIoBww_ee65IHl0v/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1df172bbaef27c67520a0551dfbd2019858b040961c4c26ea97ec0b5c0e3dac0 +size 98774 diff --git a/49E4T4oBgHgl3EQfbgzb/content/2301.05075v1.pdf b/49E4T4oBgHgl3EQfbgzb/content/2301.05075v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..40f8d5d78fe39bf37a16ec896a29c1390cd0cf20 --- /dev/null +++ b/49E4T4oBgHgl3EQfbgzb/content/2301.05075v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e84eafad470b72f96e2c52b231ab10f1515fb9ef268ed8160bf3f205e1f28203 +size 3116814 diff --git a/49E4T4oBgHgl3EQfbgzb/vector_store/index.faiss b/49E4T4oBgHgl3EQfbgzb/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..c1ef64f0a41cfe6c29560d3bf5799a623beb1d26 --- /dev/null +++ b/49E4T4oBgHgl3EQfbgzb/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3cf4c3cd1756f4868c729a5f7c07c06eb1241fa868a5186a81681f47bfe5a04 +size 2293805 diff --git a/49E4T4oBgHgl3EQfbgzb/vector_store/index.pkl b/49E4T4oBgHgl3EQfbgzb/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..97784f8f94806d427addd451621489b4364d550b --- /dev/null +++ b/49E4T4oBgHgl3EQfbgzb/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7556061973033cfed4e8e9f8952e8791f3d34f79e902cc4e2b38484854e3cd24 +size 74535 diff --git a/4NAyT4oBgHgl3EQfo_jf/content/2301.00519v1.pdf b/4NAyT4oBgHgl3EQfo_jf/content/2301.00519v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c9a4e8c8f47d2a135935c8280bdc7d3744d27251 --- /dev/null +++ b/4NAyT4oBgHgl3EQfo_jf/content/2301.00519v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cdb3e6ad9dd75d86e90a7526ba925f03ea1499a46e329638032dfc014e4de1d8 +size 844504 diff --git a/4NAyT4oBgHgl3EQfo_jf/vector_store/index.pkl b/4NAyT4oBgHgl3EQfo_jf/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..e428c5123af81a347e4a50b68e959ddcf6989ee7 --- /dev/null +++ b/4NAyT4oBgHgl3EQfo_jf/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:736538a4a2474339b3ff709988580c962c992cf3ff47f29b58c24827210d8724 +size 583770 diff --git a/4dE1T4oBgHgl3EQfAwIb/content/tmp_files/2301.02842v1.pdf.txt b/4dE1T4oBgHgl3EQfAwIb/content/tmp_files/2301.02842v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..e1c0d00688cbd5b3dcee68f8f63390eff12ae258 --- /dev/null +++ b/4dE1T4oBgHgl3EQfAwIb/content/tmp_files/2301.02842v1.pdf.txt @@ -0,0 +1,775 @@ +arXiv:2301.02842v1 [math.AG] 7 Jan 2023 +STRATIFIED BUNDLES ON THE HILBERT SCHEME OF n POINTS +SAURAV HOLME CHOUDHURY +Abstract. Let k be an algebraically closed field of characteristic p > 3 and S be a smooth projective +surface over k with k-rational point x. For n ≥ 2, let S[n] denote the Hilbert scheme of n points on S. In +this note, we compute the fundamental group scheme πalg(S[n], ˜ +nx) defined by the Tannakian category of +stratified bundles on S[n]. +1. Introduction +For a variety X over C, one has the classical notion of the fundamental group π1(Xan, x) defined using +the analytic topology on X. Over arbitrary base fields k, one has several analogues of the fundamental group +defined in terms of algebro-geometric information. +In [SGA1], Grothendieck introduced the notion of ´etale fundamental group π´et(X, x), where X is a scheme +and x is a geometric point of X, in terms of the finite etale covers of X. In [N76], Nori defined the Nori +fundamental group scheme πN(X, x), where X is a connected, reduced and complete scheme over a perfect +field k and x is a k-rational point, via Tannakian reconstruction using the category of essentially finite vector +bundles on X. The definition of πN(X, x) was extended to the case of connected and reduced k-schemes in +[N82]. Another analogue, the S-fundamental group scheme πS(X, x) was introduced and studied by Langer +in [Lan11] and [Lan12] for smooth projective varieties X over an algebraically closed field k. It is defined via +Tannakian reconstruction using the category of numerically flat vector bundles on X. The S-fundamental +group scheme for a smooth projective curve C over an algebraically closed field k was already introduced +and studied in [BPS06]. +The variant of the fundamental group scheme which is of prime importance in this note is the algebraic +fundamental group πalg(X, x). In [Gie75], Gieseker defined πalg(X, x) as the fundamental group scheme +corresponding to the Tannakian category of DX-modules, where DX is the sheaf of differential operators on +X. For X smooth over a field of positive characteristic, Gieseker introduced the notion of stratified bundles +and showed that the category of DX-modules is tensor equivalent to the category of stratified bundles on +X. Stratifed bundles were further studied in [dS07] and [BHdS21]. Precise definitions and statements will +be given in the next section. +Let S be a smooth projective surface over k. For n ≥ 2, let S[n] denote the Hilbert scheme n points on S. +It is well known that S[n] is a smooth projective variety of dimension 2n. In [PS20], the authors show that +for char k > 3 and n ≥ 2, there is an isomorphism of affine group schemes over k +π†(S, x)ab → π†(S[n], ˜ +nx) +where † = S, N or ´et. +In this note, we extend their results to the case of πalg and prove the following theorem. +Theorem. Let char k > 3 and n ≥ 2. There is an isomorphism of affine group schemes over k +f : πalg(S, x)ab → πalg(S[n], ˜ +nx) +In section 2, we recall the definition of stratified bundles and some of their basic properties. The formalism +of Tannakian reconstruction is recalled in section 3 and used to define the algebraic fundamental group +πalg(X, x). +1 + +The geometrical properties of the Hilbert scheme of n points on a smooth projective surface are in section 4. +In section 5, we prove a result about descent of stratified bundles which allows us to define the homomorphism +f by defining the associated functor of Tannakian categories. The concluding section 6 establishes the main +theorem by showing that f is an isomorphism. +Acknowledgements. We would like to thank Indranil Biswas and Ronnie Sebastian for their comments +on earlier drafts of this note. +2. Stratified bundles +Let k be a field of characteristic p and X be a noetherian scheme over k. Stratified bundles on X are +sequences of coherent sheaves on X satisfying infinite Frobenius descent. More precisely, the category of +stratified bundles on X, denoted S(X), consists of +• Objects (Ei, αi) are sequences of coherent OX-modules Ei, i ∈ N along with isomorphisms +αi : F ∗Ei+1 → Ei +for all i ∈ N, where F is the absolute Frobenius on X. +• Morphisms φ : (Ei, αi) → (Fi, βi) consists of a sequence of OX-module morphisms φi : Ei → Fi +such that φi ◦ αi = βi ◦ F ∗ (φi+1) +Let f : Y → X be a morphism and (Ei, αi) be a stratified bundle on X. Then we can define the pullback +along f, denoted f ∗(Ei, αi), as consisting of the sequence of OY coherent sheaves f ∗Ei and isomorphisms +are given by the composite maps +F ∗f ∗Ei+1 +γEi+1 +−−−→ f ∗F ∗Ei+1 +f ∗(αi) +−−−−→ f ∗Ei +where γ : F ∗f ∗ → f ∗F ∗ is the natural isomorphism of functors. +Thus S(X) is contravariant functor in X. One also has a tensor product on S(X) defined by taking term +by term tensor product. Also S(X) is an abelian category [cf. [BHdS21], Proposition 4.4]. +We recall some well known results about stratified bundles [cf [dS07], [Gie75]]. +Proposition. If (Ei, αi) is a stratified bundle on X, then Ei is a locally free OX-module for all i ∈ N. +This allows us to define duals of stratified bundles, making S(X) into an abelian rigid tensor category. +The rank of a stratified bundle (Ei, αi) is defined to be the rank of E0. The trivial stratified bundles on X +are of the form ⊕(OX, ...; F ∗, ...). +Let DX be the sheaf of differential operators on X. The category of DX modules consists of +• Objects coherent OX modules E equipped with a DX action i.e a morphism of OX-algebras +DX → Endk(E) +• Morphisms OX-linear maps E → F compatible with the DX action +A theorem of Katz [[Gie75], Theorem 1.3] shows that for X smooth over k, then the category of stratified +bundles on X and the category of DX modules are tensor equivalent to each other. +We close this section with the definition of G equivariant stratified bundles on a variety X admitting +action of a group G on it. +Definition. A stratified bundle (Ei, αi) is said to be a G-equivariant stratified bundle if Ei are G-equivariant +vector bundles and αi are G-equivariant OX module morphisms. +3. Tannakian categories and fundamental group schemes +In this section we recall the definition and basic properties of Tannakian categories. +We then recall +Gieseker’s definition of the fundamental group scheme πalg using the Tannakian formalism. +2 + +3.1. Tannakian Categories and affine group schemes. Tannakian categories were defined and studied +in [DM82] to formalize the properties of Repk(G), the category of finite dimensional k-representations of G, +an affine group scheme over k. +Definition (Neutral Tannakian Categories). A rigid abelian tensor category C with End I = k is a neutral +Tannakian category if it admits an exact faithful k-linear tensor functor ω : C → Veck. Any such functor is +said to be a fiber functor for C. +Given a neutral Tannakian category (C, ⊗, ω, I), we define the functor Aut⊗(ω) : k − algebra → Sets +such that for k-algebra R, Aut⊗(ω)(R) consists of the families (λX) for X ∈ ob(C), where λX is a R-linear +automorphism of X ⊗ R such that λX1⊗X2 = λX1 ⊗ λX2, λI = idR, and +λY ◦ (α ⊗ 1) = (α ⊗ 1) ◦ λX : X ⊗ R → Y ⊗ R +for all morphisms α : X → Y . +Theorem (Main theorem for neutral Tannakian categories, [DM82], Theorem 2.11). Let (C, ⊗) be a rigid +abelian tensor category such that k = End(I) and let ω : C → Veck be an exact faithful tensor functor. Then +• The functor Aut⊗(ω) of k-algebras is represented by an affine group scheme G. +• The functor C → Repk(G) is an equivalence of tensor categories. +Theorem. Let (C, ⊗, ω, I) and (C′, ⊗, ω′, I′) be neutral Tannakian categories which correspond to the repre- +sentation categories of the affine k group schemes G and G′ respectively. Then any functor of Tannakian +categories from C → C′ is induced by a unique morphism of affine k group schemes G′ → G. +This theorem allows us to define many variants of fundamental groups of a scheme X by considering +different Tannakian categories naturally associated with X. The following result is very useful in establishing +a given morphism between affine group schemes is an isomorphism. +Theorem ([DM82], Theorem 2.21). Let f : G → G′ be a homomorphism of group schemes over k and +Rep (f) : Rep (G′) → Rep (G) be the corresponding functor of Tannakian categories. Then +• f is faithfully flat if and only if Rep (f) is fully faithful and has essential image closed under subobjects +i.e for V ′ ∈ Rep (G′) and suboject W ⊂ Rep (f)(V ′), there is a subobject W ′ ⊂ V ′ in Rep (G′) such +that Rep (f)(W ′) ≃ W in Rep (G) +• f is closed immersion if and only if every object of Rep (G) is a subquotient of some object in the +essential image of Rep (f). +We finish by recalling a basic result on affine group schemes (we refer to section 4.1 in [PS20] for details). +Let G be a affine group scheme over k, Gab be its abelianization (i.e the maximal abelian quotient of G) +and α : G → Gab be the (faithfully flat) quotient morphism . We can then define the composite morphism +φ : Gn +αn +−−→ Gn +ab +m +−→ Gab +where m is the multiplication homomorphism. As Sn acts on the k-group scheme Gn, we can define the +notion of a Sn-invariant group morphism ψ : Gn → H for any k-group scheme H. +Lemma 3.1. Let G and H be two group schemes over k. For an integer n ≥ 2, the set of Sn-invariant +group morphisms Gn → H is in bijective correspondence with the set of group morphism Gab → H i.e any +morphism of k-group schemes φ : Gn → H which is Sn-invariant factors uniquely through a morphism +ψ : Gab → H such that φ = ψ ◦ h +3.2. The group scheme πalg(X, x). Classically, over C, the Riemann-Hilbert correspondence identifies the +category of vector bundles equipped with integrable connections on a smooth connected projective variety +X/C with the category of representations of the topological fundamental group πtop(X, x) for some chosen +base point x. Via GAGA, this gives a purely algebraic description of the category of representations of the +topological fundamental group π(X, x). This category (equipped with the fiber functor (E, ∇) → Ex) is +a neutral Tannakian category and can be identified, via the Tannakian formalism, with the representation +3 + +category of the proalgebraic completion of the topological fundamental group, denoted as πtop(X, x)alg. +Over a field k of characteristic 0, the category of flat connections on a smooth variety X is tensor equiva- +lent to the category of DX-modules. However over a field of characteristic p, the category of flat connections +on X is not as well behaved as the category of DX-modules and one defines a fundamental group scheme for +X by Tannakian formalism using the category of DX-modules. By Katz’s theorem mentioned before, the +fundamental group coincides with the one defined using S(X) below. +Let x ∈ X(k) be a k-rational point. Then the abelian rigid tensor category S(X) is neutralized by the +fiber functor +Tx : S(X) → V eck +The fundamental group scheme defined by the neutral Tannakian category (S(X), ⊗, Tx, (OX, F ∗)) is +called the algebraic fundamental group of X based at x and is denoted by πalg(X, x). +The following basic properties of πalg are well known. +• (Independence of basepoint) Let X be a geometrically connected, smooth projective k-scheme. Then +for all x1, x2 ∈ X(k), one has +πalg(X, x1) ≃ πalg(X, x2) +• (Product rule) For X1, X2 geometrically connected and smooth over k and xi ∈ Xi(k), there is an +isomorphism +πalg(X1, x1) × πalg(X2, x2) → πalg(X1 × X2, (x1, x2)) +• For X smooth and open immersion U +i−→ X such that the complement of U in X has codimension +≥ 2 and x ∈ U(k), then the homomorphism +πalg(U, x) → πalg(X, x) +associated to the restriction functor i∗ : S(X) → S(U) is an isomorphism. +4. Geometry of Hilbert Scheme of points +Let S be a smooth projective surface over k. We fix notation as follows +• Sn denotes the n-fold cartesian product of S with itself. +• S(n) denotes the nth symmetric product of S defined as the quotient Sn/Sn, where Sn denotes the +symmetric group on n letters. +• S[n] denotes the Hilbert scheme of n points on S. +Let ρ : Sn → S(n) be the quotient map and h : S[n] → S(n) be the Hilbert-Chow morphism. We write +S(n) +◦ +for the open subset of S(n) consisting of distinct points with S[n] +◦ +:= h−1(S(n) +◦ +) and Sn +◦ := ρ−1(S(n) +◦ +). +The map hn,◦ : S[n] +◦ +→ S(n) +◦ +is an isomorphism. We have the diagram: +S[n] +Sn +S(n) +hn +ρn +In general, Hilbert schemes of points on a projective variety display a lot of pathological features. But in +[Fog68] the author shows that, in the case of smooth projective surface S, S[n] is a smooth projective variety. +Thus, in this case, the Hilbert-Chow morphism h : S[n] → S(n) is a resolution of singularities. +4 + +One can consider S(n) as the set of effective 0-cycles of degree n on S(n). In this case it is easy to see that +S(n) admits a stratification by type, where the type of a 0-cycle y of degree n is a tuple (n1, . . . , nr) where y +can be written as +y = Σr +j=1njxj +where xj are distinct points of S with multiplicities n1 ≥ n2 ≥ · · · ≥ nr, where nj are positive integers. +Let C(n1, . . . , nr) denote the subset of S(n) consisting of points of the type (n1, . . . , nr). Let S(n) +∗ += +C(1, . . . , 1) � C(2, . . . , 1) denote the open subset of S(n) consisting of points of type (1, . . . , 1) and (2, . . . , 1). +Let S[n] +∗ +and Sn +∗ denote the preimage of S(n) +∗ +under h and ρ respectively. +We recall some basic properties below which we will need later (we refer to [Fog68], [PS20] for details). +• The subsets C(n1, . . . , nr) are nonsingular of dimension 2r. +• The closed subset S(n) \ S(n) +∗ +is of codimension ≥ 2 in S(n). +• The closed subset S[n] \ S[n] +∗ +is of codimension 2 in S[n]. +• The closed subset Sn \ Sn +∗ is of codimension ≥ 4 in Sn. +• The closed subset S(n) +∗ +\ S(n) +◦ +is of codimension 2 in S(n) +∗ +. +• When characteristic of k ̸= 2, for y ∈ C(2, 1, . . . , 1), the scheme theoretic fiber h−1(y) is isomorphic +to P1 +k. In fact, S[n] +∗ +is the blowup of S(n) +∗ +along C(2, 1, . . . , 1). +We end this section by recalling a result of Fogarty ([Fog77], Proposition 3.6). +Proposition. If L is a Sn-invariant line bundle on Sn, there exists a line bundle L′ on S(n) such that +h∗L′ ≃ L. +It follows that L′ in the proposition is isomorphic to σ∗(L)Sn +5. The functor between Tannakian categories +Let S be a smooth projective surface over k and (Ei, αi) be a stratified bundle on S[n]. Restricting to S[n] +∗ +gives us a functor +i∗ : S(S[n]) → S(S[n] +∗ ) +which is a equivalence of categories as S[n] +∗ +is the complement of a codimension 2 closed subset of S[n]. +Next we show that a stratified bundle on S[n] +∗ +can be pushed forward under h to get a stratified bundle on +S(n) +∗ +. First we begin by a result on descent of vector bundles along the morphism h : S[n] +∗ +→ S(n) +∗ +. Similar +results have been established by authors in [Ish83] and [PS20]. +Proposition 1. Assume char k ̸= 2. Let E be a vector bundle on S[n] +∗ +which restricts to trivial vector bundles +on the fibers of h over S(n) +∗ +. Then h∗E is a locally free OS(n) +∗ -module. Moreover the natural map +h∗h∗(E) → E +is an isomorphism. +Proof. Let x ∈ S(n) +∗ +be a point of type (2, 1, . . ., 1). Then by assumption, the fiber of h over x is isomorphic +to P1 +k. Let J be the ideal sheaf of the closed subscheme h−1(x) and Ix be the ideal sheaf of the closed point +x. We have +J = IxOS[n] +∗ +For all n ≥ 1, let Yn denote the closed subscheme of S[n] +∗ +corresponding to the ideal sheaf J n. Consider +the following short exact sequence of sheaves on S[n] +∗ +0 → J ⊗ E → E → E|Y1 → 0 +5 + +Pushing forward by h, we get the following exact sequence of sheaves on S(n) +∗ +h∗E → H0(Y1, E|Y1) → R1h∗(J ⊗ E) +We claim that the completion of R1h∗(J ⊗ E) at the maximal ideal mx in OS(n) +∗ +,x is 0. The proof uses +the theorem of formal functions which says that +(R1h∗(J ⊗ E))∧ ≃ lim +←− H1(Yn, J ⊗ E ⊗ OS[n] +∗ /J n) +We prove by induction that H1(Yn, J ⊗E ⊗OS[n] +∗ /J n) = 0. As Y1 ≃ P1 +k, the sheaves J n/J n+1 are locally +free. These sheaves are also globally generated over Y1 as we have the surjection +mn +x/mn+1 +x +⊗O +S(n) +∗ +,x OS[n] +∗ +≃ In +x /In+1 +x +⊗O +S(n) +∗ +OS[n] +∗ +։ J n/J n+1 +As J n/J n+1 is locally free on Y1 ≃ P1 +k and globally generated, it is a direct sum of line bundles each of +which has degree ≥ 0. Thus one gets the base case of induction from degree considerations, as +H1(Y1, J ⊗ E ⊗ OS[n] +∗ /J = H1(Y1, J /J 2 ⊗ EY1) = 0 +Assume that the claim is true for n. Then the proof for n + 1 follows from the long exact sequence in +cohomology attached to the short exact sequence of sheaves on Yn+1 +0 → J n+1/J n+2 ⊗ E → J /J n+2 ⊗ E → J /J n+1 ⊗ E → 0 +which gives us the exact sequence +H1(Yn+1, J n+1/J n+2 ⊗ E) → H1(Yn+1, J /J n+2 ⊗ E) → H1(Yn+1, J /J n+1 ⊗ E) +We know H1(Yn, J n+1/J n+2⊗E) = H1(Y1, J n+1/J n+2⊗E) = 0 (by degree consideration) and H1(Yn+1, J /J n+1⊗ +E) = H1(Yn, J /J n+1 ⊗ E) = 0 (by induction hypothesis), thus we get +H1(Yn+1, J /J n+2 ⊗ E) = 0 +Thus the stalk of R1h∗(J ⊗ E) at x is 0. +This shows that the natural map h∗E → H0(Y1, E|Y1) is surjective in a neighbourhood of x. Let f1, ..., fr +be a basis of H0(Y1, E|Y1). Let Spec(R) be an affine neighbourhood of x where the natural map is surjective +and let ˜fi ∈ Γ(Spec(R), h∗E) = Γ(h−1(Spec(R)), E) be lifts of fi. Using ˜fi one defines a homomorphism +O⊕r +S[n] +∗ |h−1(Spec(R)) → E +on h−1(Spec(R) which is a surjection (and hence an isomorphism) on Y1. As h is proper, there exists a +smaller affine neighbourhood U of x over which there is an isomorphism +O⊕r +V +≃ E +where V = h−1(U). Applying h∗, we get +(h∗OV )⊕r ≃ h∗E +As S(n) +∗ +is normal and h : S[n] +∗ +→ S(n) +∗ +is birational with connected fibers, by a form of Zariski’s main +theorem [cf [Har77], Corollary 11.3 and 11.4], we have that h∗OV ≃ OU and thus h∗E is locally free. The +natural morphism +h∗h∗(E) → E +is clearly an isomorphism. +□ +6 + +Let VBS(n) +∗ +be the category of locally free sheaves on S(n) +∗ +and VBh +S[n] +∗ +be the category of locally free +sheaves on S[n] +∗ +which restrict to trivial vector bundles on the fibers of h. Proposition 1 above gives us an +equivalence of categories. +Proposition 2. Assume char k ̸= 2. The pushforward functor +h∗ : VBh +S[n] +∗ +→ VBS(n) +∗ +is an equivalence of categories with the inverse given by +h∗ : VBS(n) +∗ +→ VBh +S[n] +∗ +Proof. We observe that if E′ ≃ h∗(E), then E ≃ h∗E′. This shows that h∗ is essentially surjective. The +natural map +HomS(n) +∗ (h∗E, h∗F) → HomS[n] +∗ (E, F) +is bijective. Thus h∗ is an equivalence of categories. +□ +Corollary. For all E ∈ VBh +S[n] +∗ , the natural map +F ∗h∗(E) → h∗F ∗(E) +is an isomorphism over S(n) +∗ +. +Proof. As F ∗E is also an object of VBh +S[n] +∗ , thus both sheaves are locally free of the same rank. Thus it +suffices to show that the natural map +F ∗h∗(E) → h∗F ∗(E) +is surjective. As F is faithfully flat on the smooth locus of S(n) +∗ +, the claim holds on the smooth locus. Let +x ∈ S(n) +∗ +be of type (2, 1, . . . , 1). Then the restriction of F ∗h∗(E) to x is naturally isomorphic to H0(Y1, E|Y1) +and the restriction of h∗F ∗(E) to x is H0(Y1, F ∗(E|Y1). The restriction of the natural map to x is the map +F ∗ : H0(Y1, E1) → H0(Y1, F ∗E1) +which is surjective. +□ +By Theorem 2.2 of [Gie75], we have that every stratified bundle on P1 +k is trivial. Thus the above results +give us +Proposition 3. Assume char k ̸= 2. Let (Ei, αi) be a stratified bundle on S[n] +∗ . Then h∗(Ei) is locally free +OS(n) +∗ -module for all i ∈ N. Moreover the natural map +h∗h∗(Ei) → Ei +is an isomorphism. Furthermore the natural map +F ∗h∗(Ei) → h∗F ∗(Ei) +is an isomorphism over S(n) +∗ +. +This allows us to define the pushforward of a stratified bundle (Ei, αi) on S[n] +∗ . The pushforward denoted +h∗(Ei, αi) is given by the sequence of vector bundles h∗Ei for all i ∈ N and the isomorphisms are given by +the composite +F ∗h∗(Ei+1) +ηEi+1 +−−−→ h∗F ∗(Ei+1) +h∗(αi) +−−−−→ h∗(Ei) +where η : F ∗h∗ → h∗F ∗ is the natural transformation. +7 + +Thus we get a functor +h∗ : S(S[n] +∗ ) → S(S(n) +∗ +) +h∗ is additive tensor functor as on the smooth locus S(n) +◦ +we have the isomorphisms +h∗((Ei, αi) ⊕ (Fi, βi))|S(n) +◦ +≃ h∗(Ei, αi)|S(n) +◦ +⊕ h∗(Fi, βi))|S(n) +◦ +h∗((Ei, αi) ⊗ (Fi, βi))|S(n) +◦ +≃ h∗(Ei, αi)|S(n) +◦ +⊗ h∗(Fi, βi))|S(n) +◦ +which extend to S(n) +∗ +due to codimension reasons. +The following commutative diagram shows that h∗h∗(Ei, αi) is isomorphic to (Ei, αi) as stratified bundles +with the isomorphism given by the natural morphisms h∗h∗Ei → Ei. +F ∗h∗h∗Ei+1 +h∗F ∗h∗Ei+1 +h∗h∗F ∗Ei+1 +h∗h∗Ei +F ∗Ei+1 +F ∗Ei+1 +Ei +h∗ηEi+1 +h∗h∗αi +αi +γh∗Ei+1 += +Consider the pullback functor +ρ∗ : S(S(n) +∗ +) → S(Sn +∗ ) +which takes values in the category of Sn-equivariant stratified bundles on Sn +∗ . Also we have the extension +functor +j∗ : S(Sn +∗ ) → S(Sn) +which is an equivalence of categories. Composing these functors together, we get a functor +T : S(S[n]) → S(Sn) +given by +T = j∗ ◦ ρ∗ ◦ h∗ ◦ i∗ +Clearly T is an additive tensor functor. Note that h∗ is fully faithful, ρ∗ : S(S(n) +∗ +) → S(Sn +∗ ) is fully faithful +(as ρ : Sn +◦ → S(n) +◦ +is finite ´etale) and j∗ : S(Sn +∗ ) → S(Sn) is an equivalence of categories (due to codimension +reasons). Thus T is fully faithful. +5.1. The homomorphism. Fix n distinct k-valued points x1, . . . , xk ∈ S(k). +Let ˜x ∈ S[n] such that +h(˜x) = σ(x1, . . . , xn) = z ∈ S(n) +◦ +. Then the categories S(S[n]) and S(Sn) are neutralized by the respective +fiber functors +τ˜x : S(S[n]) → Veck +(Ei, αi) �→ (E0)˜x +τ(x1,...,xn) : S(Sn) → Veck +(Fi, βi) �→ (F0)(x1,...,xn) +If T ((Ei, αi)) = (Fi, βi) that we have natural isomorphisms (E0)˜x ≃ h∗(E0)z ≃ (F0)(x1,...,xn). +Thus we have a functor of Tannakian categories +T : (S(S[n]), ⊗, τ˜x, (OS[n], d)) → (S(Sn), ⊗, τ(x1,...,xn), (OSn, d)) +8 + +which by the independence of basepoint property of S induces a functor of Tannakian categories +T : (S(S[n]), ⊗, τ ˜ +nx, (OS[n], d)) → (S(Sn), ⊗, τ(x,...,x), (OSn, d)) +and hence a morphisms of the associated fundamental group schemes +˜f : πalg(Sn, (x, . . . , x)) → πalg(S[n], ˜ +nx) +Note that by proposition 3.2 we have +πalg(Sn, (x, . . . , x)) ≃ πalg(S, x)n +. +As +T : (S(S[n]), ⊗, T ˜ +nx, (OS[n], d)) → (S(Sn), ⊗, T(x,...,x), (OSn, d)) +takes stratified bundles on S[n] to Sn-equivariant stratified bundles on Sn and a Sn-equivariant stratified +bundles on Sn corresponds to a Sn-invariant representation of πalg(S, x)n, by 3.1, ˜f factors uniquely through +f : πalg(S, x)ab → πalg(S[n], ˜ +nx) +6. Isomorphism of fundamental group schemes +In this section, we show that f is an isomorphism of affine group schemes. We begin by proving a result +about Sn-equivariant stratified line bundles on Sn. +Proposition 4. Let (Li, αi) be a Sn-equivariant stratified line bundles on Sn. Then there exists a stratified +line bundle (Li, βi) such that ρ∗(Li, βi) ≃ (Li, αi) +Proof. By Fogarty’s result mentioned above, for any Sn-equivariant line bundle Li there exists line bundle +Li ≃ ρ∗LSn +i +such that ρ∗Li ≃ Li. Pushing forward αi and taking Sn invariants we get the isomorphism +ρ∗(F ∗Li+1)Sn +ρ∗(αi)Sn +−−−−−−→ ρ∗(Li)Sn +We show that the natural homomorphism +F ∗(ρ∗(Li)Sn) → (F ∗ρ∗(Li)Sn) → (ρ∗F ∗(Li)Sn) +is an isomorphism. Pulling back the morphism under ρ, we get the commutative diagram +ρ∗F ∗((ρ∗Li)Sn) +ρ∗((ρ∗F ∗Li)Sn) +F ∗Li +F ∗Li += +where the vertical morphisms are the natural morphism which are isomorphisms by Fogarty’s theorem. +By pushing forward under ρ and taking Sn invariants we get that +F ∗(ρ∗(Li)Sn) → (ρ∗F ∗(Li)Sn) +is an isomorphism. We define βi to be the composite isomorphism +F ∗(ρ∗(Li)Sn) → (ρ∗F ∗(Li)Sn) +ρ∗(αi)Sn +−−−−−−→ ρ∗(Li)Sn +The commutative diagram also gives us that ρ∗(Li, βi) ≃ (Li, αi) +□ +9 + +6.1. Faithfully flat. Next we show that the morphism f is faithfully flat +Proposition 5. The homomorphism +f : πalg(S, x)ab → πalg(S[n], ˜ +nx) +is faithfully flat. +Proof. By [[DM82] Theorem 2.21], this is equivalent to showing that the functor +T : S(S[n]) → S(Sn) +is fully faithful and the essential image of T is closed under taking subobjects. We already know that T +is fully faithful. Let E• = (Ei, αi) be a stratified bundle on S[n] and F• := T (E•) be the corresponding +Sn-equivariant stratified bundle on Sn. If F′ +• ⊂ F• is a Sn-equivariant stratified subbundle, then we need +to show there exists E′ +• ⊂ E• such that T (E′ +•) = F′ +•. +The proof proceeds by induction on the rank of E•. If rank E• = 1, the proof is immediate. Let rank +E• ≥ 2 +Then the stratified bundles F• and F′ +• correspond to the representations +πalg(Sn, (x, . . . , x) → πalg(S, x)ab → GL(V ) +and +πalg(Sn, (x, . . . , x) → πalg(S, x)ab → GL(V ′) +respectively. +As πalg(S, x)ab is an abelian affine group scheme over k, all its irreducible representations are one dimen- +sional. Thus one gets that the πalg(S, x)ab-module V/V ′ has a one dimensional quotient W. Thus there is +a πalg(S, x)ab-module surjection V → W such that the kernel contains V ′. Let L• be the Sn-equivariant +stratified bundle corresponding to W. Thus we have a short exact sequence of Sn-equivariant stratified +bundles +0 → K• → F• → L• → 0 +where F′ +• ⊂ K•. +By proposition 1 above, we know that Li := ρ∗LSn +i +is a line bundle on S(n) and ρ∗Li = Li +We claim that the following complex of sheaves on S(n) +∗ +is exact for all i ∈ N +(1) +0 → (ρ∗Ki)Sn|S(n) +∗ +→ (ρ∗Fi)Sn|S(n) +∗ +→ (ρ∗Li)Sn|S(n) +∗ +→ 0 +It is enough to show that (ρ∗Fi)Sn|S(n) +∗ +→ (ρ∗Li)Sn|S(n) +∗ +is surjective. We note that (ρ∗Fi)Sn|S(n) +∗ += +h∗(Ei|S[n] +∗ ). Let C be the cokernel +h∗(Ei|S[n] +∗ ) → (ρ∗Li)Sn|S(n) +∗ +→ C → 0 +Pulling back under ρ, we get the following commutative diagram on Sn +∗ +ρ∗h∗(Ei|S[n] +∗ ) +ρ∗((ρ∗Li)Sn|S(n) +∗ ) +ρ∗C +0 +Fi +Li|Sn +∗ +0 += += +10 + +The rows are exact and hence ρ∗C = 0. As ρ is surjective, this implies C = 0. Thus Ki := (ρ∗Ki)Sn|S(n) +∗ +is locally free on S(n) +∗ +Pulling back the exact sequence (1) under h, we get a short exact sequence of locally free sheaves on S[n] +∗ +0 → h∗Ki|S[n] +∗ +→ Ei|S[n] +∗ +→ ˜Li|S[n] +∗ +→ 0 +where ˜Li := h∗Li. +As the complement of S[n] +∗ +in S[n] is of codimension ≥ 2 and Ei, L are locally free, the surjective morphism +Ei|S[n] +∗ +→ ˜Li|S[n] +∗ +extends to a unique morphism τi : Ei → ˜Li. This is surjective as L is of rank 1 and τ := (τi) give a +nonzero morphism of stratified bundles +E• → ˜L• +where ˜L• := h∗(ρ∗(L•)Sn). Let κ• be the kernel of the morphism E• → ˜L•. Then T (κ•) = K•. Thus, by +the induction hypothesis on rank, there exists a stratified subbundle E′ +• ⊂ κ• ⊂ E• such that T (E′ +•) = F′ +•. +□ +6.2. Closed immersion. We begin by recalling a result from [PS20]. +Let p ∈ S(n) be a point of type (n1, n2, . . . , nr). Let p′ +i, for i = 1, 2, . . . m be the points in the fiber h−1(p). +Let A be the local ring OS(n),p and B be the semilocal ring OSn ⊗OS(n) A. Then B is a finite A module and +BSn = A. +Lemma 6.1. When char k > n1, any Sn-equivariant surjective B-module homomorphism f : M → N of +finitely generated B modules descends to surjective A-module homomorphism of the Sn-invariants M Sn → +N Sn +This allows us to prove the following analogue of Proposition 5.3.6 in [PS20]. +Proposition. Let E• = (Ei, αi) be a Sn-equivariant stratified bundle on Sn +(1) Let p ∈ S(n) be a point of type (n1, n2, . . . , nr). If char k > n1, then the sheaf ρ∗ESn +i +is locally free +in a neighbourhood of p for all i. +(2) Let U denote the largest open subset where ρ∗ESn +i +is locally free, then on ρ−1(U), the natural mor- +phism +ρ∗ρ∗ESn +i +→ Ei +is an isomorphism for all i ∈ N +Proof. The first assertion is proved by induction on the rank of E•. If E• is a Sn-equivariant stratified bundle +of rank 1, then by proposition 1, ρ∗ESn +i +is locally free on S(n) for all i. In general, as E• corresponds to a +representation of the abelian group scheme πalg(S, x)ab, there exists a Sn-equivariant short exact sequence +of locally free sheaves on Sn +0 → K• → E• → L• → 0 +Pushing forward by ρ and taking Sn-invariants we get the exact sequence for all i +0 → ρ(Ki)Sn → ρ(Ei)Sn → ρ(Li)Sn +We claim that the homomorphism on the right is surjective in the neighbourhood of a point p of type +(n1, n2, . . . , nr). Surjectivity can be checked after passing to a formal neighbourhood of p and thus reduces to +lemma 6.1. By induction hypothesis on rank, both ρ(Ki)Sn and ρ(Li)Sn are locally free on a neighbourhood +of p and hence so is ρ(Ei)Sn. +11 + +The second assertion follows from the observation that the natural homomorphism +ρ∗ρ∗ESn +i +→ Ei +is an isomorphism on ρ−1(S(n) +◦ +) as as ρ : Sn +◦ → S(n) +◦ +is finite ´etale. As the complement of Sn +◦ in ρ−1(U) is of +codimension ≥ 2 and both sheaves are locally free on ρ−1(U), thus the natural morphism is an isomorphism. +□ +Proposition 6. Let char k > 3. The homomorphism +f : πalg(S, x)ab → πalg(S[n], ˜ +nx) +is faithfully flat. +Proof. By [[DM82], Theorem 2.21], it is enough to show that the functor +T : S(S[n]) → S(Sn) +is essentially surjective. Thus we want to show that for any Sn-equivariant stratified bundle E• on Sn, there +exists a stratified bundle F• on S[n] such that T (F•) = E•. +Let U be the open subset of S(n) consisting of points of type (1, 1, . . . , 1), (2, 1, . . . , 1), (3, 1, . . ., 1) and +(2, 1, 1, . . ., 1). By assumption on characteristic of k and the previous proposition, we get that ρ∗ESn +i +is +locally free on U. Also we have on ρ−1(U), the natural morphism +ρ∗ρ∗ESn +i +→ Ei +is an isomorphism. Imitating proposition 1 above, this allows us to define a stratified bundle (ρ∗ESn +i +, βi) on +U such that ρ∗(ρ∗ESn +i +, βi) ≃ E•. Pulling back under h to h−1(U) (whose complement in S[n] has codimension +≥ 3) and extending to S[n], we get a stratified bundle F• such that T (F•) = E• +□ +As f is both faithfully flat and a closed immersion, we get the following theorem +Theorem 6.2. Let char k > 3. The homomorphism +f : πalg(S, x)ab → πalg(S[n], ˜ +nx) +is an isomorphism. +References +[BHdS21] Biswas, Indranil, Ph`ung Hˆo Hai, and Jo˜ao Pedro Dos Santos. ”On the fundamental group schemes of certain quotient +varieties.” Tohoku Mathematical Journal 73, no. 4 (2021): 565-595. +[BPS06] Biswas, Indranil, A. J. Parameswaran, and S. Subramanian. ”Monodromy group for a strongly semistable principal +bundle over a curve.” Duke Mathematical Journal 132, no. 1 (2006): 1-48. +[DM82] Deligne, Pierre; Milne, James (1982), ”Tannakian categories”, in Deligne, Pierre; Milne, James; Ogus, Arthur; Shih, +Kuang-yen (eds.), Hodge Cycles, Motives, and Shimura Varieties, Lecture Notes in Mathematics, vol. 900, Springer, pp. +101–228 +[dS07] Dos Santos, Jo˜ao Pedro Pinto. ”Fundamental group schemes for stratified sheaves.” Journal of Algebra 317, no. 2 (2007): +691-713. +[Fog68] Fogarty, John. ”Algebraic families on an algebraic surface.” American Journal of Mathematics 90, no. 2 (1968): 511-521. +[Fog77] Fogarty, John. ”Line bundles on quasi-symmetric powers of varieties.” Journal of Algebra 44, no. 1 (1977): 169-180. +[Gie75] Gieseker, David. ”Flat vector bundles and the fundamental group in non-zero characteristics.” Annali della Scuola +Normale Superiore di Pisa-Classe di Scienze 2, no. 1 (1975): 1-31. +[Har77] Hartshorne, Robin. Algebraic geometry. Vol. 52. Springer Science & Business Media, 2013. +[Ish83] Ishimura, Sadao. ”A descent problem of vector bundles and its applications.” Journal of Mathematics of Kyoto University +23, no. 1 (1983): 73-83. +[Lan11] Langer, Adrian. ”On the S-fundamental group scheme.” In Annales de l’Institut Fourier, vol. 61, no. 5, pp. 2077-2119. +2011. +[Lan12] Langer, Adrian. ”On the S-fundamental group scheme. II.” Journal of the Institute of Mathematics of Jussieu 11, no. +4 (2012): 835-854. +[N76] Nori, Madhav V. ”On the representations of the fundamental group.” Compositio Mathematica 33, no. 1 (1976): 29-41. +[N82] Nori, Madhav V. ”The fundamental group-scheme.” Proceedings Mathematical Sciences 91, no. 2 (1982): 73-122. +[PS20] Paul, Arjun, and Ronnie Sebastian. ”Fundamental group schemes of Hilbert scheme of n points on a smooth projective +surface.” Bulletin des Sciences Math´ematiques 164 (2020): 102898. +12 + +[SGA1] Grothendieck, Alexander, and Michele Raynaud. ”Revˆetements´etales et groupe fondamental (SGA 1).” arXiv preprint +math/0206203 (2002). +The Institute of Mathematical Sciences, (HBNI), Chennai 600113. +Email address: sauravhc@imsc.res.in +13 + diff --git a/4dE1T4oBgHgl3EQfAwIb/content/tmp_files/load_file.txt b/4dE1T4oBgHgl3EQfAwIb/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..cf9a510b8ccf4cf01942cd6607395caa3a193455 --- /dev/null +++ b/4dE1T4oBgHgl3EQfAwIb/content/tmp_files/load_file.txt @@ -0,0 +1,457 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf,len=456 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='02842v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='AG] 7 Jan 2023 STRATIFIED BUNDLES ON THE HILBERT SCHEME OF n POINTS SAURAV HOLME CHOUDHURY Abstract.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let k be an algebraically closed field of characteristic p > 3 and S be a smooth projective surface over k with k-rational point x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' For n ≥ 2, let S[n] denote the Hilbert scheme of n points on S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In this note, we compute the fundamental group scheme πalg(S[n], ˜ nx) defined by the Tannakian category of stratified bundles on S[n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Introduction For a variety X over C, one has the classical notion of the fundamental group π1(Xan, x) defined using the analytic topology on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Over arbitrary base fields k, one has several analogues of the fundamental group defined in terms of algebro-geometric information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In [SGA1], Grothendieck introduced the notion of ´etale fundamental group π´et(X, x), where X is a scheme and x is a geometric point of X, in terms of the finite etale covers of X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In [N76], Nori defined the Nori fundamental group scheme πN(X, x), where X is a connected, reduced and complete scheme over a perfect field k and x is a k-rational point, via Tannakian reconstruction using the category of essentially finite vector bundles on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The definition of πN(X, x) was extended to the case of connected and reduced k-schemes in [N82].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Another analogue, the S-fundamental group scheme πS(X, x) was introduced and studied by Langer in [Lan11] and [Lan12] for smooth projective varieties X over an algebraically closed field k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' It is defined via Tannakian reconstruction using the category of numerically flat vector bundles on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The S-fundamental group scheme for a smooth projective curve C over an algebraically closed field k was already introduced and studied in [BPS06].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The variant of the fundamental group scheme which is of prime importance in this note is the algebraic fundamental group πalg(X, x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In [Gie75], Gieseker defined πalg(X, x) as the fundamental group scheme corresponding to the Tannakian category of DX-modules, where DX is the sheaf of differential operators on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' For X smooth over a field of positive characteristic, Gieseker introduced the notion of stratified bundles and showed that the category of DX-modules is tensor equivalent to the category of stratified bundles on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Stratifed bundles were further studied in [dS07] and [BHdS21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Precise definitions and statements will be given in the next section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let S be a smooth projective surface over k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' For n ≥ 2, let S[n] denote the Hilbert scheme n points on S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' It is well known that S[n] is a smooth projective variety of dimension 2n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In [PS20], the authors show that for char k > 3 and n ≥ 2, there is an isomorphism of affine group schemes over k π†(S, x)ab → π†(S[n], ˜ nx) where † = S, N or ´et.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In this note, we extend their results to the case of πalg and prove the following theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let char k > 3 and n ≥ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' There is an isomorphism of affine group schemes over k f : πalg(S, x)ab → πalg(S[n], ˜ nx) In section 2, we recall the definition of stratified bundles and some of their basic properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The formalism of Tannakian reconstruction is recalled in section 3 and used to define the algebraic fundamental group πalg(X, x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1 The geometrical properties of the Hilbert scheme of n points on a smooth projective surface are in section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In section 5, we prove a result about descent of stratified bundles which allows us to define the homomorphism f by defining the associated functor of Tannakian categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The concluding section 6 establishes the main theorem by showing that f is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Acknowledgements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We would like to thank Indranil Biswas and Ronnie Sebastian for their comments on earlier drafts of this note.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Stratified bundles Let k be a field of characteristic p and X be a noetherian scheme over k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Stratified bundles on X are sequences of coherent sheaves on X satisfying infinite Frobenius descent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' More precisely, the category of stratified bundles on X, denoted S(X), consists of Objects (Ei, αi) are sequences of coherent OX-modules Ei, i ∈ N along with isomorphisms αi : F ∗Ei+1 → Ei for all i ∈ N, where F is the absolute Frobenius on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Morphisms φ : (Ei, αi) → (Fi, βi) consists of a sequence of OX-module morphisms φi : Ei → Fi such that φi ◦ αi = βi ◦ F ∗ (φi+1) Let f : Y → X be a morphism and (Ei, αi) be a stratified bundle on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then we can define the pullback along f, denoted f ∗(Ei, αi), as consisting of the sequence of OY coherent sheaves f ∗Ei and isomorphisms are given by the composite maps F ∗f ∗Ei+1 γEi+1 −−−→ f ∗F ∗Ei+1 f ∗(αi) −−−−→ f ∗Ei where γ : F ∗f ∗ → f ∗F ∗ is the natural isomorphism of functors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus S(X) is contravariant functor in X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' One also has a tensor product on S(X) defined by taking term by term tensor product.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Also S(X) is an abelian category [cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [BHdS21], Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We recall some well known results about stratified bundles [cf [dS07], [Gie75]].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' If (Ei, αi) is a stratified bundle on X, then Ei is a locally free OX-module for all i ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This allows us to define duals of stratified bundles, making S(X) into an abelian rigid tensor category.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The rank of a stratified bundle (Ei, αi) is defined to be the rank of E0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The trivial stratified bundles on X are of the form ⊕(OX, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' F ∗, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let DX be the sheaf of differential operators on X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The category of DX modules consists of Objects coherent OX modules E equipped with a DX action i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='e a morphism of OX-algebras DX → Endk(E) Morphisms OX-linear maps E → F compatible with the DX action A theorem of Katz [[Gie75], Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='3] shows that for X smooth over k, then the category of stratified bundles on X and the category of DX modules are tensor equivalent to each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We close this section with the definition of G equivariant stratified bundles on a variety X admitting action of a group G on it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Definition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' A stratified bundle (Ei, αi) is said to be a G-equivariant stratified bundle if Ei are G-equivariant vector bundles and αi are G-equivariant OX module morphisms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Tannakian categories and fundamental group schemes In this section we recall the definition and basic properties of Tannakian categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We then recall Gieseker’s definition of the fundamental group scheme πalg using the Tannakian formalism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Tannakian Categories and affine group schemes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Tannakian categories were defined and studied in [DM82] to formalize the properties of Repk(G), the category of finite dimensional k-representations of G, an affine group scheme over k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Definition (Neutral Tannakian Categories).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' A rigid abelian tensor category C with End I = k is a neutral Tannakian category if it admits an exact faithful k-linear tensor functor ω : C → Veck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Any such functor is said to be a fiber functor for C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Given a neutral Tannakian category (C, ⊗, ω, I), we define the functor Aut⊗(ω) : k − algebra → Sets such that for k-algebra R, Aut⊗(ω)(R) consists of the families (λX) for X ∈ ob(C), where λX is a R-linear automorphism of X ⊗ R such that λX1⊗X2 = λX1 ⊗ λX2, λI = idR, and λY ◦ (α ⊗ 1) = (α ⊗ 1) ◦ λX : X ⊗ R → Y ⊗ R for all morphisms α : X → Y .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Theorem (Main theorem for neutral Tannakian categories, [DM82], Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='11).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let (C, ⊗) be a rigid abelian tensor category such that k = End(I) and let ω : C → Veck be an exact faithful tensor functor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then The functor Aut⊗(ω) of k-algebras is represented by an affine group scheme G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The functor C → Repk(G) is an equivalence of tensor categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let (C, ⊗, ω, I) and (C′, ⊗, ω′, I′) be neutral Tannakian categories which correspond to the repre- sentation categories of the affine k group schemes G and G′ respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then any functor of Tannakian categories from C → C′ is induced by a unique morphism of affine k group schemes G′ → G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This theorem allows us to define many variants of fundamental groups of a scheme X by considering different Tannakian categories naturally associated with X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The following result is very useful in establishing a given morphism between affine group schemes is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Theorem ([DM82], Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='21).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let f : G → G′ be a homomorphism of group schemes over k and Rep (f) : Rep (G′) → Rep (G) be the corresponding functor of Tannakian categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then f is faithfully flat if and only if Rep (f) is fully faithful and has essential image closed under subobjects i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='e for V ′ ∈ Rep (G′) and suboject W ⊂ Rep (f)(V ′), there is a subobject W ′ ⊂ V ′ in Rep (G′) such that Rep (f)(W ′) ≃ W in Rep (G) f is closed immersion if and only if every object of Rep (G) is a subquotient of some object in the essential image of Rep (f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We finish by recalling a basic result on affine group schemes (we refer to section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1 in [PS20] for details).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let G be a affine group scheme over k, Gab be its abelianization (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='e the maximal abelian quotient of G) and α : G → Gab be the (faithfully flat) quotient morphism .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We can then define the composite morphism φ : Gn αn −−→ Gn ab m −→ Gab where m is the multiplication homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As Sn acts on the k-group scheme Gn, we can define the notion of a Sn-invariant group morphism ψ : Gn → H for any k-group scheme H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let G and H be two group schemes over k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' For an integer n ≥ 2, the set of Sn-invariant group morphisms Gn → H is in bijective correspondence with the set of group morphism Gab → H i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='e any morphism of k-group schemes φ : Gn → H which is Sn-invariant factors uniquely through a morphism ψ : Gab → H such that φ = ψ ◦ h 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The group scheme πalg(X, x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Classically, over C, the Riemann-Hilbert correspondence identifies the category of vector bundles equipped with integrable connections on a smooth connected projective variety X/C with the category of representations of the topological fundamental group πtop(X, x) for some chosen base point x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Via GAGA, this gives a purely algebraic description of the category of representations of the topological fundamental group π(X, x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This category (equipped with the fiber functor (E, ∇) → Ex) is a neutral Tannakian category and can be identified, via the Tannakian formalism, with the representation 3 category of the proalgebraic completion of the topological fundamental group, denoted as πtop(X, x)alg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Over a field k of characteristic 0, the category of flat connections on a smooth variety X is tensor equiva- lent to the category of DX-modules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' However over a field of characteristic p, the category of flat connections on X is not as well behaved as the category of DX-modules and one defines a fundamental group scheme for X by Tannakian formalism using the category of DX-modules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By Katz’s theorem mentioned before, the fundamental group coincides with the one defined using S(X) below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let x ∈ X(k) be a k-rational point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then the abelian rigid tensor category S(X) is neutralized by the fiber functor Tx : S(X) → V eck The fundamental group scheme defined by the neutral Tannakian category (S(X), ⊗, Tx, (OX, F ∗)) is called the algebraic fundamental group of X based at x and is denoted by πalg(X, x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The following basic properties of πalg are well known.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' (Independence of basepoint) Let X be a geometrically connected, smooth projective k-scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then for all x1, x2 ∈ X(k), one has πalg(X, x1) ≃ πalg(X, x2) (Product rule) For X1, X2 geometrically connected and smooth over k and xi ∈ Xi(k), there is an isomorphism πalg(X1, x1) × πalg(X2, x2) → πalg(X1 × X2, (x1, x2)) For X smooth and open immersion U i−→ X such that the complement of U in X has codimension ≥ 2 and x ∈ U(k), then the homomorphism πalg(U, x) → πalg(X, x) associated to the restriction functor i∗ : S(X) → S(U) is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Geometry of Hilbert Scheme of points Let S be a smooth projective surface over k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We fix notation as follows Sn denotes the n-fold cartesian product of S with itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' S(n) denotes the nth symmetric product of S defined as the quotient Sn/Sn, where Sn denotes the symmetric group on n letters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' S[n] denotes the Hilbert scheme of n points on S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let ρ : Sn → S(n) be the quotient map and h : S[n] → S(n) be the Hilbert-Chow morphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We write S(n) for the open subset of S(n) consisting of distinct points with S[n] := h−1(S(n) ) and Sn := ρ−1(S(n) ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The map hn,◦ : S[n] → S(n) is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We have the diagram: S[n] Sn S(n) hn ρn In general, Hilbert schemes of points on a projective variety display a lot of pathological features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' But in [Fog68] the author shows that, in the case of smooth projective surface S, S[n] is a smooth projective variety.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus, in this case, the Hilbert-Chow morphism h : S[n] → S(n) is a resolution of singularities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 4 One can consider S(n) as the set of effective 0-cycles of degree n on S(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In this case it is easy to see that S(n) admits a stratification by type, where the type of a 0-cycle y of degree n is a tuple (n1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr) where y can be written as y = Σr j=1njxj where xj are distinct points of S with multiplicities n1 ≥ n2 ≥ · · · ≥ nr, where nj are positive integers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let C(n1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr) denote the subset of S(n) consisting of points of the type (n1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let S(n) ∗ = C(1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1) � C(2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1) denote the open subset of S(n) consisting of points of type (1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1) and (2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let S[n] ∗ and Sn ∗ denote the preimage of S(n) ∗ under h and ρ respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We recall some basic properties below which we will need later (we refer to [Fog68], [PS20] for details).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The subsets C(n1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr) are nonsingular of dimension 2r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The closed subset S(n) \\ S(n) ∗ is of codimension ≥ 2 in S(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The closed subset S[n] \\ S[n] ∗ is of codimension 2 in S[n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The closed subset Sn \\ Sn ∗ is of codimension ≥ 4 in Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The closed subset S(n) ∗ \\ S(n) is of codimension 2 in S(n) ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' When characteristic of k ̸= 2, for y ∈ C(2, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1), the scheme theoretic fiber h−1(y) is isomorphic to P1 k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In fact, S[n] ∗ is the blowup of S(n) ∗ along C(2, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We end this section by recalling a result of Fogarty ([Fog77], Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' If L is a Sn-invariant line bundle on Sn, there exists a line bundle L′ on S(n) such that h∗L′ ≃ L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' It follows that L′ in the proposition is isomorphic to σ∗(L)Sn 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The functor between Tannakian categories Let S be a smooth projective surface over k and (Ei, αi) be a stratified bundle on S[n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Restricting to S[n] ∗ gives us a functor i∗ : S(S[n]) → S(S[n] ∗ ) which is a equivalence of categories as S[n] ∗ is the complement of a codimension 2 closed subset of S[n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Next we show that a stratified bundle on S[n] ∗ can be pushed forward under h to get a stratified bundle on S(n) ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' First we begin by a result on descent of vector bundles along the morphism h : S[n] ∗ → S(n) ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Similar results have been established by authors in [Ish83] and [PS20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Assume char k ̸= 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let E be a vector bundle on S[n] ∗ which restricts to trivial vector bundles on the fibers of h over S(n) ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then h∗E is a locally free OS(n) ∗ -module.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Moreover the natural map h∗h∗(E) → E is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let x ∈ S(n) ∗ be a point of type (2, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=', 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then by assumption, the fiber of h over x is isomorphic to P1 k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let J be the ideal sheaf of the closed subscheme h−1(x) and Ix be the ideal sheaf of the closed point x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We have J = IxOS[n] ∗ For all n ≥ 1, let Yn denote the closed subscheme of S[n] ∗ corresponding to the ideal sheaf J n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Consider the following short exact sequence of sheaves on S[n] ∗ 0 → J ⊗ E → E → E|Y1 → 0 5 Pushing forward by h, we get the following exact sequence of sheaves on S(n) ∗ h∗E → H0(Y1, E|Y1) → R1h∗(J ⊗ E) We claim that the completion of R1h∗(J ⊗ E) at the maximal ideal mx in OS(n) ∗ ,x is 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The proof uses the theorem of formal functions which says that (R1h∗(J ⊗ E))∧ ≃ lim ←− H1(Yn, J ⊗ E ⊗ OS[n] ∗ /J n) We prove by induction that H1(Yn, J ⊗E ⊗OS[n] ∗ /J n) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As Y1 ≃ P1 k, the sheaves J n/J n+1 are locally free.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' These sheaves are also globally generated over Y1 as we have the surjection mn x/mn+1 x ⊗O S(n) ∗ ,x OS[n] ∗ ≃ In x /In+1 x ⊗O S(n) ∗ OS[n] ∗ ։ J n/J n+1 As J n/J n+1 is locally free on Y1 ≃ P1 k and globally generated, it is a direct sum of line bundles each of which has degree ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus one gets the base case of induction from degree considerations, as H1(Y1, J ⊗ E ⊗ OS[n] ∗ /J = H1(Y1, J /J 2 ⊗ EY1) = 0 Assume that the claim is true for n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then the proof for n + 1 follows from the long exact sequence in cohomology attached to the short exact sequence of sheaves on Yn+1 0 → J n+1/J n+2 ⊗ E → J /J n+2 ⊗ E → J /J n+1 ⊗ E → 0 which gives us the exact sequence H1(Yn+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J n+1/J n+2 ⊗ E) → H1(Yn+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J /J n+2 ⊗ E) → H1(Yn+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J /J n+1 ⊗ E) We know H1(Yn,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J n+1/J n+2⊗E) = H1(Y1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J n+1/J n+2⊗E) = 0 (by degree consideration) and H1(Yn+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J /J n+1⊗ E) = H1(Yn,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J /J n+1 ⊗ E) = 0 (by induction hypothesis),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' thus we get H1(Yn+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J /J n+2 ⊗ E) = 0 Thus the stalk of R1h∗(J ⊗ E) at x is 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This shows that the natural map h∗E → H0(Y1, E|Y1) is surjective in a neighbourhood of x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let f1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=', fr be a basis of H0(Y1, E|Y1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let Spec(R) be an affine neighbourhood of x where the natural map is surjective and let ˜fi ∈ Γ(Spec(R), h∗E) = Γ(h−1(Spec(R)), E) be lifts of fi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Using ˜fi one defines a homomorphism O⊕r S[n] ∗ |h−1(Spec(R)) → E on h−1(Spec(R) which is a surjection (and hence an isomorphism) on Y1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As h is proper, there exists a smaller affine neighbourhood U of x over which there is an isomorphism O⊕r V ≃ E where V = h−1(U).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Applying h∗, we get (h∗OV )⊕r ≃ h∗E As S(n) ∗ is normal and h : S[n] ∗ → S(n) ∗ is birational with connected fibers, by a form of Zariski’s main theorem [cf [Har77], Corollary 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='3 and 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='4], we have that h∗OV ≃ OU and thus h∗E is locally free.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The natural morphism h∗h∗(E) → E is clearly an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' □ 6 Let VBS(n) ∗ be the category of locally free sheaves on S(n) ∗ and VBh S[n] ∗ be the category of locally free sheaves on S[n] ∗ which restrict to trivial vector bundles on the fibers of h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition 1 above gives us an equivalence of categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Assume char k ̸= 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The pushforward functor h∗ : VBh S[n] ∗ → VBS(n) ∗ is an equivalence of categories with the inverse given by h∗ : VBS(n) ∗ → VBh S[n] ∗ Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We observe that if E′ ≃ h∗(E), then E ≃ h∗E′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This shows that h∗ is essentially surjective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The natural map HomS(n) ∗ (h∗E, h∗F) → HomS[n] ∗ (E, F) is bijective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus h∗ is an equivalence of categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' □ Corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' For all E ∈ VBh S[n] ∗ , the natural map F ∗h∗(E) → h∗F ∗(E) is an isomorphism over S(n) ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As F ∗E is also an object of VBh S[n] ∗ , thus both sheaves are locally free of the same rank.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus it suffices to show that the natural map F ∗h∗(E) → h∗F ∗(E) is surjective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As F is faithfully flat on the smooth locus of S(n) ∗ , the claim holds on the smooth locus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let x ∈ S(n) ∗ be of type (2, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then the restriction of F ∗h∗(E) to x is naturally isomorphic to H0(Y1, E|Y1) and the restriction of h∗F ∗(E) to x is H0(Y1, F ∗(E|Y1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The restriction of the natural map to x is the map F ∗ : H0(Y1, E1) → H0(Y1, F ∗E1) which is surjective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' □ By Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='2 of [Gie75], we have that every stratified bundle on P1 k is trivial.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus the above results give us Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Assume char k ̸= 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let (Ei, αi) be a stratified bundle on S[n] ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then h∗(Ei) is locally free OS(n) ∗ -module for all i ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Moreover the natural map h∗h∗(Ei) → Ei is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Furthermore the natural map F ∗h∗(Ei) → h∗F ∗(Ei) is an isomorphism over S(n) ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This allows us to define the pushforward of a stratified bundle (Ei, αi) on S[n] ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The pushforward denoted h∗(Ei, αi) is given by the sequence of vector bundles h∗Ei for all i ∈ N and the isomorphisms are given by the composite F ∗h∗(Ei+1) ηEi+1 −−−→ h∗F ∗(Ei+1) h∗(αi) −−−−→ h∗(Ei) where η : F ∗h∗ → h∗F ∗ is the natural transformation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 7 Thus we get a functor h∗ : S(S[n] ∗ ) → S(S(n) ∗ ) h∗ is additive tensor functor as on the smooth locus S(n) we have the isomorphisms h∗((Ei, αi) ⊕ (Fi, βi))|S(n) ≃ h∗(Ei, αi)|S(n) ⊕ h∗(Fi, βi))|S(n) h∗((Ei, αi) ⊗ (Fi, βi))|S(n) ≃ h∗(Ei, αi)|S(n) ⊗ h∗(Fi, βi))|S(n) which extend to S(n) ∗ due to codimension reasons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The following commutative diagram shows that h∗h∗(Ei, αi) is isomorphic to (Ei, αi) as stratified bundles with the isomorphism given by the natural morphisms h∗h∗Ei → Ei.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' F ∗h∗h∗Ei+1 h∗F ∗h∗Ei+1 h∗h∗F ∗Ei+1 h∗h∗Ei F ∗Ei+1 F ∗Ei+1 Ei h∗ηEi+1 h∗h∗αi αi γh∗Ei+1 = Consider the pullback functor ρ∗ : S(S(n) ∗ ) → S(Sn ∗ ) which takes values in the category of Sn-equivariant stratified bundles on Sn ∗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Also we have the extension functor j∗ : S(Sn ∗ ) → S(Sn) which is an equivalence of categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Composing these functors together, we get a functor T : S(S[n]) → S(Sn) given by T = j∗ ◦ ρ∗ ◦ h∗ ◦ i∗ Clearly T is an additive tensor functor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Note that h∗ is fully faithful, ρ∗ : S(S(n) ∗ ) → S(Sn ∗ ) is fully faithful (as ρ : Sn → S(n) is finite ´etale) and j∗ : S(Sn ∗ ) → S(Sn) is an equivalence of categories (due to codimension reasons).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus T is fully faithful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Fix n distinct k-valued points x1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , xk ∈ S(k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let ˜x ∈ S[n] such that h(˜x) = σ(x1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , xn) = z ∈ S(n) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then the categories S(S[n]) and S(Sn) are neutralized by the respective fiber functors τ˜x : S(S[n]) → Veck (Ei, αi) �→ (E0)˜x τ(x1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=',xn) : S(Sn) → Veck (Fi, βi) �→ (F0)(x1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=',xn) If T ((Ei, αi)) = (Fi, βi) that we have natural isomorphisms (E0)˜x ≃ h∗(E0)z ≃ (F0)(x1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=',xn).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus we have a functor of Tannakian categories T : (S(S[n]), ⊗, τ˜x, (OS[n], d)) → (S(Sn), ⊗, τ(x1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=',xn), (OSn, d)) 8 which by the independence of basepoint property of S induces a functor of Tannakian categories T : (S(S[n]), ⊗, τ ˜ nx, (OS[n], d)) → (S(Sn), ⊗, τ(x,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=',x), (OSn, d)) and hence a morphisms of the associated fundamental group schemes ˜f : πalg(Sn, (x, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , x)) → πalg(S[n], ˜ nx) Note that by proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='2 we have πalg(Sn, (x, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , x)) ≃ πalg(S, x)n .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As T : (S(S[n]), ⊗, T ˜ nx, (OS[n], d)) → (S(Sn), ⊗, T(x,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=',x), (OSn, d)) takes stratified bundles on S[n] to Sn-equivariant stratified bundles on Sn and a Sn-equivariant stratified bundles on Sn corresponds to a Sn-invariant representation of πalg(S, x)n, by 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1, ˜f factors uniquely through f : πalg(S, x)ab → πalg(S[n], ˜ nx) 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Isomorphism of fundamental group schemes In this section, we show that f is an isomorphism of affine group schemes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We begin by proving a result about Sn-equivariant stratified line bundles on Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let (Li, αi) be a Sn-equivariant stratified line bundles on Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then there exists a stratified line bundle (Li, βi) such that ρ∗(Li, βi) ≃ (Li, αi) Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By Fogarty’s result mentioned above, for any Sn-equivariant line bundle Li there exists line bundle Li ≃ ρ∗LSn i such that ρ∗Li ≃ Li.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Pushing forward αi and taking Sn invariants we get the isomorphism ρ∗(F ∗Li+1)Sn ρ∗(αi)Sn −−−−−−→ ρ∗(Li)Sn We show that the natural homomorphism F ∗(ρ∗(Li)Sn) → (F ∗ρ∗(Li)Sn) → (ρ∗F ∗(Li)Sn) is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Pulling back the morphism under ρ, we get the commutative diagram ρ∗F ∗((ρ∗Li)Sn) ρ∗((ρ∗F ∗Li)Sn) F ∗Li F ∗Li = where the vertical morphisms are the natural morphism which are isomorphisms by Fogarty’s theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By pushing forward under ρ and taking Sn invariants we get that F ∗(ρ∗(Li)Sn) → (ρ∗F ∗(Li)Sn) is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We define βi to be the composite isomorphism F ∗(ρ∗(Li)Sn) → (ρ∗F ∗(Li)Sn) ρ∗(αi)Sn −−−−−−→ ρ∗(Li)Sn The commutative diagram also gives us that ρ∗(Li, βi) ≃ (Li, αi) □ 9 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Faithfully flat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Next we show that the morphism f is faithfully flat Proposition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The homomorphism f : πalg(S, x)ab → πalg(S[n], ˜ nx) is faithfully flat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By [[DM82] Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='21], this is equivalent to showing that the functor T : S(S[n]) → S(Sn) is fully faithful and the essential image of T is closed under taking subobjects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We already know that T is fully faithful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let E• = (Ei, αi) be a stratified bundle on S[n] and F• := T (E•) be the corresponding Sn-equivariant stratified bundle on Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' If F′ ⊂ F• is a Sn-equivariant stratified subbundle, then we need to show there exists E′ ⊂ E• such that T (E′ ) = F′ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The proof proceeds by induction on the rank of E•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' If rank E• = 1, the proof is immediate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let rank E• ≥ 2 Then the stratified bundles F• and F′ correspond to the representations πalg(Sn, (x, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , x) → πalg(S, x)ab → GL(V ) and πalg(Sn, (x, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , x) → πalg(S, x)ab → GL(V ′) respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As πalg(S, x)ab is an abelian affine group scheme over k, all its irreducible representations are one dimen- sional.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus one gets that the πalg(S, x)ab-module V/V ′ has a one dimensional quotient W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus there is a πalg(S, x)ab-module surjection V → W such that the kernel contains V ′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let L• be the Sn-equivariant stratified bundle corresponding to W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus we have a short exact sequence of Sn-equivariant stratified bundles 0 → K• → F• → L• → 0 where F′ ⊂ K•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By proposition 1 above, we know that Li := ρ∗LSn i is a line bundle on S(n) and ρ∗Li = Li We claim that the following complex of sheaves on S(n) ∗ is exact for all i ∈ N (1) 0 → (ρ∗Ki)Sn|S(n) ∗ → (ρ∗Fi)Sn|S(n) ∗ → (ρ∗Li)Sn|S(n) ∗ → 0 It is enough to show that (ρ∗Fi)Sn|S(n) ∗ → (ρ∗Li)Sn|S(n) ∗ is surjective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We note that (ρ∗Fi)Sn|S(n) ∗ = h∗(Ei|S[n] ∗ ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let C be the cokernel h∗(Ei|S[n] ∗ ) → (ρ∗Li)Sn|S(n) ∗ → C → 0 Pulling back under ρ, we get the following commutative diagram on Sn ∗ ρ∗h∗(Ei|S[n] ∗ ) ρ∗((ρ∗Li)Sn|S(n) ∗ ) ρ∗C 0 Fi Li|Sn ∗ 0 = = 10 The rows are exact and hence ρ∗C = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As ρ is surjective, this implies C = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus Ki := (ρ∗Ki)Sn|S(n) ∗ is locally free on S(n) ∗ Pulling back the exact sequence (1) under h, we get a short exact sequence of locally free sheaves on S[n] ∗ 0 → h∗Ki|S[n] ∗ → Ei|S[n] ∗ → ˜Li|S[n] ∗ → 0 where ˜Li := h∗Li.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As the complement of S[n] ∗ in S[n] is of codimension ≥ 2 and Ei, L are locally free, the surjective morphism Ei|S[n] ∗ → ˜Li|S[n] ∗ extends to a unique morphism τi : Ei → ˜Li.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' This is surjective as L is of rank 1 and τ := (τi) give a nonzero morphism of stratified bundles E• → ˜L• where ˜L• := h∗(ρ∗(L•)Sn).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let κ• be the kernel of the morphism E• → ˜L•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then T (κ•) = K•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus, by the induction hypothesis on rank, there exists a stratified subbundle E′ ⊂ κ• ⊂ E• such that T (E′ ) = F′ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' □ 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Closed immersion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' We begin by recalling a result from [PS20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let p ∈ S(n) be a point of type (n1, n2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let p′ i, for i = 1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' m be the points in the fiber h−1(p).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let A be the local ring OS(n),p and B be the semilocal ring OSn ⊗OS(n) A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Then B is a finite A module and BSn = A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Lemma 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' When char k > n1, any Sn-equivariant surjective B-module homomorphism f : M → N of finitely generated B modules descends to surjective A-module homomorphism of the Sn-invariants M Sn → N Sn This allows us to prove the following analogue of Proposition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='6 in [PS20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let E• = (Ei, αi) be a Sn-equivariant stratified bundle on Sn (1) Let p ∈ S(n) be a point of type (n1, n2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' If char k > n1, then the sheaf ρ∗ESn i is locally free in a neighbourhood of p for all i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' (2) Let U denote the largest open subset where ρ∗ESn i is locally free, then on ρ−1(U), the natural mor- phism ρ∗ρ∗ESn i → Ei is an isomorphism for all i ∈ N Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The first assertion is proved by induction on the rank of E•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' If E• is a Sn-equivariant stratified bundle of rank 1, then by proposition 1, ρ∗ESn i is locally free on S(n) for all i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' In general, as E• corresponds to a representation of the abelian group scheme πalg(S, x)ab, there exists a Sn-equivariant short exact sequence of locally free sheaves on Sn 0 → K• → E• → L• → 0 Pushing forward by ρ and taking Sn-invariants we get the exact sequence for all i 0 → ρ(Ki)Sn → ρ(Ei)Sn → ρ(Li)Sn We claim that the homomorphism on the right is surjective in the neighbourhood of a point p of type (n1, n2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , nr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Surjectivity can be checked after passing to a formal neighbourhood of p and thus reduces to lemma 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By induction hypothesis on rank, both ρ(Ki)Sn and ρ(Li)Sn are locally free on a neighbourhood of p and hence so is ρ(Ei)Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 11 The second assertion follows from the observation that the natural homomorphism ρ∗ρ∗ESn i → Ei is an isomorphism on ρ−1(S(n) ) as as ρ : Sn → S(n) is finite ´etale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' As the complement of Sn in ρ−1(U) is of codimension ≥ 2 and both sheaves are locally free on ρ−1(U), thus the natural morphism is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' □ Proposition 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let char k > 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The homomorphism f : πalg(S, x)ab → πalg(S[n], ˜ nx) is faithfully flat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By [[DM82], Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='21], it is enough to show that the functor T : S(S[n]) → S(Sn) is essentially surjective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Thus we want to show that for any Sn-equivariant stratified bundle E• on Sn, there exists a stratified bundle F• on S[n] such that T (F•) = E•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let U be the open subset of S(n) consisting of points of type (1, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1), (2, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' , 1), (3, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=', 1) and (2, 1, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=', 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' By assumption on characteristic of k and the previous proposition, we get that ρ∗ESn i is locally free on U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Also we have on ρ−1(U), the natural morphism ρ∗ρ∗ESn i → Ei is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Imitating proposition 1 above, this allows us to define a stratified bundle (ρ∗ESn i , βi) on U such that ρ∗(ρ∗ESn i , βi) ≃ E•.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Pulling back under h to h−1(U) (whose complement in S[n] has codimension ≥ 3) and extending to S[n], we get a stratified bundle F• such that T (F•) = E• □ As f is both faithfully flat and a closed immersion, we get the following theorem Theorem 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Let char k > 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The homomorphism f : πalg(S, x)ab → πalg(S[n], ˜ nx) is an isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' References [BHdS21] Biswas, Indranil, Ph`ung Hˆo Hai, and Jo˜ao Pedro Dos Santos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”On the fundamental group schemes of certain quotient varieties.” Tohoku Mathematical Journal 73, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 4 (2021): 565-595.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [BPS06] Biswas, Indranil, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Parameswaran, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Subramanian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Monodromy group for a strongly semistable principal bundle over a curve.” Duke Mathematical Journal 132, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1 (2006): 1-48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [DM82] Deligne, Pierre;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Milne, James (1982), ”Tannakian categories”, in Deligne, Pierre;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Milne, James;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Ogus, Arthur;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Shih, Kuang-yen (eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ), Hodge Cycles, Motives, and Shimura Varieties, Lecture Notes in Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 900, Springer, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 101–228 [dS07] Dos Santos, Jo˜ao Pedro Pinto.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Fundamental group schemes for stratified sheaves.” Journal of Algebra 317, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2 (2007): 691-713.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Fog68] Fogarty, John.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Algebraic families on an algebraic surface.” American Journal of Mathematics 90, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2 (1968): 511-521.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Fog77] Fogarty, John.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Line bundles on quasi-symmetric powers of varieties.” Journal of Algebra 44, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1 (1977): 169-180.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Gie75] Gieseker, David.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Flat vector bundles and the fundamental group in non-zero characteristics.” Annali della Scuola Normale Superiore di Pisa-Classe di Scienze 2, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1 (1975): 1-31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Har77] Hartshorne, Robin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Algebraic geometry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 52.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Springer Science & Business Media, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Ish83] Ishimura, Sadao.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”A descent problem of vector bundles and its applications.” Journal of Mathematics of Kyoto University 23, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1 (1983): 73-83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Lan11] Langer, Adrian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”On the S-fundamental group scheme.” In Annales de l’Institut Fourier, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 61, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2077-2119.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [Lan12] Langer, Adrian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”On the S-fundamental group scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' II.” Journal of the Institute of Mathematics of Jussieu 11, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 4 (2012): 835-854.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [N76] Nori, Madhav V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”On the representations of the fundamental group.” Compositio Mathematica 33, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 1 (1976): 29-41.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [N82] Nori, Madhav V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”The fundamental group-scheme.” Proceedings Mathematical Sciences 91, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 2 (1982): 73-122.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' [PS20] Paul, Arjun, and Ronnie Sebastian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Fundamental group schemes of Hilbert scheme of n points on a smooth projective surface.” Bulletin des Sciences Math´ematiques 164 (2020): 102898.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' 12 [SGA1] Grothendieck, Alexander, and Michele Raynaud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' ”Revˆetements´etales et groupe fondamental (SGA 1).” arXiv preprint math/0206203 (2002).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' The Institute of Mathematical Sciences, (HBNI), Chennai 600113.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content=' Email address: sauravhc@imsc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} +page_content='in 13' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4dE1T4oBgHgl3EQfAwIb/content/2301.02842v1.pdf'} diff --git a/6tE1T4oBgHgl3EQfTQPr/content/2301.03077v1.pdf b/6tE1T4oBgHgl3EQfTQPr/content/2301.03077v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7cd1f0e7bbdeb179fba1de8b0e9d8b6313ce0af5 --- /dev/null +++ b/6tE1T4oBgHgl3EQfTQPr/content/2301.03077v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dbb779335fecdcd6ec06bc7912a00cb901b6477d2608a67c0d39887038ed49f2 +size 353481 diff --git a/6tE1T4oBgHgl3EQfTQPr/vector_store/index.pkl b/6tE1T4oBgHgl3EQfTQPr/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..bb40680b40d3688d059fe5f36893774f3bcccb87 --- /dev/null +++ b/6tE1T4oBgHgl3EQfTQPr/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9beccb9bc14dbdfba8de81f25eb2982df4100c2e46f24f816590559527bdd854 +size 146220 diff --git a/7NFLT4oBgHgl3EQfAC40/content/tmp_files/2301.11964v1.pdf.txt b/7NFLT4oBgHgl3EQfAC40/content/tmp_files/2301.11964v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..e51504b998f23e38457fc7cbaeac196a9caa01e1 --- /dev/null +++ b/7NFLT4oBgHgl3EQfAC40/content/tmp_files/2301.11964v1.pdf.txt @@ -0,0 +1,1078 @@ +Adversarial Networks and Machine Learning for +File Classification +Ken St. Germain1, Josh Angichiodo +Department of Cyber Science +United States Naval Academy +Annapolis, MD +1stgermai@usna.edu +Abstract—Correctly identifying the type of file under exam- +ination is a critical part of a forensic investigation. The file +type alone suggests the embedded content, such as a picture, +video, manuscript, spreadsheet, etc. In cases where a system +owner might desire to keep their files inaccessible or file type +concealed, we propose using an adversarially-trained machine +learning neural network to determine a file’s true type even +if the extension or file header is obfuscated to complicate its +discovery. Our semi-supervised generative adversarial network +(SGAN) achieved 97.6% accuracy in classifying files across +11 different types. We also compared our network against a +traditional standalone neural network and three other machine +learning algorithms. The adversarially-trained network proved +to be the most precise file classifier especially in scenarios +with few supervised samples available. Our implementation of +a file classifier using an SGAN is implemented on GitHub +(https://ksaintg.github.io/SGAN-File-Classier/). +I. INTRODUCTION +M learning can be used to determine file types based on +a file’s byte value distribution. In this work, we introduce an +adversarial learning approach to accurately identify file types +regardless of file extension, headers, or footers. By inspecting +the histogram-based distribution of byte values in a file, we +can greatly reduce the time and effort expended by subject +matter experts during the course of a forensic investigation. +Machine learning algorithms are designed to extract relevant +information from data [1], and the field of deep learning has +been shown effective in solving classification problems [2]. In +this paper we use a generative adversarial network (GAN) to +determine the type of file under investigation. Specifically, we +employ a GAN model with semi-supervised learning known +as a semi-supervised GAN (SGAN) [3] where only a small +portion of the training dataset is labeled. +A. Hiding files +Privacy advocates [4] urge users to protect their private in- +formation from criminal interception or unlawful government +overreach, and protecting the digital data stored on users’ +computers, phones, and other devices can include denying +physical access or employing encryption. While encryption has +become more commonplace and accessible [5], users desiring +more security against cryptographic weaknesses [6], [7] may +apply additional measures to safeguard their information. +By changing file extensions or removing them altogether, a +user can obfuscate the true file type. While this rudimentary +technique applied to a small number of files may not be a +challenge to computer forensic investigators, it may be more +effective if used across a large body of files composed of +varying types. +Many operating systems will select (or suggest) an appli- +cation to open a file based on the file extension [8]. For +example, Microsoft Windows will use the file extension, such +as .docx to determine the application to open the file. A +file named cat.docx suggests that the file is a document +that can be opened by Microsoft Word. However, users can +change the names and extensions of the file to any arbitrary +string of characters. A file originally created as a bitmap file +named cat.bmp and renamed to cat.docx will not open +and render correctly using Microsoft Word. +There are a variety of reasons to keep the nature of a +file unknown to all but the user. By obfuscating file types, +malware developers may hope to evade email filters or anti- +virus software [9] [10]. A user engaged in illicit activities may +desire to hinder law enforcement by complicating evidence +discovery [11]. Whatever the user’s motivation, without the +correct file extension and absent a brute-force approach, an +investigator will require a tool to efficiently discover the +appropriate program to open the file. +B. Finding files +Many file types can be determined by examining the file +header and footer information, also known as a “magic num- +ber”. The file header is the first few bytes in a file and the +footer is the last few bytes in a file. Depending on file type, +the file headers and footers will be of various lengths and have +different values. Many file types will have unique headers and +footers, yet some file types will share header and footer values, +e.g., .xls, .doc, .ppt [12]. +File headers and footers can be analyzed through command- +line tools that perform a binary or hexadecimal dump, or +by using binary or hexadecimal readers/editors to provide +insight to the file type. Alternatively, tools like Scalpel [13] +search a chunk of data that may contain multiple files, and +based on user-configured options, will perform file carving +that allows the investigator to see the chunk’s number and file +types within. Scalpel’s configurable options use header and +footer values as well as common signatures within a file. For +arXiv:2301.11964v1 [cs.LG] 27 Jan 2023 + +example, although an html file is plaintext and will not have +a header, it will likely include the text string . +Regardless of an investigator’s methods, specialized knowl- +edge is required to conclude the type of file under examination. +If the hexadecimal string D0 CF 11 E0 A1 B1 1A E1 is +found in the header, this could be one of five Microsoft Office +file types [12]. When several thousand or more files require +classification, the time demand on the most experienced in- +vestigator greatly increases. +C. Contributions +This work uses machine learning algorithms trained on +extracted file features to identify the type of file under in- +vestigation. We created histograms based on the frequency of +byte-values (ranging from zero to 255) to train and then test +our machine learning algorithms. Specifically, our contribution +provides: +• A classifier from a semi-supervised generative adversarial +network designed to identify file types +• Comparison of classifier accuracy with the performance +of a traditionally-trained multi-layer perceptron (MLP) +network +• Comparison and analysis of the neural network method +compared to the results from non-neural network machine +learning algorithms, specifically Decision Tree, extreme +gradient boosting (XGBoost), and k-Nearest Neighbor +(kNN) +To the best of our knowledge, no other work has used a +classifier of an adversarially-trained neural network to conduct +file type classification. We show improved accuracy over pre- +viously explored methods can be achieved with reduced expert +analysis required to create samples for a training dataset. +This paper provides background and discussion of related +works in Section II. We then discuss our dataset and how +we derive our samples for machine learning in Section III. +We present our SGAN architecture in Section IV and discuss +other machine learning algorithms in Section V. The results +of our work are summarized in Section VI and we provide our +conclusions and future work in Section VII. +II. BACKGROUND +This section examines previous work in file classification +and introduces the SGAN. We summarize the use of byte +values within files to determine file types and we discuss +the use of machine learning in file classification. Finally, we +discuss the nature of adversarial networks and examine the +SGAN model. +A. Classification using byte values +As an alternative to header and footer inspection, McDaniel +and Heydari used the binary content of files to identify the +type in [14]. They used several algorithms based on a byte +frequency distribution fingerprint to determine a file type, +showing that file classification can be accomplished by com- +paring a candidate file’s byte distribution to the distribution of +120 other files of known type. The accuracy of their proposed +algorithms was just under 96% when they grouped together +.acd, .doc, .xls, and .ppt file types into one class. +When these files were separately classified, the accuracy rate +dropped to 85%. Based on the binary frequency distribution +in [14], several authors have extended the research on file +classification. +In [15], Li et al. were able to improve on McDaniel and +Heydari’s accuracy in [14] using a centroid-based approach +and saw improved accuracy when truncating the files. Li used +the Manhattan Distance for each files’ byte distributions to +compare files and determine the appropriate classification. Be- +cause of file header similarity, Li created centroid models that +combined file types similar to McDaniel’s approach in [14]. +Specifically, there was one model that combined .exe and +.dll files into one class, and another model that combined +.doc, .xls, and .ppt files together in another class. +Moody and Erbacher introduced the Statistical Analysis +Data Identification (SADI) algorithm in [16]. After calculating +byte values for each file, a range of statistical information +was gathered and subsequently used to determine file types. +The accuracy of SADI had varying success with nine different +file types, reaching 76% accuracy of all file types after initial +analysis. A secondary assessment on file types that previously +did not reach greater than 92% accuracy showed improvement +when characteristic patterns were considered. +Using fragments of .pdf, .rtf, and .doc files from a +publicly-available dataset [17], Rahmet et al. leveraged longest +common sub-sequences to identify file fragments in [18]. The +authors’ algorithm successfully classified these file fragments +with 92.91% overall accuracy. +Our work extends the efforts discussed here, and we also +made use of byte values and the frequency in which they arose +in a file. The byte value distribution was provided to machine +learning algorithms, and each file type was classified. While +we also used file types that shared the same header strings +and files that did not contain headers, we created models that +differentiated the files uniquely instead of choosing to group +them together. +B. Machine learning for file classification +In [19], Amirani et al. used principle component analysis +(PCA) and neural networks to achieve file classification accu- +racy of 98.33% against a pool of six different file types. The +authors used two neural networks: a five-layer MLP network +that uses PCA features as the input, and a second three-layer +MLP network to conduct file classification. Each of their six +file types were equally represented in the dataset, with 120 +files of each type. +Konaray et al. conducted several experiments using a variety +of machine learning algorithms in [20]. The dataset used by +Konaray were composed of 13 text-based files (e.g., .html, +.py, .bat, etc.). The authors were able to achieve an +accuracy of 97.83% using the XGBoost algorithm [21]. +Comparing statistical classification algorithms such as sup- +port vector machine (SVM) and kNN with commercially +available tools, Gopal et al. showed that machine learning + +algorithms could outperform commercial products in [22]. The +authors collected byte values for their experiments using an +n-gram approach. They showed that kNN with 1-gram byte +values and SVM with 2-gram byte values greatly outperformed +commercial tools in terms of accuracy. +Inspired by the efforts in machine learning research, we +also hope to improve file classification accuracy. As we will +discuss in Section III, the dataset we used provided access to +more file types and of a wider variety than those mentioned +in the works here. The present work uses 11 types of files, +including some that are solely composed of ASCII characters +such as .txt and .html. In order to further research in this +domain, we investigated the SGAN-trained classifier. +C. Semi-supervised GAN +Semi-supervised learning requires that only a portion of the +training data be labeled. Semi-supervised learning differs from +supervised learning where all training data is labeled, and also +unsupervised learning, where no labels exist and the networks +must find their own way to organize the data. Semi-supervised +learning is valuable for large training data sets when it would +be laborious and time-intensive to manually label each file. +When training an unsupervised GAN, the discriminator, +D, is a two-class classifier that receives authentic samples +from the training dataset or spoofed samples created by +the generator, G. The generator uses random variable input +to create the fake samples and the parameters in G. The +discriminator assigns a probability from zero to one based on +its assessment that the sample is fake (0.0) or authentic (1.0). +The value function that describes this relationships from the +original work by Goodfellow [23] is given by +min +G max +D V (D, G) = Ex∼pdata(x)[log D(x)] ++ Ez∼pz(z)[log(1 − D(G(z)))] +(1) +where D(x) is the probability that x came from the data +distribution pdata(x) containing authentic training samples, +and D(G(z)) is the estimate of the probability that the dis- +criminator incorrectly identifies the fake instance as authentic. +The generator network attempts to maximize D(G(z)), while +the discriminator network tries to minimize it. The generator +creates samples, G(z), based on the parameter values in G +and the random values z provided to the generator consistent +with pz(z). +With semi-supervised learning, a small percentage of the +training data is labeled and the discriminator becomes a multi- +class classifier. For N classes, the model requires N + 1 +outputs to account for all the authentic classes plus one +additional class for the fake generated class. This can be +implemented in a variety of ways. Following Salimans et +al. [24], we can build an N-class classifier network, C, with +output logits {l1, l2, . . . , lN} prior to a softmax activation for +C. The logits vector is used as the input to a single perceptron +followed by the sigmoid activation function for D. The sigmoid +function is given as σ(z) = +1 +1+e−z , where z is the output +value of the discriminator output layer perceptron. Because +D and C share the same input and hidden layer weights, both +networks act as a single network, D/C, that is updated during +backpropagation based on their respective loss functions, J(D) +and J(C). The generator loss function is given by J(G). +Figure 1 shows a functional depiction of an SGAN in +training. The training dataset is partially labeled and provided +to the D/C model for classification by C. The remainder of the +training dataset as well as the generated samples from G are +used as input to D/C for discrimination where D will predict +whether the sample came from the training dataset or if it was +created by G. +III. DATASET +A dataset containing a variety of different files was desired +to ensure we could discern among a range of files. We used +Govdocs1, a publicly-available repository of about one million +files taken from webservers in the .gov domain [17]. The entire +Govdocs1 corpus consists of 1,000 directories, however we +only used the first three folders (000, 001, and 002) creating +a total dataset of 2,946 files, totaling 1.56 GB. We chose to +limit the dataset to ensure the processing demands would not +require exceptional computational resources. This work was +accomplished using a laptop computer with a 2.60 GHz Intel +i7 processor and 32 GB RAM. Limiting the dataset also allows +our work to be easily reproduced. +The dataset contained many common file types to include +.csv, .doc, .gif, .html, .jpg, .pdf, .txt, .xls, +etc. We noted an unequal distribution of these files such as 28 +.csv files, 254 .doc files, and 726 .pdf files. Unfortunately +there were some types that were especially underrepresented, +including one .gls file and two .java files. +A. Histograms +To capture byte value distributions, every file was converted +to a histogram. Each histogram contained 256 bins in the +range [0 , 255], representing the decimal value of each byte +in the file. For every bin, the frequency of that decimal value +occurring in the file was recorded. Histogram examples are +shown in Figure 2. In each plot, the bins are shown on the +N +classes +Unlabeled +samples +Labeled samples +Error +Error +[[Class1] +[Class2] +... +[ClassN]] +[0,1] +Noise +Fake samples +Fig. 1: Training a semi-supervised generative adversarial net- +work with N classes. + +0 +128 +255 +0 +200000 +400000 +.log +0 +128 +255 +0 +50 +100 +150 +.html +0 +128 +255 +0 +20 +40 +60 +80 +.html +0 +128 +255 +0 +200 +400 +600 +.html +0 +128 +255 +0 +20 +40 +60 +80 +.jpg +0 +128 +255 +0 +10000 +20000 +30000 +.doc +0 +128 +255 +0 +200 +400 +600 +.xml +0 +128 +255 +0 +20000 +40000 +.txt +0 +128 +255 +0 +500 +1000 +1500 +2000 +.html +0 +128 +255 +0 +500 +1000 +.txt +Fig. 2: Sample of histograms showing byte value distribution for various files. +horizontal axis while the frequency value is represented on the +vertical axis. +As Figure 2 shows, there are differences in the byte dis- +tribution between both files of the same type and files of +different types, but there are also similarities in different file +types such as .txt and .html files. Machine learning is an +appropriate tool to capture the histogram distributions and not +only differentiate among the different file types but also group +together files of matching type despite varying byte values. +B. Samples +After creating histograms for each file, we then processed +the histograms into samples. To ensure consistency across our +samples regardless of file size, we normalized each histogram, +scaling each to a cumulative distribution of 1.0. Figure 3 +shows the same .pdf file where Figure 3a is the original +and Figure 3b is normalized. +Since insufficient sample sizes for each class can precipi- +tate classification error [25], we removed the file types that +appeared less than 20 times, representing less than 0.7% of +the total. There were 14 different file types and 86 total +files removed, leaving our dataset with 11 classes and 2860 +samples. Our dataset’s composition is shown in Table I. The +sample order was then shuffled and finally split into training +0 +50 +100 +150 +200 +250 +0 +200 +400 +600 +800 +(a) +0 +50 +100 +150 +200 +250 +0.000 +0.025 +0.050 +0.075 +0.100 +0.125 +0.150 +0.175 +(b) +Fig. 3: Example histogram samples showing (a) an unscaled +.pdf file and (b) a normalized .pdf file. +and testing datasets. The training dataset used 80% of the total +samples, while the remaining 20% were reserved for testing. +IV. SGAN ARCHITECTURE +The adversarial competition in the SGAN is a minimax +game described by (1) where the discriminative model at- +tempts to correctly identify authentic training samples from +a distribution produced by the scaled histograms representing +the dataset files, pdata, and fake training samples created by +the generator. +While D and G adversarially train each other, they learn +to improve their individual performances. Additionally, C is +trained on labeled samples from the training dataset. Although +C does not directly receive unlabeled authentic or fake sam- +ples, the weights of C are affected by unsupervised training +since it shares weights with D in the D/C implementation. +The SGAN was implemented using the Python program- +ming language, Keras [26] front-end, and Tensorflow [27] +back-end. Additionally, Numpy, and Matplotlib Python li- +braries were used. The overall SGAN design is summarized +in Table II, with a total of 417,271 parameters for the dis- +criminator and the generator, and 304,779 parameters for the +classifier. The file size of the classifier was 3,634 KB. +The discriminator/classifier network, D/C, is a densely or +fully connected MLP deep neural network (DNN) with a single +input for the file histograms. Four additional fully connected +layers of size 512, 256, 128 and 64 are followed with rectified +linear unit (ReLU) activation functions. The ReLU function, +g is given by g(z) = max(0, z). The four hidden layers use +Dropout of 0.3 to prevent overfitting. Prior to the output layers, +a fully connected layer of size 11 is used to capture the number +of file types to be classified. The discriminator output layer of +size 1 is fully connected and uses a sigmoid activation function +to provide values [0.0, 1.0] as discussed in Section II-C. The +classifier output is a softmax activation connected to the 11 +TABLE I: Dataset file composition +file type +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +samples +28 +254 +40 +681 +229 +726 +207 +40 +486 +137 +32 + +TABLE II: SGAN architecture +Discriminator/Classifier: +layer +output size +activation +Input: x ∼ pdata(x) +256 +Fully Connected +512 +ReLU +Dropout = 0.3 +Fully Connected +256 +ReLU +Dropout = 0.3 +Fully Connected +128 +ReLU +Dropout = 0.3 +Fully Connected +64 +ReLU +Dropout = 0.3 +Fully Connected +11 ln = {l1, l2, . . . , l11} +Discriminator Output +1 +sigmoid +Classifier Output +11 +softmax +Generator: +layer +output +activation +Input: z ∼ pz(z) +100 +Fully Connected +32 +ReLU +Dropout = 0.3 +Fully Connected +64 +ReLU +Dropout = 0.3 +Fully Connected +128 +ReLU +Dropout = 0.3 +Fully Connected +256 +ReLU +Output +256 +sigmoid +node layer. The softmax function indicates the most likely class +to which the input belongs. The learning rate for D/C was +0.0005 using the Adam [28] optimizer and training was done +with batches of 32 samples. +The generator network, G, has a single input with 100 nodes +fully connected to the first hidden layer of size 32. Two +additional hidden layers of sizes 64 and 128 are again fully +connected using ReLU activations. Finally, a layer of size 256 +is connected to the output layer and sigmoid activation that +ultimately creates the fake histograms samples. The learning +rate for G was 0.0005 using the Adam optimizer. +V. MACHINE LEARNING ALGORITHMS +In order to illustrate the SGAN’s performance when clas- +sifying files, we used additional machine learning algorithms. +We assessed another neural network, the decision trees learn- +ing method, the XGBoost algorithm, and the nearest neighbors +algorithm. The same training and testing dataset were used +for each machine learning model. The SGAN was the most +complex to train due to using multiple neural networks and +no convergence to a global minima. +In terms of structure, the closest model to the SGAN is a +supervised learning-based neural network. We created an MLP +network with identical architecture to our SGAN classifier. +The standalone MLP network was trained in a fully supervised +manner to accurately select the correct file type based on input. +Both the SGAN and standalone MLP models were trained with +a batch size of 32 samples, and training was limited to no more +than 300 epochs. Following training, the best classifiers were +selected based on their accuracy against the training dataset. +These classifiers were then evaluated on the test dataset as +reported in Section VI. +Decision trees are a supervised learning approach that can +be used to accomplish multi-class classification [29]. Using +the features of the histograms, the decision tree algorithm +examines the parametric values in each sample and attempts +to accurately classify the file based on a series of decisions +based on learned thresholds. +The XGBoost algorithm was implemented as a classifier. +XGBoost is a supervised learning tool that can be used to +help us predict the correct file type category. With multiple +classes, the multi-class logistic loss function was used to train +the model. +Finally, the nearest neighbors classification algorithm com- +pares measurements of the input data and training data [29] +based on previously stored training information. The classifi- +cation result is determined by the number of samples selected, +k, with the smallest Euclidean distance among the sample +attributes. We iterated k from one to six to determine the most +appropriate number of neighbors to consider when deciding +the classification. +VI. RESULTS +Our results are summarized in Table III. The SGAN was +most accurate among all other machine learning algorithms +regardless of the number of supervised samples used in train- +ing. When using the entirety of the training data for training +the SGAN classifier, we achieved the highest classification +performance with the SGAN at 97.552% accuracy. Figure 4 +shows the confusion matrix of the SGAN when the classifier +had access to the entire training data. We see that the SGAN +performed worst at identifying .xml files at 83% accuracy, +confusing them with .html files. Looking over our dataset, +we note that .xml files were among the fewest number of +samples available for training. For some test samples, the +SGAN confused .ppt files with .doc files, and some .pdf +files were misidentified as .jpg files. The standalone MLP +network was nearly as accurate, reaching 96.15%. +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +Predicted label +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +True label +1.0 +1.0 +1.0 +0.97 +1.0 +0.96 +0.97 +1.0 +0.98 +1.0 +0.83 +SGAN File Classifier +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +Fig. 4: Confusion matrix for fully-supervised SGAN. + +TABLE III: Classification Accuracy +Number of +supervised samples +SGAN +Standalone +MLP +Decision Tree +XGBoost +kNN, k = 1 +kNN, k = 2 +kNN, k = 3 +kNN, k = 4 +kNN, k = 5 +kNN, k = 6 +2288 +0.97552 +0.96154 +0.90734 +0.90384 +0.88986 +0.82692 +0.874126 +0.83042 +0.85490 +0.81293 +1144 +0.93357 +0.92132 +0.86363 +0.87413 +0.86713 +0.79720 +0.84091 +0.75350 +0.81469 +0.76049 +500 +0.91783 +0.9021 +0.82168 +0.77972 +0.84965 +0.71504 +0.76573 +0.62063 +0.74650 +0.65734 +100 +0.87413 +0.81469 +0.48252 +0.65559 +0.71504 +0.44406 +0.61189 +0.48427 +0.52800 +0.38990 +50 +0.81993 +0.62062 +0.26573 +0.56818 +0.66084 +0.38112 +0.54895 +0.43007 +0.30944 +0.08741 +If we reduce the number of supervised samples provided +to our machine learning algorithms, we expect our testing +accuracy will be somewhat reduced. In the course of a forensic +investigation, subject matter expertise and a finite amount of +time must be prioritized, and since creating a fully-labeled +dataset is resource intensive, a worthy goal might be to balance +diminishing returns from further training a machine learning +algorithm against the time requirements needed for other tasks. +When drastically reducing the training input down to a sample +size of 50, only 2.2% of the training dataset, the SGAN +achieved 81.99% accuracy while the standalone MLP dropped +to 62.06% accuracy. The confusion matrices for this case are +shown in Figure 5. +Comparing Figure 5a and Figure 5b, we see that with +fewer samples, both neural networks continued to struggle +in categorizing .xml and .gif files. However, with 50 +supervised samples, both the SGAN and standalone MLP had +more confusion between .xml files and .html files. The +.gif files were incorrectly predicted to be .ppt files as +before, but also as .jpg files. We also see .xls files were +incorrectly categorized as .doc files, which is notable as +they are both Microsoft products and share the same header +information. +The decision tree, XGBoost, and kNN algorithms performed +relatively poorly with respect to classification accuracy com- +pared to the neural networks, especially as the number of +supervised samples were reduced. This is likely due to the +number of dimensions under assessment with our training +and testing samples. The “curse of dimensionality” [30] can +sometimes be overcome with enough samples, so reducing +a training dataset has a predictably deleterious effect on +performance. +The fully supervised SGAN model is implemented on +GitHub (https://ksaintg.github.io/SGAN-File-Classier/). Re- +searchers can make use of this implementation to test how +altering file headers, changing byte values, deleting portions +of files, etc. will affect classification accuracy. To determine if +the file headers would change the SGAN accuracy, we tested +our fully-supervised SGAN with a test dataset with altered file +headers. Except for .xml, .html, and .txt files which do +not make use of file headers, we replaced the first six bytes +of each test dataset file with the hexadecimal string AA BB +CC DD EE FF. The test accuracy for determining the file +type between files with altered and unaltered file headers was +nearly identical. Training for up to 300 epochs on a fully- +supervised training dataset with unaltered headers, there was +only a disparity in altered vs. unaltered test accuracy at epoch +135, where the difference was 0.14%. +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +Predicted label +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +True label +1.0 +0.8 +0.62 +0.81 +0.9 +0.93 +0.93 +0.85 +0.64 +0.65 +0.5 +SGAN File Classifier +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +(a) +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +Predicted label +.csv +.doc +.gif +.html +.jpg +.pdf +.ppt +.ps +.txt +.xls +.xml +True label +1.0 +0.86 +0.62 +0.67 +0.67 +0.75 +0.86 +0.46 +0.3 +0.39 +0.5 +Standalone MLP File Classifier +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +(b) +Fig. 5: Confusion matrices for (a) SGAN and (b) stan- +dalone MLP trained with 50 supervised samples. +. +VII. CONCLUSION AND FUTURE WORK +The adversarial training of a neural network produced +encouraging results in terms of classification accuracy. While +the neural networks were more complex to train than the +other machine learning algorithms, the accuracy results were +far superior. Though the SGAN was the most complex of all +the models, its accuracy was the best at correctly classifying +files based on their byte value distribution, especially with + +few supervised samples. Once trained, the time difference in +classifying the dataset between any of the algorithms was +indistinguishable. This work leads to future research using +additional neural network architectures and using our spoofed +histograms from the generator network to improve other +machine learning algorithms. +REFERENCES +[1] M. P. Deisenroth, A. A. Faisal, and C. S. Ong, Mathematics for Machine +Learning. +Cambridge University Press, Apr. 2020. +[2] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. +MIT +Press, 2016. [Online]. Available: http://www.deeplearningbook.org +[3] A. Odena, “Semi-Supervised Learning with Generative Adversarial +Networks,” arXiv:1606.01583 [cs, stat], Oct. 2016. [Online]. Available: +http://arxiv.org/abs/1606.01583 +[4] H. Fakhoury and D. Kayyali, “Know Your Rights,” Electronic Frontier +Foundation, Oct. 2014. [Online]. Available: https://www.eff.org/issues/ +know-your-rights +[5] A. Peterson, “Everything you need to know about encryption: Hint, +you’re already using it.” Washington Post, Dec. 2015. [Online]. Avail- +able: https://www.washingtonpost.com/news/the-switch/wp/2015/12/08/ +you-already-use-encryption-heres-what-you-need-to-know-about-it/ +[6] A. Barenghi, L. Breveglieri, I. Koren, and D. Naccache, “Fault Injection +Attacks on Cryptographic Devices: Theory, Practice, and Countermea- +sures,” Proceedings of the IEEE, vol. 100, no. 11, pp. 3056–3076, Nov. +2012. +[7] A. Mukherjee, S. A. A. Fakoorian, J. Huang, and A. L. Swindlehurst, +“Principles of Physical Layer Security in Multiuser Wireless Networks: +A Survey,” IEEE Communications Surveys Tutorials, vol. 16, no. 3, pp. +1550–1573, 2014. +[8] “Filename +extension +definition,” +The +Linux +Information +Project +(LINFO), Jul. 2006. [Online]. Available: http://www.linfo.org/filename +extension.html +[9] Microsoft, +“Blocked +attachments +in +Outlook.” +[Online]. +Avail- +able: +https://support.microsoft.com/en-us/office/blocked-attachments- +in-outlook-434752e1-02d3-4e90-9124-8b81e49a8519 +[10] B. +Collins, +“Don’t +Click +On +These +5 +Danger- +ous +Email +Attachments,” +Jan. +2021. +[Online]. +Avail- +able: https://www.forbes.com/sites/barrycollins/2021/01/16/dont-click- +on-these-5-dangerous-email-attachments/ +[11] L. Freeh, “The Impact of Encryption on Public Safety,” Statement +Before +the +Permanent +Select +Committee +on +Intelligence +United +States House of Representatives, Sep. 1997. [Online]. Available: +https://irp.fas.org/congress/1997 hr/h970909f.htm +[12] D. +Rentz, +“The +Microsoft +Compound +Document +File +Format,” +Aug. +2007. +[Online]. +Available: +http://www.openoffice.org/sc/ +compdocfileformat.pdf +[13] G. G. Richard III and V. Roussev, “Scalpel: A Frugal, High Performance +File Carver.” in The Digital Forensics Research Conference, New +Orleans, LA, USA, Aug. 2005. [Online]. Available: https://dfrws.org/ +presentation/scalpel-a-frugal-high-performance-file-carver/ +[14] M. McDaniel and M. Heydari, “Content based file type detection +algorithms,” in 36th Annual Hawaii International Conference on System +Sciences, 2003. Proceedings of the, Jan. 2003. +[15] W.-J. Li, K. Wang, S. Stolfo, and B. Herzog, “Fileprints: identifying file +types by n-gram analysis,” in Proceedings from the Sixth Annual IEEE +SMC Information Assurance Workshop, Jun. 2005, pp. 64–71. +[16] S. J. Moody and R. F. Erbacher, “S ´ADI - Statistical Analysis for +Data Type Identification,” in 2008 Third International Workshop on +Systematic Approaches to Digital Forensic Engineering, May 2008, pp. +41–54. +[17] S. Garfinkel, P. Farrell, V. Roussev, and G. Dinolt, “Bringing +science to digital forensics with standardized forensic corpora,” Digital +Investigation, vol. 6, pp. S2–S11, Sep. 2009. [Online]. Available: +https://linkinghub.elsevier.com/retrieve/pii/S1742287609000346 +[18] R. F. Rahmat, F. Nicholas, S. Purnamawati, and O. S. Sitompul, +“File Type Identification of File Fragments using Longest Common +Subsequence (LCS),” Journal of Physics: Conference Series, vol. 801, +p. 012054, Jan. 2017, publisher: IOP Publishing. [Online]. Available: +https://doi.org/10.1088/1742-6596/801/1/012054 +[19] M. C. Amirani, M. Toorani, and A. Beheshti, “A new approach to +content-based file type detection,” in 2008 IEEE Symposium on Comput- +ers and Communications, Jul. 2008, pp. 1103–1108, iSSN: 1530-1346. +[20] S. K. Konaray, A. Toprak, G. M. Pek, H. Akc¸ekoce, and D. Kılınc¸, +“Detecting File Types Using Machine Learning Algorithms,” in 2019 +Innovations in Intelligent Systems and Applications Conference (ASYU), +Oct. 2019, pp. 1–4. +[21] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” +in Proceedings of the 22nd ACM SIGKDD International Conference +on Knowledge Discovery and Data Mining, ser. KDD ’16. +New York, +NY, USA: Association for Computing Machinery, Aug. 2016, pp. +785–794. [Online]. Available: https://doi.org/10.1145/2939672.2939785 +[22] S. Gopal, Y. Yang, K. Salomatin, and J. Carbonell, “Statistical Learning +for File-Type Identification,” in 2011 10th International Conference on +Machine Learning and Applications and Workshops, vol. 1, Dec. 2011, +pp. 68–73. +[23] I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, +S. Ozair, A. Courville, and Y. Bengio, “Generative Adversarial Nets,” in +Proceedings of the 27th International Conference on Neural Information +Processing Systems - Volume 2, ser. NIPS’14. +Cambridge, MA, USA: +MIT Press, 2014, pp. 2672–2680. +[24] T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, +and +X. +Chen, +“Improved +Techniques +for +Training +GANs,” +in +Advances in Neural Information Processing Systems 29, D. D. Lee, +M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett, Eds. +Curran +Associates, Inc., 2016, pp. 2234–2242. [Online]. Available: http: +//papers.nips.cc/paper/6125-improved-techniques-for-training-gans.pdf +[25] S. Raudys and A. Jain, “Small sample size effects in statistical pattern +recognition: recommendations for practitioners,” IEEE Transactions on +Pattern Analysis and Machine Intelligence, vol. 13, no. 3, pp. 252–264, +Mar. 1991, conference Name: IEEE Transactions on Pattern Analysis +and Machine Intelligence. +[26] F. Chollet, et al., Keras, 2015, https://keras.io. [Online]. Available: +https://keras.io +[27] M. +Abadi, +et +al., +“TensorFlow: +Large-Scale +Machine +Learning +on +Heterogeneous +Systems,” +2015. +[Online]. +Available: +https:// +www.tensorflow.org/ +[28] D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” +Proceedings of the 3rd International Conference on Learning Represen- +tations (ICLR), Dec. 2014. +[29] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, +O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, +J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and +d. Duchesnay, “Scikit-learn: Machine Learning in Python,” Journal of +Machine Learning Research, vol. 12, no. 85, pp. 2825–2830, 2011. +[Online]. Available: http://jmlr.org/papers/v12/pedregosa11a.html +[30] K. Murphy, Probabilistic Machine Learning: An introduction. +MIT +Press, 2022. [Online]. Available: https://probml.github.io/pml-book/ +book1.html + diff --git a/7NFLT4oBgHgl3EQfAC40/content/tmp_files/load_file.txt b/7NFLT4oBgHgl3EQfAC40/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..df5c288126d598d5ff0509cd1712f27b53f6751f --- /dev/null +++ b/7NFLT4oBgHgl3EQfAC40/content/tmp_files/load_file.txt @@ -0,0 +1,800 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf,len=799 +page_content='Adversarial Networks and Machine Learning for File Classification Ken St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Germain1, Josh Angichiodo Department of Cyber Science United States Naval Academy Annapolis, MD 1stgermai@usna.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='edu Abstract—Correctly identifying the type of file under exam- ination is a critical part of a forensic investigation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The file type alone suggests the embedded content, such as a picture, video, manuscript, spreadsheet, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In cases where a system owner might desire to keep their files inaccessible or file type concealed, we propose using an adversarially-trained machine learning neural network to determine a file’s true type even if the extension or file header is obfuscated to complicate its discovery.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Our semi-supervised generative adversarial network (SGAN) achieved 97.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='6% accuracy in classifying files across 11 different types.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We also compared our network against a traditional standalone neural network and three other machine learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The adversarially-trained network proved to be the most precise file classifier especially in scenarios with few supervised samples available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Our implementation of a file classifier using an SGAN is implemented on GitHub (https://ksaintg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='io/SGAN-File-Classier/).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' INTRODUCTION M learning can be used to determine file types based on a file’s byte value distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In this work, we introduce an adversarial learning approach to accurately identify file types regardless of file extension, headers, or footers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' By inspecting the histogram-based distribution of byte values in a file, we can greatly reduce the time and effort expended by subject matter experts during the course of a forensic investigation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Machine learning algorithms are designed to extract relevant information from data [1], and the field of deep learning has been shown effective in solving classification problems [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In this paper we use a generative adversarial network (GAN) to determine the type of file under investigation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Specifically, we employ a GAN model with semi-supervised learning known as a semi-supervised GAN (SGAN) [3] where only a small portion of the training dataset is labeled.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Hiding files Privacy advocates [4] urge users to protect their private in- formation from criminal interception or unlawful government overreach, and protecting the digital data stored on users’ computers, phones, and other devices can include denying physical access or employing encryption.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' While encryption has become more commonplace and accessible [5], users desiring more security against cryptographic weaknesses [6], [7] may apply additional measures to safeguard their information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' By changing file extensions or removing them altogether, a user can obfuscate the true file type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' While this rudimentary technique applied to a small number of files may not be a challenge to computer forensic investigators, it may be more effective if used across a large body of files composed of varying types.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Many operating systems will select (or suggest) an appli- cation to open a file based on the file extension [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' For example, Microsoft Windows will use the file extension, such as .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='docx to determine the application to open the file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A file named cat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='docx suggests that the file is a document that can be opened by Microsoft Word.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' However, users can change the names and extensions of the file to any arbitrary string of characters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A file originally created as a bitmap file named cat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='bmp and renamed to cat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='docx will not open and render correctly using Microsoft Word.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' There are a variety of reasons to keep the nature of a file unknown to all but the user.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' By obfuscating file types, malware developers may hope to evade email filters or anti- virus software [9] [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A user engaged in illicit activities may desire to hinder law enforcement by complicating evidence discovery [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Whatever the user’s motivation, without the correct file extension and absent a brute-force approach, an investigator will require a tool to efficiently discover the appropriate program to open the file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Finding files Many file types can be determined by examining the file header and footer information, also known as a “magic num- ber”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The file header is the first few bytes in a file and the footer is the last few bytes in a file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Depending on file type, the file headers and footers will be of various lengths and have different values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Many file types will have unique headers and footers, yet some file types will share header and footer values, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=', .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' File headers and footers can be analyzed through command- line tools that perform a binary or hexadecimal dump, or by using binary or hexadecimal readers/editors to provide insight to the file type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Alternatively, tools like Scalpel [13] search a chunk of data that may contain multiple files, and based on user-configured options, will perform file carving that allows the investigator to see the chunk’s number and file types within.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Scalpel’s configurable options use header and footer values as well as common signatures within a file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' For arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='11964v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='LG] 27 Jan 2023 example, although an html file is plaintext and will not have a header, it will likely include the text string .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Regardless of an investigator’s methods, specialized knowl- edge is required to conclude the type of file under examination.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' If the hexadecimal string D0 CF 11 E0 A1 B1 1A E1 is found in the header, this could be one of five Microsoft Office file types [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' When several thousand or more files require classification, the time demand on the most experienced in- vestigator greatly increases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Contributions This work uses machine learning algorithms trained on extracted file features to identify the type of file under in- vestigation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We created histograms based on the frequency of byte-values (ranging from zero to 255) to train and then test our machine learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Specifically,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' our contribution provides: A classifier from a semi-supervised generative adversarial network designed to identify file types Comparison of classifier accuracy with the performance of a traditionally-trained multi-layer perceptron (MLP) network Comparison and analysis of the neural network method compared to the results from non-neural network machine learning algorithms,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' specifically Decision Tree,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' extreme gradient boosting (XGBoost),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' and k-Nearest Neighbor (kNN) To the best of our knowledge,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' no other work has used a classifier of an adversarially-trained neural network to conduct file type classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We show improved accuracy over pre- viously explored methods can be achieved with reduced expert analysis required to create samples for a training dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' This paper provides background and discussion of related works in Section II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We then discuss our dataset and how we derive our samples for machine learning in Section III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We present our SGAN architecture in Section IV and discuss other machine learning algorithms in Section V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The results of our work are summarized in Section VI and we provide our conclusions and future work in Section VII.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' BACKGROUND This section examines previous work in file classification and introduces the SGAN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We summarize the use of byte values within files to determine file types and we discuss the use of machine learning in file classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Finally, we discuss the nature of adversarial networks and examine the SGAN model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Classification using byte values As an alternative to header and footer inspection, McDaniel and Heydari used the binary content of files to identify the type in [14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' They used several algorithms based on a byte frequency distribution fingerprint to determine a file type, showing that file classification can be accomplished by com- paring a candidate file’s byte distribution to the distribution of 120 other files of known type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The accuracy of their proposed algorithms was just under 96% when they grouped together .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='acd, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls, and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt file types into one class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' When these files were separately classified, the accuracy rate dropped to 85%.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Based on the binary frequency distribution in [14], several authors have extended the research on file classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In [15], Li et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' were able to improve on McDaniel and Heydari’s accuracy in [14] using a centroid-based approach and saw improved accuracy when truncating the files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Li used the Manhattan Distance for each files’ byte distributions to compare files and determine the appropriate classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Be- cause of file header similarity, Li created centroid models that combined file types similar to McDaniel’s approach in [14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Specifically, there was one model that combined .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='exe and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='dll files into one class, and another model that combined .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls, and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt files together in another class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Moody and Erbacher introduced the Statistical Analysis Data Identification (SADI) algorithm in [16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' After calculating byte values for each file, a range of statistical information was gathered and subsequently used to determine file types.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The accuracy of SADI had varying success with nine different file types, reaching 76% accuracy of all file types after initial analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A secondary assessment on file types that previously did not reach greater than 92% accuracy showed improvement when characteristic patterns were considered.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Using fragments of .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='rtf, and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc files from a publicly-available dataset [17], Rahmet et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' leveraged longest common sub-sequences to identify file fragments in [18].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The authors’ algorithm successfully classified these file fragments with 92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='91% overall accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Our work extends the efforts discussed here, and we also made use of byte values and the frequency in which they arose in a file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The byte value distribution was provided to machine learning algorithms, and each file type was classified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' While we also used file types that shared the same header strings and files that did not contain headers, we created models that differentiated the files uniquely instead of choosing to group them together.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Machine learning for file classification In [19], Amirani et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' used principle component analysis (PCA) and neural networks to achieve file classification accu- racy of 98.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='33% against a pool of six different file types.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The authors used two neural networks: a five-layer MLP network that uses PCA features as the input, and a second three-layer MLP network to conduct file classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Each of their six file types were equally represented in the dataset, with 120 files of each type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Konaray et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' conducted several experiments using a variety of machine learning algorithms in [20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The dataset used by Konaray were composed of 13 text-based files (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=', .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='py, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='bat, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The authors were able to achieve an accuracy of 97.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='83% using the XGBoost algorithm [21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Comparing statistical classification algorithms such as sup- port vector machine (SVM) and kNN with commercially available tools, Gopal et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' showed that machine learning algorithms could outperform commercial products in [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The authors collected byte values for their experiments using an n-gram approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' They showed that kNN with 1-gram byte values and SVM with 2-gram byte values greatly outperformed commercial tools in terms of accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Inspired by the efforts in machine learning research, we also hope to improve file classification accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' As we will discuss in Section III, the dataset we used provided access to more file types and of a wider variety than those mentioned in the works here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The present work uses 11 types of files, including some that are solely composed of ASCII characters such as .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In order to further research in this domain, we investigated the SGAN-trained classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Semi-supervised GAN Semi-supervised learning requires that only a portion of the training data be labeled.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Semi-supervised learning differs from supervised learning where all training data is labeled, and also unsupervised learning, where no labels exist and the networks must find their own way to organize the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Semi-supervised learning is valuable for large training data sets when it would be laborious and time-intensive to manually label each file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' When training an unsupervised GAN, the discriminator, D, is a two-class classifier that receives authentic samples from the training dataset or spoofed samples created by the generator, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The generator uses random variable input to create the fake samples and the parameters in G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The discriminator assigns a probability from zero to one based on its assessment that the sample is fake (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0) or authentic (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The value function that describes this relationships from the original work by Goodfellow [23] is given by min G max D V (D, G) = Ex∼pdata(x)[log D(x)] + Ez∼pz(z)[log(1 − D(G(z)))] (1) where D(x) is the probability that x came from the data distribution pdata(x) containing authentic training samples, and D(G(z)) is the estimate of the probability that the dis- criminator incorrectly identifies the fake instance as authentic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The generator network attempts to maximize D(G(z)), while the discriminator network tries to minimize it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The generator creates samples, G(z), based on the parameter values in G and the random values z provided to the generator consistent with pz(z).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' With semi-supervised learning, a small percentage of the training data is labeled and the discriminator becomes a multi- class classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' For N classes, the model requires N + 1 outputs to account for all the authentic classes plus one additional class for the fake generated class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' This can be implemented in a variety of ways.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Following Salimans et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [24], we can build an N-class classifier network, C, with output logits {l1, l2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' , lN} prior to a softmax activation for C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The logits vector is used as the input to a single perceptron followed by the sigmoid activation function for D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The sigmoid function is given as σ(z) = 1 1+e−z , where z is the output value of the discriminator output layer perceptron.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Because D and C share the same input and hidden layer weights, both networks act as a single network, D/C, that is updated during backpropagation based on their respective loss functions, J(D) and J(C).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The generator loss function is given by J(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Figure 1 shows a functional depiction of an SGAN in training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The training dataset is partially labeled and provided to the D/C model for classification by C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The remainder of the training dataset as well as the generated samples from G are used as input to D/C for discrimination where D will predict whether the sample came from the training dataset or if it was created by G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' DATASET A dataset containing a variety of different files was desired to ensure we could discern among a range of files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We used Govdocs1, a publicly-available repository of about one million files taken from webservers in the .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gov domain [17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The entire Govdocs1 corpus consists of 1,000 directories, however we only used the first three folders (000, 001, and 002) creating a total dataset of 2,946 files, totaling 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='56 GB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We chose to limit the dataset to ensure the processing demands would not require exceptional computational resources.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' This work was accomplished using a laptop computer with a 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='60 GHz Intel i7 processor and 32 GB RAM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Limiting the dataset also allows our work to be easily reproduced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The dataset contained many common file types to include .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We noted an unequal distribution of these files such as 28 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv files, 254 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc files, and 726 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Unfortunately there were some types that were especially underrepresented, including one .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gls file and two .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='java files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Histograms To capture byte value distributions, every file was converted to a histogram.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Each histogram contained 256 bins in the range [0 , 255], representing the decimal value of each byte in the file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' For every bin, the frequency of that decimal value occurring in the file was recorded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Histogram examples are shown in Figure 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In each plot, the bins are shown on the N classes Unlabeled samples Labeled samples Error Error [[Class1] [Class2] .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [ClassN]] [0,1] Noise Fake samples Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1: Training a semi-supervised generative adversarial net- work with N classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 0 128 255 0 200000 400000 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='log 0 128 255 0 50 100 150 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html 0 128 255 0 20 40 60 80 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html 0 128 255 0 200 400 600 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html 0 128 255 0 20 40 60 80 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg 0 128 255 0 10000 20000 30000 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc 0 128 255 0 200 400 600 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml 0 128 255 0 20000 40000 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt 0 128 255 0 500 1000 1500 2000 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html 0 128 255 0 500 1000 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2: Sample of histograms showing byte value distribution for various files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' horizontal axis while the frequency value is represented on the vertical axis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' As Figure 2 shows, there are differences in the byte dis- tribution between both files of the same type and files of different types, but there are also similarities in different file types such as .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Machine learning is an appropriate tool to capture the histogram distributions and not only differentiate among the different file types but also group together files of matching type despite varying byte values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Samples After creating histograms for each file, we then processed the histograms into samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' To ensure consistency across our samples regardless of file size, we normalized each histogram, scaling each to a cumulative distribution of 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Figure 3 shows the same .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf file where Figure 3a is the original and Figure 3b is normalized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Since insufficient sample sizes for each class can precipi- tate classification error [25], we removed the file types that appeared less than 20 times, representing less than 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='7% of the total.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' There were 14 different file types and 86 total files removed, leaving our dataset with 11 classes and 2860 samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Our dataset’s composition is shown in Table I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The sample order was then shuffled and finally split into training 0 50 100 150 200 250 0 200 400 600 800 (a) 0 50 100 150 200 250 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='000 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='025 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='050 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='075 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='100 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='125 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='150 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='175 (b) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 3: Example histogram samples showing (a) an unscaled .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf file and (b) a normalized .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf file.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' and testing datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The training dataset used 80% of the total samples, while the remaining 20% were reserved for testing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' SGAN ARCHITECTURE The adversarial competition in the SGAN is a minimax game described by (1) where the discriminative model at- tempts to correctly identify authentic training samples from a distribution produced by the scaled histograms representing the dataset files, pdata, and fake training samples created by the generator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' While D and G adversarially train each other, they learn to improve their individual performances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Additionally, C is trained on labeled samples from the training dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Although C does not directly receive unlabeled authentic or fake sam- ples, the weights of C are affected by unsupervised training since it shares weights with D in the D/C implementation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The SGAN was implemented using the Python program- ming language, Keras [26] front-end, and Tensorflow [27] back-end.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Additionally, Numpy, and Matplotlib Python li- braries were used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The overall SGAN design is summarized in Table II, with a total of 417,271 parameters for the dis- criminator and the generator, and 304,779 parameters for the classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The file size of the classifier was 3,634 KB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The discriminator/classifier network, D/C, is a densely or fully connected MLP deep neural network (DNN) with a single input for the file histograms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Four additional fully connected layers of size 512, 256, 128 and 64 are followed with rectified linear unit (ReLU) activation functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The ReLU function, g is given by g(z) = max(0, z).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The four hidden layers use Dropout of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 to prevent overfitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Prior to the output layers, a fully connected layer of size 11 is used to capture the number of file types to be classified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The discriminator output layer of size 1 is fully connected and uses a sigmoid activation function to provide values [0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0] as discussed in Section II-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The classifier output is a softmax activation connected to the 11 TABLE I: Dataset file composition file type .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml samples 28 254 40 681 229 726 207 40 486 137 32 TABLE II: SGAN architecture Discriminator/Classifier: layer output size activation Input: x ∼ pdata(x) 256 Fully Connected 512 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 256 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 128 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 64 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 11 ln = {l1, l2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' , l11} Discriminator Output 1 sigmoid Classifier Output 11 softmax Generator: layer output activation Input: z ∼ pz(z) 100 Fully Connected 32 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 64 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 128 ReLU Dropout = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 Fully Connected 256 ReLU Output 256 sigmoid node layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The softmax function indicates the most likely class to which the input belongs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The learning rate for D/C was 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0005 using the Adam [28] optimizer and training was done with batches of 32 samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The generator network, G, has a single input with 100 nodes fully connected to the first hidden layer of size 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Two additional hidden layers of sizes 64 and 128 are again fully connected using ReLU activations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Finally, a layer of size 256 is connected to the output layer and sigmoid activation that ultimately creates the fake histograms samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The learning rate for G was 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0005 using the Adam optimizer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' MACHINE LEARNING ALGORITHMS In order to illustrate the SGAN’s performance when clas- sifying files, we used additional machine learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We assessed another neural network, the decision trees learn- ing method, the XGBoost algorithm, and the nearest neighbors algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The same training and testing dataset were used for each machine learning model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The SGAN was the most complex to train due to using multiple neural networks and no convergence to a global minima.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In terms of structure, the closest model to the SGAN is a supervised learning-based neural network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We created an MLP network with identical architecture to our SGAN classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The standalone MLP network was trained in a fully supervised manner to accurately select the correct file type based on input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Both the SGAN and standalone MLP models were trained with a batch size of 32 samples, and training was limited to no more than 300 epochs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Following training, the best classifiers were selected based on their accuracy against the training dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' These classifiers were then evaluated on the test dataset as reported in Section VI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Decision trees are a supervised learning approach that can be used to accomplish multi-class classification [29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Using the features of the histograms, the decision tree algorithm examines the parametric values in each sample and attempts to accurately classify the file based on a series of decisions based on learned thresholds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The XGBoost algorithm was implemented as a classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' XGBoost is a supervised learning tool that can be used to help us predict the correct file type category.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' With multiple classes, the multi-class logistic loss function was used to train the model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Finally, the nearest neighbors classification algorithm com- pares measurements of the input data and training data [29] based on previously stored training information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The classifi- cation result is determined by the number of samples selected, k, with the smallest Euclidean distance among the sample attributes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We iterated k from one to six to determine the most appropriate number of neighbors to consider when deciding the classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' VI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' RESULTS Our results are summarized in Table III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The SGAN was most accurate among all other machine learning algorithms regardless of the number of supervised samples used in train- ing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' When using the entirety of the training data for training the SGAN classifier, we achieved the highest classification performance with the SGAN at 97.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='552% accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Figure 4 shows the confusion matrix of the SGAN when the classifier had access to the entire training data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We see that the SGAN performed worst at identifying .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml files at 83% accuracy, confusing them with .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Looking over our dataset, we note that .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml files were among the fewest number of samples available for training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' For some test samples, the SGAN confused .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt files with .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc files, and some .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf files were misidentified as .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The standalone MLP network was nearly as accurate, reaching 96.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='15%.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml Predicted label .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml True label 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='97 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='96 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='97 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='98 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='83 SGAN File Classifier 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 4: Confusion matrix for fully-supervised SGAN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' TABLE III: Classification Accuracy Number of supervised samples SGAN Standalone MLP Decision Tree XGBoost kNN, k = 1 kNN, k = 2 kNN, k = 3 kNN, k = 4 kNN, k = 5 kNN, k = 6 2288 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='97552 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='96154 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='90734 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='90384 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='88986 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='82692 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='874126 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='83042 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='85490 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='81293 1144 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='93357 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='92132 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='86363 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='87413 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='86713 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='79720 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='84091 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='75350 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='81469 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='76049 500 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='91783 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='9021 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='82168 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='77972 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='84965 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='71504 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='76573 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='62063 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='74650 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='65734 100 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='87413 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='81469 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='48252 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='65559 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='71504 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='44406 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='61189 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='48427 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='52800 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='38990 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='81993 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='62062 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='26573 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='56818 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='66084 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='38112 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='54895 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='43007 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='30944 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='08741 If we reduce the number of supervised samples provided to our machine learning algorithms, we expect our testing accuracy will be somewhat reduced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' In the course of a forensic investigation, subject matter expertise and a finite amount of time must be prioritized, and since creating a fully-labeled dataset is resource intensive, a worthy goal might be to balance diminishing returns from further training a machine learning algorithm against the time requirements needed for other tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' When drastically reducing the training input down to a sample size of 50, only 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='2% of the training dataset, the SGAN achieved 81.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='99% accuracy while the standalone MLP dropped to 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='06% accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The confusion matrices for this case are shown in Figure 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Comparing Figure 5a and Figure 5b, we see that with fewer samples, both neural networks continued to struggle in categorizing .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' However, with 50 supervised samples, both the SGAN and standalone MLP had more confusion between .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml files and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif files were incorrectly predicted to be .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt files as before, but also as .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' We also see .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls files were incorrectly categorized as .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc files, which is notable as they are both Microsoft products and share the same header information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The decision tree, XGBoost, and kNN algorithms performed relatively poorly with respect to classification accuracy com- pared to the neural networks, especially as the number of supervised samples were reduced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' This is likely due to the number of dimensions under assessment with our training and testing samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The “curse of dimensionality” [30] can sometimes be overcome with enough samples, so reducing a training dataset has a predictably deleterious effect on performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The fully supervised SGAN model is implemented on GitHub (https://ksaintg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='io/SGAN-File-Classier/).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Re- searchers can make use of this implementation to test how altering file headers, changing byte values, deleting portions of files, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' will affect classification accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' To determine if the file headers would change the SGAN accuracy, we tested our fully-supervised SGAN with a test dataset with altered file headers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Except for .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html, and .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt files which do not make use of file headers, we replaced the first six bytes of each test dataset file with the hexadecimal string AA BB CC DD EE FF.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' The test accuracy for determining the file type between files with altered and unaltered file headers was nearly identical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Training for up to 300 epochs on a fully- supervised training dataset with unaltered headers, there was only a disparity in altered vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' unaltered test accuracy at epoch 135, where the difference was 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='14%.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml Predicted label .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml True label 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='62 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='81 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='9 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='93 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='93 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='85 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='64 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='65 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='5 SGAN File Classifier 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 (a) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml Predicted label .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='csv .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='doc .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='gif .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='jpg .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ppt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='ps .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='txt .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xls .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='xml True label 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='86 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='62 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='67 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='67 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='75 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='86 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='46 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='39 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='5 Standalone MLP File Classifier 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='0 (b) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 5: Confusion matrices for (a) SGAN and (b) stan- dalone MLP trained with 50 supervised samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' VII.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' CONCLUSION AND FUTURE WORK The adversarial training of a neural network produced encouraging results in terms of classification accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' While the neural networks were more complex to train than the other machine learning algorithms, the accuracy results were far superior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Though the SGAN was the most complex of all the models, its accuracy was the best at correctly classifying files based on their byte value distribution, especially with few supervised samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Once trained, the time difference in classifying the dataset between any of the algorithms was indistinguishable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' This work leads to future research using additional neural network architectures and using our spoofed histograms from the generator network to improve other machine learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' REFERENCES [1] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Deisenroth, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Faisal, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Ong, Mathematics for Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Cambridge University Press, Apr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [2] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Goodfellow, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Bengio, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Courville, Deep Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' MIT Press, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='deeplearningbook.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org [3] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Odena, “Semi-Supervised Learning with Generative Adversarial Networks,” arXiv:1606.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='01583 [cs, stat], Oct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/abs/1606.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='01583 [4] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Fakhoury and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Kayyali, “Know Your Rights,” Electronic Frontier Foundation, Oct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='eff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/issues/ know-your-rights [5] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Peterson, “Everything you need to know about encryption: Hint, you’re already using it.” Washington Post, Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Avail- able: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='washingtonpost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='com/news/the-switch/wp/2015/12/08/ you-already-use-encryption-heres-what-you-need-to-know-about-it/ [6] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Barenghi, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Breveglieri, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Koren, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Naccache, “Fault Injection Attacks on Cryptographic Devices: Theory, Practice, and Countermea- sures,” Proceedings of the IEEE, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 100, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 11, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 3056–3076, Nov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [7] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Mukherjee, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Fakoorian, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Huang, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Swindlehurst, “Principles of Physical Layer Security in Multiuser Wireless Networks: A Survey,” IEEE Communications Surveys Tutorials, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 16, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1550–1573, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [8] “Filename extension definition,” The Linux Information Project (LINFO), Jul.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='linfo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/filename extension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html [9] Microsoft, “Blocked attachments in Outlook.” [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Avail- able: https://support.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='microsoft.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='com/en-us/office/blocked-attachments- in-outlook-434752e1-02d3-4e90-9124-8b81e49a8519 [10] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Collins, “Don’t Click On These 5 Danger- ous Email Attachments,” Jan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Avail- able: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='forbes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='com/sites/barrycollins/2021/01/16/dont-click- on-these-5-dangerous-email-attachments/ [11] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Freeh, “The Impact of Encryption on Public Safety,” Statement Before the Permanent Select Committee on Intelligence United States House of Representatives, Sep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://irp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='fas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/congress/1997 hr/h970909f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='htm [12] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Rentz, “The Microsoft Compound Document File Format,” Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='openoffice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/sc/ compdocfileformat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf [13] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Richard III and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Roussev, “Scalpel: A Frugal, High Performance File Carver.” in The Digital Forensics Research Conference, New Orleans, LA, USA, Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://dfrws.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/ presentation/scalpel-a-frugal-high-performance-file-carver/ [14] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' McDaniel and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Heydari, “Content based file type detection algorithms,” in 36th Annual Hawaii International Conference on System Sciences, 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Proceedings of the, Jan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [15] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Li, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Wang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Stolfo, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Herzog, “Fileprints: identifying file types by n-gram analysis,” in Proceedings from the Sixth Annual IEEE SMC Information Assurance Workshop, Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2005, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 64–71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [16] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Moody and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Erbacher, “S ´ADI - Statistical Analysis for Data Type Identification,” in 2008 Third International Workshop on Systematic Approaches to Digital Forensic Engineering, May 2008, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 41–54.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [17] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Garfinkel, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Farrell, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Roussev, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Dinolt, “Bringing science to digital forensics with standardized forensic corpora,” Digital Investigation, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 6, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' S2–S11, Sep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://linkinghub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='elsevier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='com/retrieve/pii/S1742287609000346 [18] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Rahmat, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Nicholas, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Purnamawati, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Sitompul, “File Type Identification of File Fragments using Longest Common Subsequence (LCS),” Journal of Physics: Conference Series, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 801, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 012054, Jan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2017, publisher: IOP Publishing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='1088/1742-6596/801/1/012054 [19] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Amirani, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Toorani, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Beheshti, “A new approach to content-based file type detection,” in 2008 IEEE Symposium on Comput- ers and Communications, Jul.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2008, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1103–1108, iSSN: 1530-1346.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [20] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Konaray, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Toprak, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Pek, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Akc¸ekoce, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Kılınc¸, “Detecting File Types Using Machine Learning Algorithms,” in 2019 Innovations in Intelligent Systems and Applications Conference (ASYU), Oct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2019, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1–4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [21] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Chen and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' KDD ’16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' New York, NY, USA: Association for Computing Machinery, Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2016, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 785–794.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='1145/2939672.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='2939785 [22] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Gopal, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Yang, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Salomatin, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Carbonell, “Statistical Learning for File-Type Identification,” in 2011 10th International Conference on Machine Learning and Applications and Workshops, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1, Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2011, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 68–73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [23] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Goodfellow, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Pouget-Abadie, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Mirza, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Xu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Warde-Farley, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Ozair, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Courville, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Bengio, “Generative Adversarial Nets,” in Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2, ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' NIPS’14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Cambridge, MA, USA: MIT Press, 2014, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2672–2680.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [24] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Salimans, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Goodfellow, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Zaremba, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Cheung, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Radford, and X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Chen, “Improved Techniques for Training GANs,” in Advances in Neural Information Processing Systems 29, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Lee, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Sugiyama, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Luxburg, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Guyon, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Garnett, Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Curran Associates, Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=', 2016, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2234–2242.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: http: //papers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='nips.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='cc/paper/6125-improved-techniques-for-training-gans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='pdf [25] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Raudys and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Jain, “Small sample size effects in statistical pattern recognition: recommendations for practitioners,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 13, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 252–264, Mar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 1991, conference Name: IEEE Transactions on Pattern Analysis and Machine Intelligence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [26] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Chollet, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=', Keras, 2015, https://keras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='io.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://keras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='io [27] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Abadi, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=', “TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems,” 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https:// www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='tensorflow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/ [28] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Kingma and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Ba, “Adam: A Method for Stochastic Optimization,” Proceedings of the 3rd International Conference on Learning Represen- tations (ICLR), Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [29] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Pedregosa, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Varoquaux, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Gramfort, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Michel, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Thirion, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Grisel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Blondel, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Prettenhofer, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Weiss, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Dubourg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Vanderplas, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Passos, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Cournapeau, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Brucher, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Perrot, and d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Duchesnay, “Scikit-learn: Machine Learning in Python,” Journal of Machine Learning Research, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 12, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 85, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' 2825–2830, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: http://jmlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='org/papers/v12/pedregosa11a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html [30] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Murphy, Probabilistic Machine Learning: An introduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' MIT Press, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content=' Available: https://probml.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='io/pml-book/ book1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} +page_content='html' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/7NFLT4oBgHgl3EQfAC40/content/2301.11964v1.pdf'} diff --git a/7dAzT4oBgHgl3EQf-f5K/content/2301.01934v1.pdf b/7dAzT4oBgHgl3EQf-f5K/content/2301.01934v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e77489c0a598603cfc185a154a783974017dec40 --- /dev/null +++ b/7dAzT4oBgHgl3EQf-f5K/content/2301.01934v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87e1654535495ea9a22c317d942ef6d93439d80075a8e048b7c11e79f2579e3f +size 413180 diff --git a/7dAzT4oBgHgl3EQf-f5K/vector_store/index.pkl b/7dAzT4oBgHgl3EQf-f5K/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..54437befbf8c51ae6a524c12a73b9255de1bd065 --- /dev/null +++ b/7dAzT4oBgHgl3EQf-f5K/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8bf9e1b0b552c4b1ceea5149c6714f3678459dfc571e56f655e2ed4f9edf64c0 +size 156522 diff --git a/7tFJT4oBgHgl3EQfmyyq/content/2301.11589v1.pdf b/7tFJT4oBgHgl3EQfmyyq/content/2301.11589v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8dac90302e1dc62a3d38e6d9cd62220cb7c29697 --- /dev/null +++ b/7tFJT4oBgHgl3EQfmyyq/content/2301.11589v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c3aae8f93028f9716dfefaa7522dabf30fe3a061bc708fe0a2e940a5ba3abdd +size 461275 diff --git a/7tFJT4oBgHgl3EQfmyyq/vector_store/index.faiss b/7tFJT4oBgHgl3EQfmyyq/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..2e43ef19ef08a0533f1bbac26aa454e25c39816c --- /dev/null +++ b/7tFJT4oBgHgl3EQfmyyq/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36f496b05a3365bccccf816ceef6d00bb8bfed0d62d4ed1888df939575af695c +size 1900589 diff --git a/7tFJT4oBgHgl3EQfmyyq/vector_store/index.pkl b/7tFJT4oBgHgl3EQfmyyq/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..32543872a93fe247427592b9af7b646612923966 --- /dev/null +++ b/7tFJT4oBgHgl3EQfmyyq/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9c93f954a2ac5ec68d697a8ce6a675ff4f85d55eb02e89018f99a43df230d53 +size 85902 diff --git a/8tFQT4oBgHgl3EQfITXl/content/tmp_files/2301.13252v1.pdf.txt b/8tFQT4oBgHgl3EQfITXl/content/tmp_files/2301.13252v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..208a92880c30f4de83e3d2f7935851f94265c90c --- /dev/null +++ b/8tFQT4oBgHgl3EQfITXl/content/tmp_files/2301.13252v1.pdf.txt @@ -0,0 +1,595 @@ +Received 20 December 2022; +Revised 06 January 2023; +Accepted 06 January 2023 +DOI: xxx/xxxx +PROCEEDINGS +Equations of State for Dense Matter and Atrophysical Constraints +Rafael Bán Jacobsen1 | Verônica Dexheimer2 | Ricardo Luciano Sonego Farias1 +1Universidade Federal de Santa Maria +(UFSM), Santa Maria, Brazil +2Department of Physics, Kent State +University, Kent, OH 44243, USA +ABSTRACT: This conference proceeding presents an overview of the modern +approaches in the study of baryonic matter at high densities, focusing on the use of +online repositories such as CompOSE and MUSES for the calculation of neutron +star properties. In this context, relevant astrophysical constraints for the equations of +state (mass-radius relation, speed of sound, tidal deformability) are discussed. +KEYWORDS: +Neutron Star EoS, Dense matter, Astrophysical constraints, CompOSE, MUSES +1 +GENERAL ASPECTS OF THE +EQUATION OF STATE FOR DENSE MATTER +The study of the properties of compressed baryonic matter, +or, more specifically, strongly interacting matter at high densi- +ties, is a mostly relevant topic for current research in Physics, +with implications both in the microscopic and in the large scale +realms of nature. In the first domain, heavy-ion collision exper- +iments, such as those carried out by the Relativistic Heavy +Ion Collider (RHIC) at Brookhaven National Laboratory and +the Large Hadron Collider (LHC) at CERN, provide numerous +data on the behavior of baryonic matter at extreme condi- +tions of density and temperature. Additionally, in the second +domain, astronomic observations of neutron stars, from both +orbiting and ground based observatories, spanning the electro- +magnetic spectrum from 훾-rays to radio wavelengths and now +also including gravitational waves, can unveil significant prop- +erties of baryonic matter at high densities, since neutron stars +contain compressed baryonic matter in their centers. These +remnants of massive stars after core-collapse supernova explo- +sions are typically about 12 kilometers across and may contain +up to 2 solar masses (2푀⊙), implying core densities as high as +10 times nuclear saturation density (∼ 1015푔∕푐푚3). +In both cases, linking data to theoretical description of bary- +onic matter depends on the equation of state (EoS) adopted. +In a broad sense, an EoS is a thermodynamic equation relat- +ing state variables (and usually including the pressure). In the +specific field of nuclear astrophysics, it is also expected that +an EoS provides a full thermodynamic list of variables (e.g., +chemical potentials, entropy per baryon), particle composition +of the system (the proportion of the different types of lep- +tons, nucleons, and hyperons), microscopic information (e.g., +effective masses and pairing gaps) and stellar properties (e.g., +maximum mass and radius, tidal deformability). +EoS input tables for astrophysical simulations usually +includes baryon number density (푛퐵), charge fraction (푌푄), and +temperature (푇 ) as independent variables. A 1-dimensional +EoS table depends only on the parameter 푛퐵 and may describe +cold isospin-symmetric matter (푇 = 0 and 푌푄 = 0.5), cold +neutron matter (푇 = 0 and 푌푄 = 0.0), or cold 훽-equilibrated +matter (푇 = 0 and 푌푄 determined by the conditions of 훽- +equilibrium and charge neutrality). A 2-dimensional EoS table +depends on two of the three aforementioned independent vari- +ables and may describe, for example, dense matter at zero tem- +perature (varying 푛퐵 and 푌푄 with 푇 = 0), symmetric matter +(varying 푛퐵 and 푇 with 푌푄 = 0.5), neutron matter (varying 푛퐵 +and 푇 with 푌푄 = 0), and 훽-equilibrated matter (varying 푛퐵 and +푇 , and calculating 푌푄 according to 훽-equilibrium and charge +neutrality). Nonetheless, a 3-dimensional EoS table depends +on all three free parameters and serves for general purposes. +Namely, a 3-dimensional EoS table is required for supernova +and mergers simulations as long as, differently from neutron +stars, the matter in proto-neutron stars and in hypermassive +stars is hot and not 훽-equilibrated. +A complete EoS for neutron stars is expected to describe +a system with nuclei in the lower density regime, evolving to +bulk hadronic matter (nucleons, hyperons, deconfined quarks) +at higher densities. Inside neutron stars, this corresponds to the +crust and core, respectively (see Fig. 1 ). +arXiv:2301.13252v1 [astro-ph.HE] 30 Jan 2023 + +2 +V. Dexheimer, R.B. Jacobsen, R.L.S. Farias +FIGURE 1 Schematic structure of a neutron star. Figure +modified from Weber et al. (2014). +An EoS for dense and hot matter must be based on a quan- +tum relativistic description, because this framework ensures +respect to causality, as long as vector interactions are not too +strong. A realistic dense and hot EoS must also obey a series +of nuclear and quantum chromodynamics (QCD) constraints: +• To reproduce chiral symmetry restoration, as demanded +by QCD at large densities and temperatures (with a cor- +respondent decrease in the overall baryonic masses); +• To reproduce lattice QCD results at finite temperature +(which are provided at any isospin and strangeness, but +are restricted to low density relative to the temperature); +• To be in agreement with the (nearly) isospin-symmetric +and zero net strangeness heavy-ion collision physics at +finite temperature; +• To reproduce perturbative QCD results in the relevant +regime. +• To +reproduce +standard +zero-temperature +isospin- +symmetric nuclear physics results around saturation +density. +2 +MODERN SOURCES FOR EQUATIONS +OF STATE +In order to face the challenge of finding an adequate EoS for +dense matter in this variety of phenomena, online repositories +of equations of state have been built in recent years. CompOSE +and MUSES are among these modern sources for 1-, 2-, and +3-dimensional EoS tables. +2.1 +COMPOSE +CompOSE (CompStar Online Supernovae Equations of +State)1 is the largest repository of this kind, offering almost 300 +equations of state, divided in families (cold neutron star EoS, +cold matter EoS, neutron matter EoS, general purpose EoS, +and neutron star crust EoS) and their subgroups (models with +hyperons and delta resonances, hybrid quark-hadron models, +models with hyperons, models with kaon condensate, nucle- +onic models, and quark models). The repository also provides +a software to interpolate data, calculate additional quantities, +and graph EoS dependencies. Data tables, associated software +and the manual, can be freely downloaded, cf. Dexheimer et +al. (2022); Typel et al. (2022). +Paradigmatic examples of the usefulness of such a database +can be found in studies that carry out comparisons of the +predictions made by different models for the same phys- +ical system. For instance, a set of microscopic, covariant +density-functional, and non-relativistic Skyrme-type equations +of state, obtained from CompOSE, has been employed to +study the structure of purely nucleonic 훽-equilibrated neu- +tron stars at finite temperature (Wei, Burgio, Raduta, & +Schulze, 2021). Considering the agreement with presently +available astrophysical observational constraints, this study +showed that the magnitude of thermal effects depends on +the nucleon effective mass as well as on the stiffness of +the cold equation of state. Regarding the equations of state +themselves, an appropriate quantity to analyze in this con- +text is the relative thermal pressure, defined as 푝푟푎푡푖표 += +푝푡ℎ∕푝0 = [푝(휌퐵, 푥푇 , 푇 ) − 푝(휌퐵, 푥0, 0)] ∕푝(휌퐵, 푥0, 0), where 휌퐵 +is the baryonic density, 푇 is temperature and 푥0 and 푥푇 are +the respective proton fractions of cold and hot matter. The +ratio of thermal pressure as a function of density is shown +in the upper panel of Fig. 2 for the different equations of +state studied. Moreover, in order to appreciate the astrophysi- +cal implications of these equations of state, the relative change +of the maximum gravitational neutron-star mass, defined as +푀푟푎푡푖표 = (푀ℎ표푡 +푚푎푥 − 푀푐표푙푑 +푚푎푥 +) ∕푀푐표푙푑, can be plotted as a function +of the thermal pressure ratio. The result is shown in the lower +panel of Fig. 2 . +2.2 +MUSES +MUSES (Modular Unified Solver of the Equation of State)2 is +a large collaboration project that is developing a new cyber- +infrastructure to provide novel tools to answer critical inter- +disciplinary questions in nuclear physics, gravitational wave +astrophysics and heavy-ion physics. The MUSES collabora- +tion consists of many researchers and technical professionals +1https://compose.obspm.fr +2https://muses.physics.illinois.edu/ + +NeutronStar +Surface +.Hydrogen/Heliumplasma +Ironnuclei +OuterCrust +.lons +.Electrongas +InnerCrust +Heavyions +Relativisticelectrongas +Superfluidneutrons +OuterCore +.Neutrons,protons +.Electrons,muons +InnerCore +Neutrons +Superconductingprotons +Electrons,muons +Hyperons(Z,A,三) +Deltas (△) +Boson(元,K)condensates +Deconfined(u,d,s)quarks/color- +superconductingquarkmatterV. Dexheimer, R.B. Jacobsen, R.L.S. Farias +3 +��� +��� +��� +��� +��� +��� +��� +���� +��� +� +����� +������� +����� +��� +���� +���� +������� +������� +�������� +������ +����� +��������� +������� +������ +������ +������ +����� +����� +���� +���� +���� +���� +���� +���� +���� +���� +����� +����� +���� +���� +���� +���� +���� +���� +�� +�� +�� +� +�� +������������ +�������� +��������� +������������ +����������� +����������� +����������� +������������ +����������� +������������ +�������� +���������� +������� +������� +������� +������� +�������� +��������� +��������� +��������� +���������� +���������� +��������������� +�������������� +�������������� +�� +�� +�� +� +� +� +� +������ +������ +� +� +� +�� +� +�� +�� +�� +�� +�� +�� +�� +�� +FIGURE 2 Upper panel: Ratio of thermal pressure as a func- +tion of density for several equations of state. Lower panel: +Relative change of the maximum gravitational mass as a func- +tion of the pressure ratio at the center of the star. Figures +modified from Fig.6 and Fig.8 in (Wei et al., 2021). +across dozens of institutions spread across the globe who are +building and using a collaborative platform which is modular +because, while at low baryonic chemical potential the EoS is +known from first principles, at high there will be different mod- +els for the user to choose; besides, it is unified in as much as +different modules will be merged together to ensure maximal +coverage of the phase diagram. Building up MUSES, physi- +cists and computer scientists will work together to develop the +software that generates equations of state over large ranges of +temperature and chemical potentials to cover the whole QCD +phase diagram. The group of users is composed by interested +scientists from different communities, who provide input to the +future open-source cyberinfrastructure. +3 +ASTROPHYSICAL CONSTRAINTS +Any consistent EoS has to pass the test posed by the astrophys- +ical constraints related to neutron stars, the most fundamen- +tal being the mass-radius relation for these compact objects. +Nonetheless, many relevant features cannot be appreciated on +such a basis; for example, the possible existence of different +exotic matter associated with different phase transitions inside +a neutron star can easily be seen in the speed of sound (푐푆) +behavior but not necessarily in the mass-radius relation. As a +matter of fact, 2푀⊙ stars demand a stiff EoS (with 푐푆 +2 ←→ 1 +in natural units) at intermediate densities; on the other hand, +푐푆 +2 +←→ 1∕3 from below at asymptotically large densities +because of the conformal limit of massless free quarks. Thus, +a non-monotonic behavior is expected for 푐푆, implying the +occurrence of bumps related to the softening of the EoS due to +new degrees of freedom, cf.Bedaque & Steiner (2015). +Figure 3 , adapted from Tan, Dore, Dexheimer, Noronha- +Hostler, & Yunes (2022), show how bumps (that also appear +in realistic microscopic models) can be produced under a con- +trolled 푐푠 parametrization, allowing a correlation between the +density at which the bump appears and curves in the neutron +star mass-radius diagram. Thus, this more systematic para- +metric form for the speed of sound can help to determine +neutron-star composition; besides, maximum stellar mass and +radius can determine width, density, and height of the bumps. +The non-smooth structure of the speed of 푐푆 related to phase +transitions in dense matter makes feasible the constitution of +ultra-heavy neutron stars (with masses larger than 2.5푀⊙). +These stars pass all observational and theoretical constraints, +including those imposed by recent LIGO/Virgo gravitational- +wave observations and NICER X-ray observations. +Another observational test that may be used to constrain +equations of state is the evaluation of tidal deformabilities in +neutron stars inferred from gravitational-wave measurements. +In a coalescing binary of neutron stars, the gravitational field +of one star perturbs the field of the other (and vice-versa), caus- +ing an acceleration in their inspiral. This change in the inspiral +rate shapes the gravitational-wave emitted, and this wave thus +provides information about the tidal deformabilities Λ1,2 of +the neutron stars. Considering a sequence of central densi- +ties for a given EoS and a fixed mass ratio, one can construct +the binary Love relations (BLRs) Λ푠 and Λ푎, definined with +the symmetric and anti-symmetric tidal deformabilities Λ푠,푎 = +(Λ1 ±Λ2)∕2. Due to phase transitions and the consequent non- +smooth structure of the speed of sound 푐푆, which may tilt the +mass-radius diagram, peculiar structures (such as slopes, hills, +drops and swooshes) are created in the BLRs (Tan, Dexheimer, +Noronha-Hostler, & Yunes, 2022), as shown in Figure 4 . +The change in slope in the BLRs may be observable already +during the fifth LIGO observing run if a sufficiently loud and + +4 +V. Dexheimer, R.B. Jacobsen, R.L.S. Farias +ε=3p +causal limit +□ +●×× +▲ +□●××▲ +max central nB/nsat +(a) +eos1 +eos2 +eos3 +eos4 +0 +2 +4 +6 +8 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +1.2 +nB/nsat +cs +2 +J0740+6620 +GW190814 +GW170817 +J0030+0451 +(c) +10 +11 +12 +13 +14 +15 +16 +0.5 +1.0 +1.5 +2.0 +2.5 +3.0 +R [km] +M [M⊙] +ε=3p +causal limit +□ ● ×× ▲ +□●××▲ +max central nB/nsat +(a) +eos1 +eos2 +eos3 +eos4 +0 +2 +4 +6 +8 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +1.2 +nB/nsat +cs +2 +J0740+6620 +GW190814 +GW170817 +J0030+0451 +(c) +10 +11 +12 +13 +14 +15 +16 +0.5 +1.0 +1.5 +2.0 +2.5 +3.0 +R [km] +M [M⊙] +FIGURE 3 Upper panels: Speed of sound (left) and mass- +radius diagram (right) for a subfamily of equations of state with +peaks in the speed of sound of different widths at the same +location. Lower panels: Speed of sound (left) and mass-radius +diagram (right) for a subfamily of equations of state with peaks +in the speed of sound of the same width at different locations. +Modified from Fig.4 in Tan, Dore, et al. (2022). +low mass neutron-star binary is detected. The detection of +drops and swooshes is more challenging, because both occur +at very small Λ푎, and such detection would require very low +uncertainties in the measurements, which are achievable only +if an exceptionally loud signal is detected. +4 +CONCLUSIONS +From the recent developments here reported, one my infer that +new tight constraints from experiment, observation and the- +ory are slowly determining dense matter and neutron-star core +properties. In this context, EoS repositories (such as Com- +pOSE and MUSES) help speeding up the understanding of +dense matter. Furthermore, astrophysical constraints must be +taken into account and, in this context, gravitational waves are +providing new ways to study the dense matter EoS. Besides +the basic mass-radius relation of neutron stars, more specific +and subtle quantities (such as the speed of sound and tidal +deformabilities for these objects) can be used to probe differ- +ent equations of state. Advances in this field can be expected +shortly, since LIGO, Virgo, and KAGRA are coordinating a +new observing run in March 2023. Thus, open questions in +nuclear astrophysics may soon find their answers and induce +further interrogations about the intimate structure of matter. +EoS5 +EoS6 +EoS7 +EoS8 +0 +1 +2 +3 +4 +5 +6 +7 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +n/nsat +cs +2 +(a) +J0740+6620 +J0030+0451 +Am +IL/MD +Am +IL/MD +10 +11 +12 +13 +14 +15 +16 +0.8 +1.2 +1.6 +2. +2.4 +R [km] +M [M⊙] +(b) +type (B.1) +type (B.2) +60 +80 +100 +120 +140 +-100 +-50 +0 +50 +100 +Λs +Λa +(c) +type (A) +type (C) +0 +50 +100 +150 +200 +0 +2 +4 +6 +8 +10 +12 +Λs +Λa +(d) +FIGURE 4 Upper panels: Speed of sound (left) and mass- +radius diagram (right) for for different equations of state. +First-order phase transitions (푐푆 = 0) introduce a second sta- +ble branch in the mass-radius curves. Lower panels: BLRs +between stars (with a mass ratio 0.75) in the same branch (types +A and C) or in different branches (types B.1 and B.2) produce +a slope, hill, drop, and swoosh. Figure modified from Fig.1 +in Tan, Dexheimer, et al. (2022). +ACKNOWLEDGEMENTS +V. D. acknowledges support from the National Science Foun- +dation under grants PHY1748621, MUSES OAC-2103680, +and NP3M PHY-2116686. R.L.S.F. acknowledges support +from Conselho Nacional de Desenvolvimento Científico e +Tecnológico (CNPq), Grant No. 309598/2020-6 and Fundação +de Amparo à Pesquisa do Estado do Rio Grande do Sul +(FAPERGS), Grants Nos. 19/2551- 0000690-0 and 19/2551- +0001948-3. +REFERENCES +Bedaque, P., & Steiner, A. W. 2015, Phys. Rev. Lett., 114(3), 031103. +Dexheimer, V., Mancini, M., Oertel, M., Providência, C., Tolos, L., +& Typel, S. 2022, Particles, 5(3), 346. +Tan, H., Dexheimer, V., Noronha-Hostler, J., & Yunes, N. +2022, +Physical Review Letters, 128(16). +Tan, H., Dore, T., Dexheimer, V., Noronha-Hostler, J., & Yunes, N. +2022, Physical Review D, 105(2). +Typel, S., Oertel, M., Klähn, T. et al. 2022, Eur. Phys. J. A, 58(11), +221. +Weber, F., Contrera, G. A., Orsaria, M. G., Spinella, W., & Zubairi, +O. 2014, Modern Physics Letters A, 29, 1430022. +Wei, J.-B., Burgio, G. F., Raduta, A. R., & Schulze, H.-J. +2021, +Physical Review C, 104(6). + diff --git a/8tFQT4oBgHgl3EQfITXl/content/tmp_files/load_file.txt b/8tFQT4oBgHgl3EQfITXl/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..219960bbeb1b64424dd69889b8b8e15e45791125 --- /dev/null +++ b/8tFQT4oBgHgl3EQfITXl/content/tmp_files/load_file.txt @@ -0,0 +1,344 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf,len=343 +page_content='Received 20 December 2022;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Revised 06 January 2023;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Accepted 06 January 2023 DOI: xxx/xxxx PROCEEDINGS Equations of State for Dense Matter and Atrophysical Constraints Rafael Bán Jacobsen1 | Verônica Dexheimer2 | Ricardo Luciano Sonego Farias1 1Universidade Federal de Santa Maria (UFSM),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Santa Maria,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Brazil 2Department of Physics,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Kent State University,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Kent,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' OH 44243,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' USA ABSTRACT: This conference proceeding presents an overview of the modern approaches in the study of baryonic matter at high densities,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' focusing on the use of online repositories such as CompOSE and MUSES for the calculation of neutron star properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In this context, relevant astrophysical constraints for the equations of state (mass-radius relation, speed of sound, tidal deformability) are discussed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' KEYWORDS: Neutron Star EoS, Dense matter, Astrophysical constraints, CompOSE, MUSES 1 GENERAL ASPECTS OF THE EQUATION OF STATE FOR DENSE MATTER The study of the properties of compressed baryonic matter, or, more specifically, strongly interacting matter at high densi- ties, is a mostly relevant topic for current research in Physics, with implications both in the microscopic and in the large scale realms of nature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In the first domain, heavy-ion collision exper- iments, such as those carried out by the Relativistic Heavy Ion Collider (RHIC) at Brookhaven National Laboratory and the Large Hadron Collider (LHC) at CERN, provide numerous data on the behavior of baryonic matter at extreme condi- tions of density and temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Additionally, in the second domain, astronomic observations of neutron stars, from both orbiting and ground based observatories, spanning the electro- magnetic spectrum from 훾-rays to radio wavelengths and now also including gravitational waves, can unveil significant prop- erties of baryonic matter at high densities, since neutron stars contain compressed baryonic matter in their centers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' These remnants of massive stars after core-collapse supernova explo- sions are typically about 12 kilometers across and may contain up to 2 solar masses (2푀⊙), implying core densities as high as 10 times nuclear saturation density (∼ 1015푔∕푐푚3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In both cases, linking data to theoretical description of bary- onic matter depends on the equation of state (EoS) adopted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In a broad sense, an EoS is a thermodynamic equation relat- ing state variables (and usually including the pressure).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In the specific field of nuclear astrophysics, it is also expected that an EoS provides a full thermodynamic list of variables (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', chemical potentials, entropy per baryon), particle composition of the system (the proportion of the different types of lep- tons, nucleons, and hyperons), microscopic information (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', effective masses and pairing gaps) and stellar properties (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', maximum mass and radius, tidal deformability).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' EoS input tables for astrophysical simulations usually includes baryon number density (푛퐵), charge fraction (푌푄), and temperature (푇 ) as independent variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' A 1-dimensional EoS table depends only on the parameter 푛퐵 and may describe cold isospin-symmetric matter (푇 = 0 and 푌푄 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5), cold neutron matter (푇 = 0 and 푌푄 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0), or cold 훽-equilibrated matter (푇 = 0 and 푌푄 determined by the conditions of 훽- equilibrium and charge neutrality).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' A 2-dimensional EoS table depends on two of the three aforementioned independent vari- ables and may describe, for example, dense matter at zero tem- perature (varying 푛퐵 and 푌푄 with 푇 = 0), symmetric matter (varying 푛퐵 and 푇 with 푌푄 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5), neutron matter (varying 푛퐵 and 푇 with 푌푄 = 0), and 훽-equilibrated matter (varying 푛퐵 and 푇 , and calculating 푌푄 according to 훽-equilibrium and charge neutrality).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Nonetheless, a 3-dimensional EoS table depends on all three free parameters and serves for general purposes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Namely, a 3-dimensional EoS table is required for supernova and mergers simulations as long as, differently from neutron stars, the matter in proto-neutron stars and in hypermassive stars is hot and not 훽-equilibrated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' A complete EoS for neutron stars is expected to describe a system with nuclei in the lower density regime, evolving to bulk hadronic matter (nucleons, hyperons, deconfined quarks) at higher densities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Inside neutron stars, this corresponds to the crust and core, respectively (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 1 ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='13252v1 [astro-ph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='HE] 30 Jan 2023 2 V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Dexheimer, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Jacobsen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Farias FIGURE 1 Schematic structure of a neutron star.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Figure modified from Weber et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' An EoS for dense and hot matter must be based on a quan- tum relativistic description, because this framework ensures respect to causality, as long as vector interactions are not too strong.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' A realistic dense and hot EoS must also obey a series of nuclear and quantum chromodynamics (QCD) constraints: To reproduce chiral symmetry restoration, as demanded by QCD at large densities and temperatures (with a cor- respondent decrease in the overall baryonic masses);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' To reproduce lattice QCD results at finite temperature (which are provided at any isospin and strangeness, but are restricted to low density relative to the temperature);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' To be in agreement with the (nearly) isospin-symmetric and zero net strangeness heavy-ion collision physics at finite temperature;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' To reproduce perturbative QCD results in the relevant regime.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' To reproduce standard zero-temperature isospin- symmetric nuclear physics results around saturation density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2 MODERN SOURCES FOR EQUATIONS OF STATE In order to face the challenge of finding an adequate EoS for dense matter in this variety of phenomena, online repositories of equations of state have been built in recent years.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' CompOSE and MUSES are among these modern sources for 1-, 2-, and 3-dimensional EoS tables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='1 COMPOSE CompOSE (CompStar Online Supernovae Equations of State)1 is the largest repository of this kind, offering almost 300 equations of state, divided in families (cold neutron star EoS, cold matter EoS, neutron matter EoS, general purpose EoS, and neutron star crust EoS) and their subgroups (models with hyperons and delta resonances, hybrid quark-hadron models, models with hyperons, models with kaon condensate, nucle- onic models, and quark models).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The repository also provides a software to interpolate data, calculate additional quantities, and graph EoS dependencies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Data tables, associated software and the manual, can be freely downloaded, cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Dexheimer et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' (2022);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Typel et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Paradigmatic examples of the usefulness of such a database can be found in studies that carry out comparisons of the predictions made by different models for the same phys- ical system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' For instance, a set of microscopic, covariant density-functional, and non-relativistic Skyrme-type equations of state, obtained from CompOSE, has been employed to study the structure of purely nucleonic 훽-equilibrated neu- tron stars at finite temperature (Wei, Burgio, Raduta, & Schulze, 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Considering the agreement with presently available astrophysical observational constraints, this study showed that the magnitude of thermal effects depends on the nucleon effective mass as well as on the stiffness of the cold equation of state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Regarding the equations of state themselves, an appropriate quantity to analyze in this con- text is the relative thermal pressure, defined as 푝푟푎푡푖표 = 푝푡ℎ∕푝0 = [푝(휌퐵, 푥푇 , 푇 ) − 푝(휌퐵, 푥0, 0)] ∕푝(휌퐵, 푥0, 0), where 휌퐵 is the baryonic density, 푇 is temperature and 푥0 and 푥푇 are the respective proton fractions of cold and hot matter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The ratio of thermal pressure as a function of density is shown in the upper panel of Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2 for the different equations of state studied.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Moreover, in order to appreciate the astrophysi- cal implications of these equations of state, the relative change of the maximum gravitational neutron-star mass, defined as 푀푟푎푡푖표 = (푀ℎ표푡 푚푎푥 − 푀푐표푙푑 푚푎푥 ) ∕푀푐표푙푑, can be plotted as a function of the thermal pressure ratio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The result is shown in the lower panel of Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 MUSES MUSES (Modular Unified Solver of the Equation of State)2 is a large collaboration project that is developing a new cyber- infrastructure to provide novel tools to answer critical inter- disciplinary questions in nuclear physics, gravitational wave astrophysics and heavy-ion physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The MUSES collabora- tion consists of many researchers and technical professionals 1https://compose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='obspm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='fr 2https://muses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='illinois.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='edu/ NeutronStar Surface .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='Hydrogen/Heliumplasma Ironnuclei OuterCrust .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='lons .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='Electrongas InnerCrust Heavyions Relativisticelectrongas Superfluidneutrons OuterCore .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='Neutrons,protons .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='Electrons,muons InnerCore Neutrons Superconductingprotons Electrons,muons Hyperons(Z,A,三) Deltas (△) Boson(元,K)condensates Deconfined(u,d,s)quarks/color- superconductingquarkmatterV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Dexheimer, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Jacobsen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Farias ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='����������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='���������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='��������������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�������������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�������������� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='������ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='FIGURE 2 Upper panel: Ratio of thermal pressure as a func- ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='tion of density for several equations of state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Lower panel: Relative change of the maximum gravitational mass as a func- tion of the pressure ratio at the center of the star.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Figures modified from Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='6 and Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='8 in (Wei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' across dozens of institutions spread across the globe who are building and using a collaborative platform which is modular because, while at low baryonic chemical potential the EoS is known from first principles, at high there will be different mod- els for the user to choose;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' besides, it is unified in as much as different modules will be merged together to ensure maximal coverage of the phase diagram.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Building up MUSES, physi- cists and computer scientists will work together to develop the software that generates equations of state over large ranges of temperature and chemical potentials to cover the whole QCD phase diagram.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The group of users is composed by interested scientists from different communities, who provide input to the future open-source cyberinfrastructure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 3 ASTROPHYSICAL CONSTRAINTS Any consistent EoS has to pass the test posed by the astrophys- ical constraints related to neutron stars, the most fundamen- tal being the mass-radius relation for these compact objects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Nonetheless, many relevant features cannot be appreciated on such a basis;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' for example, the possible existence of different exotic matter associated with different phase transitions inside a neutron star can easily be seen in the speed of sound (푐푆) behavior but not necessarily in the mass-radius relation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' As a matter of fact, 2푀⊙ stars demand a stiff EoS (with 푐푆 2 ←→ 1 in natural units) at intermediate densities;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' on the other hand, 푐푆 2 ←→ 1∕3 from below at asymptotically large densities because of the conformal limit of massless free quarks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Thus, a non-monotonic behavior is expected for 푐푆, implying the occurrence of bumps related to the softening of the EoS due to new degrees of freedom, cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='Bedaque & Steiner (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Figure 3 , adapted from Tan, Dore, Dexheimer, Noronha- Hostler, & Yunes (2022), show how bumps (that also appear in realistic microscopic models) can be produced under a con- trolled 푐푠 parametrization, allowing a correlation between the density at which the bump appears and curves in the neutron star mass-radius diagram.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Thus, this more systematic para- metric form for the speed of sound can help to determine neutron-star composition;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' besides, maximum stellar mass and radius can determine width, density, and height of the bumps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The non-smooth structure of the speed of 푐푆 related to phase transitions in dense matter makes feasible the constitution of ultra-heavy neutron stars (with masses larger than 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5푀⊙).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' These stars pass all observational and theoretical constraints, including those imposed by recent LIGO/Virgo gravitational- wave observations and NICER X-ray observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Another observational test that may be used to constrain equations of state is the evaluation of tidal deformabilities in neutron stars inferred from gravitational-wave measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In a coalescing binary of neutron stars, the gravitational field of one star perturbs the field of the other (and vice-versa), caus- ing an acceleration in their inspiral.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' This change in the inspiral rate shapes the gravitational-wave emitted, and this wave thus provides information about the tidal deformabilities Λ1,2 of the neutron stars.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Considering a sequence of central densi- ties for a given EoS and a fixed mass ratio, one can construct the binary Love relations (BLRs) Λ푠 and Λ푎, definined with the symmetric and anti-symmetric tidal deformabilities Λ푠,푎 = (Λ1 ±Λ2)∕2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Due to phase transitions and the consequent non- smooth structure of the speed of sound 푐푆, which may tilt the mass-radius diagram, peculiar structures (such as slopes, hills, drops and swooshes) are created in the BLRs (Tan, Dexheimer, Noronha-Hostler, & Yunes, 2022), as shown in Figure 4 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The change in slope in the BLRs may be observable already during the fifth LIGO observing run if a sufficiently loud and 4 V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Dexheimer, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Jacobsen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Farias ε=3p causal limit □ ×× ▲ □●××▲ max central nB/nsat (a) eos1 eos2 eos3 eos4 0 2 4 6 8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 nB/nsat cs 2 J0740+6620 GW190814 GW170817 J0030+0451 (c) 10 11 12 13 14 15 16 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 R [km] M [M⊙] ε=3p causal limit □ ● ×× ▲ □●××▲ max central nB/nsat (a) eos1 eos2 eos3 eos4 0 2 4 6 8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 nB/nsat cs 2 J0740+6620 GW190814 GW170817 J0030+0451 (c) 10 11 12 13 14 15 16 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 R [km] M [M⊙] FIGURE 3 Upper panels: Speed of sound (left) and mass- radius diagram (right) for a subfamily of equations of state with peaks in the speed of sound of different widths at the same location.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Lower panels: Speed of sound (left) and mass-radius diagram (right) for a subfamily of equations of state with peaks in the speed of sound of the same width at different locations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Modified from Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='4 in Tan, Dore, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' low mass neutron-star binary is detected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' The detection of drops and swooshes is more challenging, because both occur at very small Λ푎, and such detection would require very low uncertainties in the measurements, which are achievable only if an exceptionally loud signal is detected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 4 CONCLUSIONS From the recent developments here reported, one my infer that new tight constraints from experiment, observation and the- ory are slowly determining dense matter and neutron-star core properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' In this context, EoS repositories (such as Com- pOSE and MUSES) help speeding up the understanding of dense matter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Furthermore, astrophysical constraints must be taken into account and, in this context, gravitational waves are providing new ways to study the dense matter EoS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Besides the basic mass-radius relation of neutron stars, more specific and subtle quantities (such as the speed of sound and tidal deformabilities for these objects) can be used to probe differ- ent equations of state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Advances in this field can be expected shortly, since LIGO, Virgo, and KAGRA are coordinating a new observing run in March 2023.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Thus, open questions in nuclear astrophysics may soon find their answers and induce further interrogations about the intimate structure of matter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' EoS5 EoS6 EoS7 EoS8 0 1 2 3 4 5 6 7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='0 n/nsat cs 2 (a) J0740+6620 J0030+0451 Am IL/MD Am IL/MD 10 11 12 13 14 15 16 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='6 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='4 R [km] M [M⊙] (b) type (B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='1) type (B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2) 60 80 100 120 140 100 50 0 50 100 Λs Λa (c) type (A) type (C) 0 50 100 150 200 0 2 4 6 8 10 12 Λs Λa (d) FIGURE 4 Upper panels: Speed of sound (left) and mass- radius diagram (right) for for different equations of state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' First-order phase transitions (푐푆 = 0) introduce a second sta- ble branch in the mass-radius curves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Lower panels: BLRs between stars (with a mass ratio 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='75) in the same branch (types A and C) or in different branches (types B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='1 and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='2) produce a slope, hill, drop, and swoosh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Figure modified from Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='1 in Tan, Dexheimer, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' ACKNOWLEDGEMENTS V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' acknowledges support from the National Science Foun- dation under grants PHY1748621, MUSES OAC-2103680, and NP3M PHY-2116686.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' acknowledges support from Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), Grant No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 309598/2020-6 and Fundação de Amparo à Pesquisa do Estado do Rio Grande do Sul (FAPERGS), Grants Nos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 19/2551- 0000690-0 and 19/2551- 0001948-3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' REFERENCES Bedaque, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', & Steiner, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2015, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', 114(3), 031103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Dexheimer, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Mancini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Oertel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Providência, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Tolos, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', & Typel, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2022, Particles, 5(3), 346.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Tan, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Dexheimer, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Noronha-Hostler, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', & Yunes, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2022, Physical Review Letters, 128(16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Tan, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Dore, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Dexheimer, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Noronha-Hostler, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', & Yunes, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2022, Physical Review D, 105(2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Typel, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Oertel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Klähn, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2022, Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' A, 58(11), 221.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Weber, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Contrera, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Orsaria, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Spinella, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', & Zubairi, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2014, Modern Physics Letters A, 29, 1430022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' Wei, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='-B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Burgio, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', Raduta, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=', & Schulze, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} +page_content=' 2021, Physical Review C, 104(6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/8tFQT4oBgHgl3EQfITXl/content/2301.13252v1.pdf'} diff --git a/99E0T4oBgHgl3EQfgAA0/content/tmp_files/2301.02411v1.pdf.txt b/99E0T4oBgHgl3EQfgAA0/content/tmp_files/2301.02411v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..b3f19d471b422fef03928a9f4042adcdb27b34e1 --- /dev/null +++ b/99E0T4oBgHgl3EQfgAA0/content/tmp_files/2301.02411v1.pdf.txt @@ -0,0 +1,3029 @@ +arXiv:2301.02411v1 [math.AG] 6 Jan 2023 +QUELQUES ASPECTS ARITHMÉTIQUES ET GÉOMÉTRIQUES +DES CYCLES ALGÉBRIQUES ET DES MOTIFS +GIUSEPPE ANCONA +Habilitation à diriger les recherches +Soutenue le 17 Novembre 2022 devant le jury composé de : +Yves André, examinateur +Joseph Ayoub, rapporteur +Jean-Benoît Bost, examinateur +Anna Cadoret, rapporteur +François Charles, rapporteur +Carlo Gasbarri, examinateur +Rutger Noot, garant +Table des matières +Remerciements +3 +0. +Introduction +7 +1. +A quoi servent les motifs ? +10 +2. +Cohomologies de Weil +12 +3. +Conjectures standard et moins standard +18 +4. +Exemples +23 +5. +Autodualité et conservativité +29 +6. +Positivité en caracteristique positive +32 +7. +Périodes p-adiques à la André +39 +8. +Motifs des schémas en groupes commutatifs +46 +9. +Construction de classes algébriques +52 +Références +57 +1 + + +MÉMOIRE HDR +3 +Remerciements +On m’avait souvent décrit l’Habilitation comme une formalité par laquelle il +fallait passer. Je me rends compte qu’en fait d’une part j’ai pris plaisir à rédiger +ce mémoire et d’autre part c’est avec un peu d’émotion que je m’approche à cette +soutenance. Je serai entouré par des collègues et des amis qui me sont chers et +j’aimerais par ces quelques lignes les remercier. +Merci aux trois rapporteurs d’avoir accepté de consacrer leur temps à la lecture +de mon mémoire. Joseph Ayoub a été mon encadrant de postdoc et c’est une figure +centrale de la théorie des motifs. J’ai bénéficié de nombreuses discussions avec lui +et c’est un grand honneur pour moi l’attention qu’il a porté à mes travaux. +J’ai rencontré Anna Cadoret pendant ma thèse à l’occasion d’un groupe de travail +organisé par son ANR. J’ai le souvenir d’une atmosphère détendue et ouverte vers +les jeunes, comme il arrive rarement dans notre métier (atmosphère que l’on a +essayé de reproduire avec la RéGA, puis le JAVA). J’ai toujours aimé l’originalité +des questions qu’elle sait se poser et je me réjouis de ses appréciations à ce mémoire. +François Charles était 2 ans avant moi à l’ENS, il a représenté pour plusieurs +d’entre nous une référence. J’ai toujours été impressionné par ses fortes intuitions et +je le remercie pour les échanges que l’on a eu et pour son intérêt pour mes résultats. +Rutger Noot a été rapporteur de ma thèse et m’a beaucoup encouragé après : +je lui en suis reconnaissant. Depuis que je suis à Strasbourg, j’ai pu profiter de +sa culture mathématique à plusieurs reprises. C’était pour moi le choix naturel de +garant et je le remercie d’avoir accepté. +Il n’y a probablement aucune section de ce mémoire où le nom d’Yves André +n’apparaisse pas. Bien que j’essaie constamment d’élargir mes intérêts et m’ouvrir +à de nouveaux domaines, je finis toujours par y découvrir un théorème fondamental +qui lui est dû. Sa présence dans mon jury est un honneur pour moi. +J’ai suivi un cours de Jean-Benoît Bost en M2 : sa clarté et sa vision sont devenues +une référence pour moi. Je le remercie pour son soutien depuis cette époque ; c’est +un plaisir pour moi de l’avoir dans mon jury d’Habilitation. +Les déjeuners à la cantine sont souvent enrichis par les digressions mathématiques +(toujours passionnées) de Carlo Gasbarri. Je le remercie pour celles que l’on a eu, +celles que l’on aura et d’avoir accepté de faire partie du jury. +Je remercie Emiliano Ambrosi, Pierre Baumann, Mattia Cavicchi, Frédéric Cha- +poton, Dragos Fratila, Florence Lecomte et Rutger Noot pour leurs relectures aux +premières versions de ce texte. +Certains des résultats présentés ici sont issus de collaborations avec des ma- +thématiciens qui m’ont beaucoup appris. Je les remercie tous ici, à partir de ma +première collaboratrice, Annette Huber, avec laquelle nous continuons à beaucoup +échanger, jusqu’au plus jeune, Mattia Cavicchi, mathématicien doué à qui je sou- +haite tout le meilleur pour ses débuts de carrière. +Mes recherches mathématiques n’auraient pas pu avoir lieu sans les échanges, les +conseils et les remarques de mes chers amis mathématiciens Olivier Benoist, Yohan +Brunebarbe, Javier Fresán et Marco Maculan. + +4 +GIUSEPPE ANCONA +Quand je repense à mon passé mathématique je dois remercier mon directeur de +thèse Jörg Wildeshaus, ainsi que Frédéric Déglise qui était chercheur à Paris 13 à +l’époque de ma thèse et qui m’a beaucoup appris. +Pendant ces années à l’IRMA j’ai eu le plaisir de discuter de mathématiques +avec de nombreux collègues, tant dans mon équipe que dans les autres. Je pense +en particulier à Emilano Ambrosi, Dragos Fratila, Robert Laterveer, Yohann Le +Floch, Adriano Marmora, Pierre Py et Ana Rechtman. +L’ambiance au laboratoire est bonne, j’ai échangé avec à peu près tout le monde +à l’occasion d’enseignements, conseils ou autre et j’ai souvent trouvé de l’empathie +et de la bienveillance. Chers collègues qui lisez ces lignes, je vous en remercie ! +Le travail d’enseignant-chercheur a besoin d’un soutien administratif et technique +considérable. Nous sommes gâté de ce point de vue-là à Strasbourg : je remercie en +particulier Alexandra Carminati, Sandrine Cerdan, Pascale Igot, Delphine Karles- +kind, Jessica Maurer-Spoerk, Alexis Palaticky et Alain Sartout. +Je suis ému par la présence de plusieurs amis et collègues qui n’habitent pas +à Strasbourg et qui ont fait de la route pour venir voir ma soutenance : Javier, +Mattia, Olivier, Quentin et Yohan, merci ! Je suis également touché par la présence +d’amis qui ont décidé de passer une heure de leur vie à me voir délirer devant des +tableaux bleus : Joanne, Cédric, François... cela me touche beaucoup. +Enfin, merci à tout ce qui est non-mathématique dans ma vie et qui la rend si +heureuse. Je pense aux amis du tango de la belle association Hermosa : merci pour +la bonne ambiance des milongas. Je pense aux amis du foot du terrain 5 : merci +pour les matchs intenses et pour les après-matchs encore plus intenses. Je pense +aux jogging avec Gianluca (accompagnés toujours de questionnements et conseils +réciproques). Je pense aux jeunes parents avec lesquels on partage les apéros où +l’on ne peut jamais terminer nos phrases ainsi que les fêtes regrettées le lendemain. +Et évidemment je pense à ma famille pétillante. Merci à toutes les générations de la +Nonna Michelina qui me chantait l’opéra pendant qu’elle préparait ses pâtes jusqu’à +mes filles qui sautent toujours à mon cou. Merci à Anna qui a toujours envie de me +voir donner un exposé et toujours la joie de le célébrer. Merci à ma mère qui nous +a préparé un super pot (vous verrez...), à mon père qui me parlait de sciences en +voiture pendant qu’il m’accompagnait au judo, à mes frères et à notre chambre à +trois où on pratiquait tout sport possible, à la famille de Maglie qui nous a toujours +accompagné avec amour et à ma belle famille avec qui on partage voyages, bonnes +bouffes et discussions sur le monde. + +A Anna, compagna di viaggio. +A Camilla e Gaia, il nostro viaggio. + + +MÉMOIRE HDR +7 +0. Introduction +Le protagoniste de ce mémoire est un morphisme d’anneaux gradués 1 appelé +application classe de cycle +clX : CH(X) −→ H(X). +Ici X est une variété projective et lisse définie sur un corps k, H est une cohomologie +de Weil 2 et CH(X) est l’anneau de Chow à coefficients rationnels. Il faut penser +à CH(X) comme un invariant de X de nature algébrique et à H(X) comme un +invariant de nature topologique : l’application clX compare ces différentes natures. +Les questions autour de clX peuvent se diviser grossièrement en trois classes : +(1) Décrire l’image de clX, +(2) Décrire le noyau de clX, +(3) Décrire la structure multiplicative de clX. +Par exemple, des conjectures qui entrent dans la première classe sont Hodge, +Tate, la conjecture des périodes de Grothendieck, ou encore les conjectures stan- +dard de type Künneth et de type Lefschetz. Dans la deuxième classe on trouve la +conjecture de Bloch–Beilinson ou la conjecture de nilpotence. Certaines conjectures +sont à cheval entre la première et la deuxième classe, par exemple la conjecture de +conservativité ou la conjecture sur la dimension finie de Kimura. Rentrent dans la +troisième classe les conjectures standard de type Hodge et de type « hom = num ». +Pour décomposer l’étude dans ces trois classes on peut d’abord factoriser clX +(comme morphisme d’anneaux) et obtenir le diagramme suivant : +CH(X) +p +�� +clX +�❖ +❖ +❖ +❖ +❖ +❖ +❖ +❖ +❖ +❖ +❖ +CH(X)/ ker clX � � +i +� H(X). +L’anneau quotient CH(X)/ ker clX est également noté CH(X)/ hom et appelé an- +neaux des cycles modulo l’équivalence homologique. +Les questions appartenant à la classe (1) ci-dessus reviennent alors à l’étude de +l’injection i et celles de la classe (2) à la projection p. Pour exprimer la classe (3) +on considère l’équivalence numérique 3 et on complète le diagramme ainsi : +1. La graduation dans CH(X) est induite par la codimension des sous-variétés et la multipli- +cation est appelée produit d’intersection. +2. Par exemple si k = C on pourra choisir la cohomologie singulière ou pour k de caractéristique +p ≥ 0 et ℓ un nombre premier tel que ℓ ̸= p on pourra choisir la cohomologie ℓ-adique. +3. Cette équivalence rend tous les points de la variété équivalents et les cycles de codimension +complémentaire sont mis en dualité par le produit d’intersection. + +8 +GIUSEPPE ANCONA +CH(X) +p +�� +clX +�▼ +▼ +▼ +▼ +▼ +▼ +▼ +▼ +▼ +▼ +CH(X)/ hom� � i +� +q +�� +H(X). +CH(X)/ num +(0.1) +La compréhension de l’algèbre CH(X)/ num et de la projection q encode une bonne +partie des questions de la classe (3) ci-dessus. +Les conjectures principales sur les cycles algébriques s’expriment mieux si on +passe aux motifs. Cela revient, grosso-modo, à considérer le diagramme (0.1) pour +toutes les variétés X à la fois : +CHM(k) +π +�� +R +�▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +Mot(k)� � +I +� +π′ +�� +GrVect. +NUM(k) +(0.2) +Ici R indique la réalisation des motifs de Chow vers les espaces vectoriels gra- +dués : c’est une collection d’applications classe de cycle. On a de plus des foncteurs +pleins π, π′ de projection vers les motifs homologiques et les motifs numériques et +un foncteur fidèle I des motifs homologiques vers les espaces vectoriels gradués, +également appelé réalisation. +Le théorie des motifs est utile non seulement à la formulation de questions sur +les cycles algébriques mais aussi à la démonstration de résultats sur ces derniers, +voir la Section 1 pour une discussion de ce point. En guise d’exemple, voici une +petite liste de résultats - certains issus de mes travaux - où l’on remarquera que +les motifs n’apparaissent pas dans les énoncés et pourtant sont cruciaux dans leurs +preuves. +Théorème 0.1. +(1) (Kahn [Kah03]) L’application classe de cycle est injective +pour les produits de courbes elliptiques sur un corps fini. +(2) (Kimura [Kim05]) Soit S un surface complexe projective et lisse dominée +par un produit de courbes. Si l’application classe de cycle est surjective alors +elle est injective. +(3) [Anc21] Soit A une variété abélienne de dimension quatre, alors le produit +d’intersection +CH2(A)/num × CH2(A)/num −→ Q + +MÉMOIRE HDR +9 +est de signature (ρ2 − ρ1 + 1; ρ1 − 1), où ρn = dimQ(CHn(A)/num). +(4) [AHPL16] Soient S un schéma de base régulier et G un S-schéma en groupes +commutatifs. Alors l’action du morphisme +nG : G −→ G +de multiplication par n décompose l’espace CH(G) en une somme finie de +sous-espaces propres (de plus les valeurs propres sont des puissances expli- +cites de n et la décomposition ne dépend pas de l’entier n ≥ 2 choisi). +Par sa nature même, la théorie des motifs se mélange aux différentes cohomo- +logies de Weil que l’on peut considérer. Ainsi, en fonction du corps de base k, on +se retrouve à utiliser la théorie de Hodge (k = C), les représentations galoisiennes +(k corps de nombres), l’arithmétique des nombres de Weil (k fini) ou encore de +la théorie de Hodge p-adique (en caractéristique mixte). Cela dégage les aspects +arithmétiques et géométriques de la théorie. +Ces différentes théories cohomologiques ont des analogies, que l’on retrouve dans +les motifs par des théorèmes ou des conjectures, ainsi que des différences, que la +théorie des motifs vise à réparer, voir la Section 2. +On peut distinguer les différents résultats dans le domaine des motifs d’une +part par la partie du diagramme 0.2 que l’on étudie et d’autre part par la nature +géométrique ou arithmétique du corps de base k qui est concerné. Pour aider la +lecture du texte qui suivra, voici une répartition des travaux présentés. +Arithmétique +Géométrie +CHM(k) +§5 [Anc22] +§8 [AEWH15, AHPL16] +Mot(k) +§7 [AF22] +§9 [ACLS22] +NUM(k) +§6 [Anc21] +Organisation du texte. Les premières sections du texte fournissent une intro- +duction partielle à la théorie. Nous avons déjà mentionné que les motifs sont utiles +à l’étude des cycles algébrique et donné le Théorème 0.1 comme exemple, mais +nous n’avons pas dit pourquoi ils sont utiles : c’est le but de la Section 1. La Sec- +tion 2 présente la théorie de Hodge et ses pendants arithmétiques ; on insiste sur +les analogies mais surtout sur les différences. La Section 3 donne des conjectures +sur les cycles algébriques et la Section 4 des exemples de motifs. Ces questions +et ces exemples sont repris dans les sections successives où l’on présente différents +résultats organisés selon le diagramme ci-dessus. Dans la Section 5 on démontre la +conjecture de conservativité pour les motifs provenant de variétés abéliennes défi- +nies sur un corps fini. La Section §6 concerne la conjecture standard de type Hodge +et montre le Théorème 0.1(3). Dans la Section §7 on introduit une nouvelle classe de +périodes p-adiques qui surgit dans l’étude des classes algébriques en caractéristique +mixte. Dans la Section §8 on montre le Théorème 0.1(4) qui nécessite l’utilisation +de techniques motiviques modernes. La Section §9 étudie les classes algébriques de +certaines variétés hyper-kähler qui admettent une fibration lagrangienne. + +10 +GIUSEPPE ANCONA +1. A quoi servent les motifs ? +Dans un article du même titre [Del94a], Deligne expliquait que les motifs n’ont +qu’une utilité essentiellement philosophique permettant de transférer des idées d’une +cohomologie à l’autre, grosso-modo en appliquant le diagramme 0.2 à différentes +cohomologies. Aujourd’hui on comprend que les motifs sont aussi un vrai outil tech- +nique, comme l’avait envisagé Grothendieck. Deligne même revoit sa position dans +son Bourbaki sur les multizétas et explique que les travaux de Brown sont « un des +cas où la philosophie des motifs est non seulement un guide précieux, mais permet +des démonstrations » [Del13]. +J’aimerais expliquer ici l’utilité des motifs notamment dans la théorie des cycles +algébriques : une liste d’énoncés où leur utilisation est cruciale a déjà été donnée +dans le Théorème 0.1. Plusieurs avertissements sont tout de même nécessaires. Pre- +mièrement, je ne prétends pas que cet outil soit l’unique possible, beaucoup de +résultats intéressants sur les anneaux de Chow ont été obtenus par d’autres mé- +thodes. Deuxièmement, à l’heure actuelle les applications les plus impressionnantes +des motifs apparaissent plutôt dans la théorie des périodes [Bro12, Ayo15] - on +peut espérer que les applications majeures de la théorie aux anneaux de Chow sont +encore à venir. +Dans sa construction de la théorie des motifs purs (i.e. pour les variétés propres +et lisses) Grothendieck avait imaginé un pont entre les cycles algébriques et, par +exemple, la cohomologie ℓ-adique. Son idée était que la compréhension des premiers +aurait impliqué ainsi des résultats sur la deuxième, par exemple les conjectures +standard auraient impliqué les conjectures de Weil. En un sens cela semblait la +direction raisonnable : les cycles étaient là depuis plus longtemps (on pourrait dire +depuis le théorème de Bézout) et leur définition pouvait les faire paraître comme +plus accessibles. +On comprend aujourd’hui qu’ils sont plus mystérieux que ce que l’on aurait pu +imaginer. Ceci est devenu flagrant probablement avec le théorème de Mumford +[Mum68] qui montre que le groupe des 0-cycles ne peut pas, en général, être para- +métré par une variété de type fini. En revanche beaucoup de progrès ont été faits +sur la cohomologie. Ce pont maintient donc toute son utilité mais il faut plutôt le +parcourir dans l’autre direction : on essaiera d’exploiter des informations cohomo- +logiques et de les transposer sur les cycles via les motifs. +La question naturelle qui se pose est alors pourquoi l’application classe de cycle +ne serait pas elle-même suffisante pour un tel pont entre anneaux de Chow et +cohomologie? Une réponse courte est que la cohomologie a des propriétés agréables +(Künneth, Poincaré,. . .) que les anneaux de Chow n’ont pas. Les motifs sont une +façon de réorganiser les applications classe de cycle de sorte à ce que l’on ait encore +ces propriétés agréables. +1.1. Motifs purs. Pour expliquer l’idée derrière la construction, prenons la situa- +tion suivante (inspirée par le formalisme tannakien). Soit φ : H → G un morphisme +de groupes et étudions le foncteur induit +f = φ∗ : RepF (G) −→ RepF (H) +(1.1) + +MÉMOIRE HDR +11 +sur les représentations F-linéaires pour F un corps fixé. Supposons avoir à notre +disposition pour cette étude uniquement la collection d’applications +cV : V G −→ V H ⊂ f(V ) +(1.2) +pour chaque V ∈ Rep(G). La donnée de cette collection est certainement moins +agréable que la donnée de f. On peut tout de même retrouver f en remarquant que +son action sur les morphismes est donnée par +cW⊗V ∨ : HomG(V, W) = (W ⊗ V ∨)G −→ (W ⊗ V ∨)H = HomH(V, W). +(1.3) +Cette idée de passer de l’invariant à l’équivariant est l’étape essentielle dans la +construction de Grothendieck des motifs. Dans ce cas pour chaque variété X, V sera +son motif h(X) (objet abstrait de la catégorie en construction), V G sera l’anneau +de Chow CH(X), f(V ) sera la cohomologie singulière (ou ℓ-adique,. . .) H(X), V H +seront les classes de Hodge Hdg(X) (ou les classes Galois invariantes, . . .) et cV +sera l’application classe de cycle clX. En résumant : +V +⇝ +h(X), +V G +⇝ +CH(X), +f(V ) +⇝ +H(X), +V H +⇝ +Hdg(X), +cV +⇝ +clX . +(1.4) +Pour que la construction dans (1.3) soit applicable il faut donner un sens au dual +d’un motif et au produit tensoriel de deux motifs. C’est ici qu’il est nécessaire de +considérer des variétés propres et lisses. En effet la formule de Künneth et la dualité +de Poincaré suggèrent h(X) ⊗ h(Y ) = h(X × Y ) et h(X)∨ = h(X) (à un twist de +Tate près). L’espace Hom(h(X), h(Y )) sera alors contrôlé 4 par CH(X × Y ). +Une fois que la construction de cette catégorie est faite on pourra imaginer - +et essayer de démontrer - des analogies entre H(X) et h(X) qui ne seraient pas +raisonnables avec CH(X) pour ensuite déduire des informations sur les anneaux +de Chow en passant aux Hom dans la catégorie. Par exemple on pourra imaginer +que H(X) et h(X) ont « la même dimension », ce qui ne peut pas avoir lieu avec +CH(X) - voir l’analogie (1.1). Cette idée est à la base de la notion de dimension +dans les motifs due à Kimura et O’Sullivan qui est l’ingrédient essentiel dans la +preuve du Théorème 0.1(1) et joue également un rôle dans les parties (2) et (4) du +même théorème. +1.2. Motifs mixtes. La théorie a beaucoup évoluée depuis ses fondations. Comme +la théorie de Hodge, qui a évolué d’abord avec les structures de Hodge mixtes as- +sociées à des variétés qui ne sont pas forcément projectives ou lisses, pour arriver +jusqu’aux modules de Hodge mixtes qui visent à étudier des familles de variétés +sur des bases générales, également la théorie des motifs a eu une accélération si- +gnificative avec la catégorie triangulée des motifs mixtes de Voevodsky jusqu’aux +motifs relatifs sur une base générale [Ayo07, CD19]. Ces catégories sont liées par le +formalisme des six foncteurs, tout comme les modules de Hodge mixtes. De plus, +certains Hom dans ces catégories permettent de retrouver les anneaux de Chow. +4. Pour les variétés générales qui ne sont pas projectives et lisses l’argument ci-dessus suggère +qu’il n’est pas raisonnable d’espérer un lien entre Hom(h(X), h(Y )) et des anneaux de Chow. Ef- +fectivement dans les motifs de Voevodsky ces Hom n’ont pas d’interprétation en terme d’invariants +classiques, voir aussi la Remarque 8.7(4). + +12 +GIUSEPPE ANCONA +On dispose également de foncteurs de réalisation, par exemple vers les faisceaux +constructibles, qui permettent de retrouver les applications classes de cycles. +Tout comme dans le cas des variétés projectives et lisses expliqué plus haut, ces +catégories ont l’avantage d’avoir des analogies avec les catégories cohomologiques. +Par exemple on dispose d’une filtration de poids sur les motifs tout comme en théo- +rie de Hodge mixte [Bon10]. (Une telle structure n’a pas de bons analogues dans +les anneaux de Chow : par exemple un ouvert d’un espace affine a un anneau de +Chow trivial alors que la filtration de poids en cohomologie peut être non triviale.) +Un autre exemple est le résultat suivant d’Ayoub [Ayo14, Proposition 3.24] : une +application f entre motifs au-dessus d’une base S est un isomorphisme si et seule- +ment si la restriction de f en tout point de S l’est. Cet énoncé est bien entendu +inspiré de son pendant pour les faisceaux constructibles ou étales. +Ces nouvelles catégories présentent un deuxième avantage : on dispose main- +tenant de beaucoup plus de flexibilité, analogue à celle permise par les modules +de Hodge mixtes. Par exemple, si on veut étudier l’anneau de Chow d’une variété +projective et lisse X, on pourrait avoir envie de stratifier X et d’étudier chaque +strate, ou de fibrer X au-dessus d’une base et d’étudier comment les fibres varient. +Ce genre de construction mène très souvent à des variétés qui ne sont pas lisses et +pour lesquelles les anneaux de Chow et leur fonctorialité ne sont pas définis : ces +catégories de motifs permettent, entre autre, de contourner ce problème. +Ces techniques ont permis la construction de certains cycles « concrets », par +exemple certains cycles prédits par la conjecture de Hodge joint au théorème de +décomposition, qui étaient inaccessible par des méthodes directes. Notamment cela +a été appliqué aux variétés de Shimura [Wil17], aux fibrés en quadriques [CDN22] +et aux variétés hyper-kähler qui admettent une fibration lagrangienne [ACLS22]. +1.3. Complexes motiviques. Une troisième raison pour laquelle les motifs sont +utiles à l’étude des cycles algébriques vient de leur définition moderne (depuis Voe- +vodsky). Dans la théorie de Grothendieck les motifs sont des symboles formels et +leur lien avec les cycles algébriques a lieu par construction. Dans les catégories mo- +dernes les motifs sont des complexes de faisceaux et il est possible d’en construire +un certain nombre explicitement. Leur lien avec les cycles algébriques est loin d’être +une tautologie et c’est en fait un des résultats plus profond de la théorie. On peut +alors espérer que certaines questions délicates sur les cycles algébriques puissent +devenir concrètes dans leur pendant faisceautique. C’est ce qui se passe notam- +ment dans la construction de la réalisation de Betti [Ayo10] ou dans la preuve du +Théorème 0.1(4), voir Section 8. +2. Cohomologies de Weil +Dans cette section on rappelle la définition classique de structure de Hodge +(Définition 2.1) ainsi qu’une formulation équivalente qui se prête à mieux décrire les +propriétés de positivité (Définition 2.2). Le but principal de la section est de rappeler +ces propriétés de positivité ainsi que des propriétés d’autodualité des structures de +Hodge puis de montrer que leurs analogues en cohomologie ℓ-adique sont faux en +général (Remarque 2.16 et Exemple 2.17). La définition de polarisation est cruciale, +on essaie de la justifier dans la Remarque 2.12. + +MÉMOIRE HDR +13 +D’autres différences entre la théorie de Hodge et ses analogues arithmétiques +sont éparpillées un peu partout dans le texte, voir par exemple la Conjecture 3.18 +ou la Remarque 6.9. +Définition 2.1. (Structure de Hodge, définition classique.) Une structure de Hodge +pure de poids n ∈ Z est la donnée d’un Q-espace vectoriel de dimension finie V +muni d’une décomposition +V ⊗Q C = +� +p+q=n +p,q∈Z +V p,q +(2.1) +vérifiant V p,q = V q,p. +Si on considère les espaces V {p,q} = V ⊗QR∩(V p,q +V q,p) on trouve la définition +équivalente suivante. +Définition 2.2. (Structure de Hodge, définition équivalente.) Une structure de +Hodge pure de poids n ∈ Z est un triplet formé d’un Q-espace vectoriel de dimension +finie V , d’une décomposition +V ⊗Q R = +� +p≤q +p+q=n +p,q∈Z +V {p,q} +(2.2) +en sous-espace réels et d’une structure complexe sur les espaces V {p,q} pour p ̸= q. +Les paires (p, q) qui apparaissent dans la décomposition sont appelées les types +de V . +Remarque 2.3. (Lien entre les définitions.) Les deux présentations ci-dessus sont +bien équivalentes. On remarque que, pour p < q, l’ espace V {p,q} ⊗ C est muni de +deux structures complexes, par conséquent on a une décomposition +V {p,q} ⊗ C = V p,q ⊕ V q,p, +où V p,q est l’espace où les structures coïncident et V q,p est l’espace où elles sont +conjuguées 5. +La Définition 2.2 est plus pratique pour exprimer des propriétés de positivité 6, +voir par exemple Définition 2.6. +Exemple 2.4. +(1) (Cohomologie singulière.) Si X est une variété projective et +lisse sur les nombres complexes sa cohomologie singulière est munie d’une +5. On pourrait inverser le rôle de ces deux espaces, dans ce cas il faudrait changer les signes +dans la Définition 2.9 pour avoir les mêmes conventions de signe classiques. +6. En revanche la structure tensorielle s’exprime mieux avec la Définition 2.1, par la règle +(V ⊗ W )p,q = +� +a+c=p +b+d=q +V a,b ⊗ W c,d. +Dans le cas réel on remarque que V {a,b} ⊗ W {c,d} a deux structures complexes quand a < b +et c < d. Ceci induit une décomposition en deux espaces comme précédemment, celui où les +structures coïncident contribue à (V ⊗ W )a+c,b+d et l’autre à (V ⊗ W )m,M où m et M sont le +minimum et le maximum de la paire {a + d, b + c}. + +14 +GIUSEPPE ANCONA +structure de Hodge fonctorielle en X, plus précisément Hn(X) est pure de +poids n. Tous les théorèmes que l’on peut imaginer (Poincaré, Künneth, +Lefschetz,. . .) respectent cette structure supplémentaire. +(2) (Variétés abéliennes.) Dans le cas particulier d’une variété abélienne com- +plexe A dont la variété analytique sous-jacente est Cg/Λ on a H1(A)⊗R = +Λ⊗R = Cg, ce qui fait apparaître explicitement la structure complexe dans +la Définition 2.2 de l’espace H1(A) ⊗ R. +(3) (Twist de Tate.) En poids 2n il existe une seule structure de Hodge de +dimension 1 à isomorphisme non unique près. On fixe une telle structure +de Hodge et on la note Q(−n). Pour une structure de Hodge V on note +V (−n) = V ⊗ Q(−n) et on l’appelle twist de Tate ; on choisit les Q(−n) de +sorte à avoir l’identification Q(a) ⊗ Q(b) = Q(a + b). +Traditionnellement on choisit Q(n) = (2iπ)nQ ⊂ C, puisque c’est la +structure de Hodge qui apparaît naturellement dans la cohomologie sin- +gulière de degré maximale. Ce choix particulier ne joue pas de rôle dans +les constructions qui suivront, sauf pour la notion de polarisation, voir Re- +marque 2.7(1). +Remarque 2.5. (Autodualité.) Une structure de Hodge n’est pas, en général, au- +toduale dans le sens où V et V ∨ ne sont pas en général isomorphes, même à un +twist près (le seul twist pour lequel cette autodualité est raisonnable étant le poids). +Par exemple prenons une structure de Hodge de poids 0 et supposons que la +décomposition (2.2) ait deux facteurs non nuls V ⊗Q R = V {0,0} ⊕ V {−1,+1}. Sup- +posons de plus que V {0,0} soit compatible avec la structure rationnelle et V {−1,+1} +ne le soit pas, c’est-à-dire que l’on ait l’égalité dimQ(V {0,0}∩V ) = dimR V {0,0} mais +dimQ(V {−1,+1} ∩ V ) < dimR V {−1,+1}. +Si V était autodual on aurait en particulier une forme bilinéaire sur V pour +laquelle V {0,0} et V {−1,+1} seraient l’orthogonal l’un de l’autre. Or comme une telle +forme bilinéaire serait définie sur Q, l’espace V {−1,+1} serait forcé à être également +compatible avec la structure rationnelle. +Il se trouve que toutes les structures de Hodge provenant de la géométrie algé- +brique sont autoduales. En fait même plus, elles sont munies d’une forme bilinéaire +« aussi définie que possible », c’est l’objet de la définition de polarisation, rappelée +ci-dessous. +Définition 2.6. (Polarisation - poids pair.) Une polarisation sur une structure de +Hodge V pure de poids 2n est une forme bilinéaire symétrique b sur le Q-espace +vectoriel V telle que : +(1) Les espaces V {p,q} sont orthogonaux deux à deux par rapport à b, +(2) L’adjointe par rapport à b de la structure complexe sur V {p,q} est sa conju- +guée, +(3) La restriction de b aux facteurs V {n−a,n+a} est définie (−1)a-positive 7. +7. C’est-à-dire définie positive si a est pair et définie négative si a est impair. + +MÉMOIRE HDR +15 +Remarque 2.7. +(1) Les deux premières propriétés de la définition sont équi- +valentes à l’existence d’un morphisme de structures de Hodge +b : V ⊗ V −→ Q(−2n). +Inversement, étant donné un tel b il est nécessaire de choisir une identi- +fication Q(−2n) ⊗Q R ∼= R pour pouvoir exprimer la dernière propriété (de +positivité). L’Exemple 2.4(3) fixe un tel choix. +(2) Dans le cas V = H2n(X) on déduit qu’une polarisation est définie positive +sur les classes algébriques, car elle l’est sur V {n,n}. C’est utile de connaître +les signes sur les autres facteurs même pour la compréhension des classes +algébriques, voir l’Exemple 4.9 et la Section 6. +(3) Une polarisation est automatiquement non dégénérée donc les structures de +Hodge qui la possèdent sont autoduales. D’autre part il y a des structures +de Hodge autoduales, même simples, qui n’admettent pas de polarisation, +voir l’Exemple 2.8. Autrement dit la polarisabilité est une notion plus forte +que l’autodualité. +Exemple 2.8. (Autodualité vs polarisation.) Construisons une structure de Hodge +simple et autoduale qui n’admet pas de polarisation. Soit A une variété abélienne +complexe de dimension 4 et très générale. Alors la partie primitive V = H4,prim(A) +est une structure de Hodge simple, de type (0, 4), (1, 3) et (2, 2) et polarisable (par +le Corollaire 2.14). Définissons une nouvelle structure de Hodge W où le Q-espace +vectoriel est le même que V , ainsi que la décomposition, mais on inverse le rôle de +la partie (0, 4) avec la partie (1, 3), c’est-à-dire +W {0,4} = V {1,3}, +W {1,3} = V {0,4} +et +W {2,2} = V {2,2}. +On prétend que W satisfait aux propriétés requises. Premièrement remarquons +qu’une polarisation sur V induit un accouplement b sur W qui rend W autoduale. +D’autre part ce b ne peut pas être une polarisation car les signes de la Définition +2.6 ne sont pas respectés. Il reste à montrer que W n’admet pas de polarisation et +pour cela il suffit de voir que b est l’unique accouplement sur W à scalaire près et +donc que EndHS(W) = Q · Id . Or, par construction, on a EndHS(W) = EndHS(V ). +Pour comprendre les endomorphismes de V , on prend le point de vue tannakien. +Le groupe tannakien associé à H1(A) est GSp8 et la représentation sous-jacente à +H4,prim(A) ⊂ H1(A)⊗4 est géométriquement irréductible, donc EndHS(V ) = Q · Id . +Nous rappelons ci-dessous la définition de polarisation dans le cas de poids im- +pair, elle est un peu moins agréable. Même si on s’intéresse seulement aux classes +algébriques les groupes de cohomologie de degré impair peuvent être utiles (par +exemple la cohomologie d’une variété abélienne A est contrôlée par son H1) mais +il est souvent suffisant de se rappeler uniquement que la définition de polarisation +est stable par produit tensoriel (et donc une polarisation sur H1(A) en induira une +sur Hn(A) via Hn(A) = ΛnH1(A)). +Définition 2.9. (Polarisation - poids impair.) Une polarisation sur une structure +de Hodge V pure de poids 2n+1 est une forme bilinéaire alternée b sur le Q-espace +vectoriel V telle que : +(1) Les espaces V {p,q} soient orthogonaux entre eux par rapport à b, + +16 +GIUSEPPE ANCONA +(2) L’adjointe par rapport à b de la structure complexe sur V {p,q} est sa conju- +guée, +(3) La forme bilinéaire symétrique b(·, i·) est définie (−1)a-positive sur les fac- +teurs V {n−a,n+1+a}. +Remarque 2.10. +(1) La Remarque 2.7 s’applique également dans ce cas. +(2) On pourrait vouloir travailler avec b(i·, ·) à la place de b(·, i·), dans ce cas +les signes s’inverserait. On trouve plus agréable d’avoir de la positivité sur +la « partie centrale » V {n,n+1}, c’est analogue à ce qui se passe dans la +Définition 2.6 avec la partie centrale V {n,n}. +Exemple 2.11. (Polarisations issues de la géométrie.) Soit X une variété algé- +brique complexe de dimension dX. D’une part la dualité de Poincaré fournit une +identification Hn(X) = H2dX−n(X)∨(−dX). D’autre part, le théorème de Lefschetz +difficile donne un isomorphisme Hn(X) ∼= H2dX−n(X)(dX −n), au moyen du choix +d’une section hyperplane L. En combinant les deux on obtient une forme bilinéaire +non-dégénérée +bL : Hn(X) ⊗ Hn(X) −→ Q(−2n). +Celle-ci est bien un morphisme de structure de Hodge, mais elle n’a pas les signes +demandés par une polarisation. Pour les obtenir, il faut modifier les signes de cer- +tains facteurs de la décomposition de Lefschetz. +Par exemple écrivons la décomposition de Lefschetz en degré six +H6 = H6,prim ⊕ H4,prim(−1) ⊕ H2,prim(−2) ⊕ H0,prim(−3), +elle est orthogonale par rapport à l’accouplement bL ci-dessus. La décomposition +de Hodge de chacune de ces quatre structures de Hodge est encore orthogonale; la +signature de bL est la suivante (où les cases vides sont pour les sous-espaces qui +sont toujours réduits à zéro). +signes bL +H6,prim +H4,prim(−1) +H2,prim(−2) +H0,prim(−3) +(3, 3) +- ++ +- ++ +(2, 4) ++ +- ++ +(1, 5) +- ++ +(0, 6) ++ +Pour obtenir une polarisation il faudra donc changer le signe sur les facteurs +H6,prim et H2,prim(−2). +Remarque 2.12. (Pourquoi la définition de polarisation?) Considérons le change- +ment de signe expliqué dans l’exemple ci-dessus. Il serait tentant, à première vue, +de changer encore de signe, cette fois-ci par rapport à la décomposition en V {p,q}, +de sorte à avoir une forme quadratique définie positive dans la Définition 2.6. Le +problème est que cette deuxième décomposition n’est pas en général définie sur Q +et donc ce changement de signes donnerait une forme bilinéaire qui n’est pas définie +sur Q. +De manière générale, pour les structures de Hodge provenant de la géométrie, +on ne peut pas espérer avoir une forme quadratique qui soit à la fois définie sur +Q, compatible avec la décomposition de Hodge et définie positive, voir l’Exemple +2.15. Il faut alors imaginer la Définition 2.6 comme la meilleure approximation d’un + +MÉMOIRE HDR +17 +produit scalaire qui puisse exister pour les structures de Hodge issues des variétés +algébriques. D’ailleurs la proposition ci-dessous montre que la polarisation a toutes +les conséquences que l’on aimerait déduire d’un produit scalaire. +Proposition 2.13. Soient V une structure de Hodge, b une polarisation sur V +et W ⊂ V une sous-structure de Hodge. Alors la restriction de b à W est une +polarisation. De plus l’orthogonal W ⊥ ⊂ V de W par rapport à b est une sous- +structure de Hodge et on a l’égalité de structures de Hodge V = W ⊕ W ⊥. +Corollaire 2.14. Soit X une variété complexe projective et lisse. Alors tout sous- +structure de Hodge de la cohomologie singulière de X est polarisable. +Exemple 2.15. (Polarisations vs produits scalaires.) Construisons une structure +de Hodge V de poids zéro et d’origine géométrique telle que HomHS(V ⊗ V, Q(0)) +ne contient pas une forme quadratique définie positive. +Soit E une courbe elliptique non CM et considérons H1(E). On a une décom- +position de structures de Hodge H1(E) ⊗ H1(E)∨ = Q(0) ⊕ V , où V a dimension +3 et types (0, 0) et (−1, 1). +On prétend que HomHS(V, V ∨) = HomHS(V ⊗V, Q(0)) est un Q-espace vectoriel +de dimension 1. Pour le montrer on prend le point de vue tannakien. On peut voir +que le groupe tannakien associé à H1(E) est GL2 et que H1(E) est la représenta- +tion standard. Le groupe GL2 agit sur V et on a l’identification HomHS(V, V ∨) = +HomGL2(V, V ∨). D’autre part, la théorie classique des représentations nous dit que +V est géométriquement irréductible et donc que HomGL2(V, V ∨) est un Q-espace +vectoriel de dimension (au plus) 1. +D’autre part l’espace HomHS(V ⊗V, Q(0)) contient une polarisation, par le Corol- +laire 2.14, et donc cet espace est formé uniquement de multiples d’une polarisation. +En particulier HomHS(V ⊗ V, Q(0)) ne peut pas contenir une forme quadratique +définie positive. +On peut aussi en déduire que H2(E × E) n’admet pas une forme quadratique +définie positive qui respecte la structure de Hodge car V (−1) en est un facteur +direct. +Le reste de la section insiste sur les différences entre théorie de Hodge et cohomo- +logie ℓ-adique, au regard notamment de la notion de polarisation et des propriétés +d’autodualité. +Remarque 2.16. (Polarisations en cohomologie ℓ-adique?) La proposition ci-dessus +dit en particulier que les structures de Hodge issues de la géométrie algébrique +forment une catégorie semi-simple et que chaque objet est autodual à un twist +près. C’est une grande différence avec la cohomologie ℓ-adique : la semi-simplicité +est seulement conjecturale et l’autodualité est fausse en général, voir l’Exemple +2.17. +La notion même de polarisation n’a pas d’analogue : on ne peut même pas +formuler des propriétés de positivité analogues à celles de la Définition 2.6 pour la +simple raison que la notion de positif n’a pas de sens dans Qℓ. +On peut construire des accouplements sur la cohomologie ℓ-adique de la même +façon qu’en théorie de Hodge, comme dans l’Exemple 2.11. On ne connait pas +de formulation, même conjecturale, qui décrirait cette Qℓ-forme quadratique. Par + +18 +GIUSEPPE ANCONA +ailleurs les invariants d’une telle forme quadratique, comme son symbole de Hilbert, +ne contrôlent pas ceux des sous-formes quadratiques (contrairement à ce qui se passe +avec la signature). Ceci suggère que même si on trouvait une propriété analogue à +la Définition 2.6 pour les groupes de cohomologies Hn +ℓ (X) elle pourrait ne pas être +valable pour les facteurs directs de Hn +ℓ (X). +Exemple 2.17. (Non autodualité en cohomologie ℓ-adique.) Soient k un corps de +type fini et E une courbe elliptique définie sur k telle que End(E)⊗Q soit un corps +quadratique imaginaire. Prenons un nombre premier ℓ différent de la caractéristique +de k et tel que End(E) ⊗ Qℓ = Qℓ ⊕ Qℓ. Alors l’action de End(E) ⊗ Qℓ donne une +décomposition de représentations galoisiennes H1 +ℓ (E) = V ⊕ W. Le cup-produit in- +duit une autodualité sur le H1 +ℓ (E), elle réalise W comme dual de V par la positivité +de Rosati. +D’autre part on prétend que V et W ne sont pas isomorphes comme repré- +sentation de Galois, ce qui impliquera en particulier que V n’est pas autodual. +En effet s’ils étaient isomorphes on aurait EndGal H1 +ℓ (E) = M2×2(Qℓ), or on a +EndGal H1 +ℓ (E) ∼= End(E) ⊗ Qℓ comme prédit par la conjecture de Tate, montrée +dans ce context par Tate, Faltings et Zarhin. +Remarque 2.18. L’exemple ci-dessus dépend du nombre premier ℓ choisi. On s’at- +tend à ce que l’on ne puisse pas trouver une représentation d’origine géométrique, +« indépendante de ℓ » et non autoduale. Cette idée est rendue précise par les motifs, +voir la Conjecture 3.5 et la remarque qui la suit. +3. Conjectures standard et moins standard +Soient k un corps de base et H∗ une cohomologie de Weil. Considérons le dia- +gramme (0.2) qui représente des catégories de motifs : +CHM(k) +π +�� +R +�▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +Mot(k)� � +I +� +π′ +�� +GrVect. +NUM(k) +on les considère toujours à coefficients rationnels. Pour chaque variété projective +et lisse X, de dimension dX, on considère son motif h(X), on utilisera ce même +symbole dans les différentes catégories, on précisera la catégorie concernée si cela +est important. +Le but de cette section est de donner une collection de conjectures qui permettent +d’avoir une intuition sur les motifs. Ces conjectures sont essentiellement classiques, +hormis la Conjecture 3.5 d’autodualité et la Conjecture 3.7 de positivité qui sont +nouvelles à notre connaissance. +La section est organisée en trois sous-sections. La première présente les conjec- +tures principales de la théorie (qui essentiellement entrainent les autres conjectures). +La deuxième sous-section présente les conjectures dite standard de Grothendieck. +La dernière porte sur la notion de dimension finie de Kimura. + +MÉMOIRE HDR +19 +Avec les techniques actuelles ces conjectures sont hors de portée en toute généra- +lité, mais il est possible en démontrer des cas particuliers, comme on le verra dans +les sections successives. +Conjectures principales. +Conjecture 3.1. (Chow–Künneth.) Il existe une décomposition (non unique) +h(X) = +2dX +� +n=0 +hn(X) +dans la catégorie CHM(k) telle que R(hn(X)) = Hn(X). +Remarque 3.2. +(1) (Non unicité.) Le facteur h0(X) existe toujours. Pour le +construire il suffit de considérer une application constante de X vers X : +elle sera bien l’identité sur le H0(X) et nulle sur les autres groupes. Le +facteur ainsi défini dépend de l’image de cette application constante, ou +plus précisément de sa classe modulo équivalence rationnelle. En particulier +une telle décomposition ne peut pas être unique en général. +(2) (Autodualité.) On peut conjecturer l’existence d’une décomposition de Chow– +Künneth qui ait la propriété supplémentaire d’être autoduale, c’est-à-dire +que si on considère la dualité de Poincaré h(X)∨ = h(X)(dX) alors le fac- +teur hn(X)∨ correspond à h2dX−n(X)(dX). D’un point de vue de cycles +algébriques cela veut dire que les projecteurs pn sont donnés par une collec- +tion de cycles dans X×X telle que σ∗pn = p2d−n, où σ : X×X → X×X est +l’inversion des deux facteurs. Ce n’est pas automatique de construire une +décomposition autoduale à partir d’une décomposition de Chow–Künneth : +même si l’on pose p2d−n = σ∗pn on pourrait avoir que les projecteurs pn et +σ∗pn ne sont pas orthogonaux. +Conjecture 3.3. (Conservativité.) Tous les foncteurs du diagramme (0.2) sont +conservatifs, c’est–à-dire un morphisme entre motifs est en fait un isomorphisme +si son image via un de ces foncteurs du diagramme l’est. +Définition 3.4. On dit qu’un motif est de poids n s’il est facteur direct d’un motif +de la forme hn(X). +Conjecture 3.5. (Autodualité.) Si M est un motif de poids n (toujours à coeffi- +cients rationnels) alors il existe un isomorphisme (non unique) +M ∨ ∼= M(n). +Remarque 3.6. +(1) (Autodualité des motifs vs autodualité en cohomologie.) +Les groupes de cohomologies Hn(X) jouissent de ce type d’autodualité via +la dualité de Poincaré et l’isomorphisme de Lefschetz difficile (qui dépend +du choix d’une section hyperplane), mais en général les facteurs directs +de Hn(X) n’ont pas cette propriété d’autodualité, voir l’Exemple 2.17. +Ceci n’est pas en contradiction avec la conjecture ci-dessus : les exemples +construits ne sont pas la réalisation d’un motif à coefficients rationnels. + +20 +GIUSEPPE ANCONA +(2) (Lien avec les conjectures classiques.) Cette conjecture est nouvelle à notre +connaissance. Nous trouvons sa formulation naturelle et elle nous a guidé +dans l’étude de la conjecture de conservativité, voir la Section 5. Elle suit par +ailleurs de conjectures classiques de positivité (notamment la Conjecture +3.15 que l’on verra plus loin), de la même manière que l’autodualité pour +les structures de Hodge suit des propriétés de positivité des polarisations, +voir la Proposition 2.13 et la remarque qui la suit. +Conjecture 3.7. (Positivité.) Supposons que le corps de base k soit de caractéris- +tique p. Soient M un motif homologique sur k et +q : Sym2 M → +1 +un morphisme dans Mot(k). Supposons que q soit la réduction modulo p d’une +application �q : Sym2 � +M → +1 définie en caractéristique zéro. Définissons qZ comme +la restriction de q à toutes les classes algébriques Z(M) = Hom(1, M) de M et qB +comme la réalisation singulière de �q. +Supposons que qB soit une polarisation. Alors qZ est définie positive. +Remarque 3.8. +(1) (Sur la relevabilité.) L’hypothèse de relevabilité à la ca- +ractéristique zéro est vérifiée dans des cas intéressants, par exemple les +variétés abéliennes. Dans ce cas, les classes algébriques qui se relèvent à +la caractéristique zéro vérifient automatiquement la conjecture, voir la Re- +marque 2.7. Soulignons tout de même qu’il y a en général des classes algé- +briques qui ne sont pas relevables, même dans le cas des variétés abéliennes. +(2) (Lien avec les conjecture classique.) Cette conjecture est nouvelle et nous ne +savons pas si elle peut se déduire de conjectures classiques. Nous trouvons +sa formulation naturelle et elle nous a guidé dans l’étude de la conjecture +standard de type Hodge, voir la Section 6. +Conjectures standard. +Conjecture 3.9. (Künneth.) Dans la catégorie Mot(k) des motifs homologiques, +il existe une décomposition +h(X) = +2dX +� +n=0 +hn(X) +telle que la réalisation de hn(X) soit Hn(X). +Remarque 3.10. (Lien avec Chow–Künneth.) Cette conjecture est bien sûr une +conséquence de la Conjecture 3.1. +Remarquons qu’une décomposition de Künneth est automatiquement unique et +autoduale par définition d’équivalence homologique : il s’agit de la graduation de +GrVect et de la dualité de Poincaré en cohomologie. C’est une différence avec la +décomposition de Chow–Künneth, voir la Remarque 3.2. +Conjecture 3.11. (Lefschetz.) Soient dX la dimension de X et L une section +hyperplane de X. Alors pour tout n ≤ dX il existe une correspondance γn dans +X × X dont la réalisation en degré dX + n induit un isomorphisme +R(γn) : HdX+n(X) +∼ +−−→ HdX−n(X)(−n) + +MÉMOIRE HDR +21 +qui est l’inverse du cup produit par Ln. +Remarque 3.12. (Lien avec l’autodualité.) La Conjecture standard de type Les- +chetz 3.11 implique la Conjecture standard de type Künneth 3.9, c’est un argument +classique de Kleiman [Kle68]. +Inversement, si X vérifie Künneth alors la conjecture de type Lefschetz est équi- +valente à l’autodualité hn(X)∨ ∼= hn(X)(n) du motif homologique hn(X). Cette +équivalence se déduit de la Proposition 4.5. Elle montre en particulier que la conjec- +ture de type Lefschetz ne dépend pas de la section hyperplane L choisie et elle suit +de la Conjecture d’autodualité 3.5. +Conjecture 3.13. (hom = num.) Le foncteur π′ est une équivalence. +La quatrième et dernière des conjectures standard est celle de type Hodge. Pour +la formuler il est nécessaire d’introduire la proposition suivante. +Proposition 3.14. (cf. [Anc21, §3]) Supposons que X vérifie la conjecture stan- +dard de type Lefschetz (Conjecture 3.11). Choisissons une section hyperplane L. +Alors le motif homologique hn(X) admet une décomposition en parties primitives +et il est possible de construire un accouplement +qX,n,L : hn(X) ⊗ hn(X) → Q(−n) +de façon analogue à la construction d’une polarisation sur la cohomologie singulière +d’une variété algébrique complexe 8, voir l’Exemple 2.11. +De plus, si on restreint l’accouplement +qX,2n,L(2n) : h2n(X)(n) ⊗ h2n(X)(n) → Q +aux classes algébriques Zn(X)/ hom = HomMot(k)(1, h2n(X)(n)) on obtient une Q- +forme quadratique +qZ,hom : Zn(X)/ hom ⊗ Zn(X)/ hom → Q +dont le noyau est formé exactement par les cycles numériquement triviaux 9. En +particulier qZ,hom induit une forme quadratique sur les cycles modulo équivalence +numérique +qZ : Zn(X)/ num ⊗ Zn(X)/ num → Q +qui est non dégénérée. +Conjecture 3.15. (Conjecture standard de type Hodge.) La forme quadratique +qZ : Zn(X)/ num ⊗ Zn(X)/ num → Q +introduite ci-dessus est définie positive. +Remarque 3.16. (Lien avec la conjecture de positivité.) La conjecture standard +de type Hodge (Conjecture 3.15) ne demande pas la relevabilité de X à la caracté- +ristique zéro et en ce sens elle est plus générale que la Conjecture de positivité 3.7. +D’autre part pour les variétés relevables la Conjecture 3.7 est plus générale que la +8. En particulier, si k = C et H∗ est la cohomologie singulière, R(qX,n,L) est la polarisation +classique induite par L. +9. Autrement dit, la conjecture hom = num pour X est équivalente à dire que qZ,hom est non +dégénérée. + +22 +GIUSEPPE ANCONA +conjecture standard de type Hodge, car elle s’applique à des polarisations abstraites +qui ne seraient pas forcément celles provenant d’une section hyperplane. +Remarque 3.17. (Positivité en caractéristique zéro.) En caractéristique zéro, la +forme quadratique introduite ci-dessus qZ : Zn(X)/ hom ⊗ Zn(X)/ hom → Q est +définie positive : c’est une conséquence des propriétés de positivité d’une polari- +sation (dites relations de Hodge–Riemann), voir Remarque 2.7. En particulier, en +caractéristique zéro, la conjecture standard de type Lefschetz implique les autres +conjectures standard. +Une autre différence entre la caractéristique zéro et la caractéristique positive se +trouve dans la conjecture suivante. +Conjecture 3.18. Considérons l’application classe de cycle ℓ-adique clX : CH(X) → +Hℓ(X). Alors le Q-espace vectoriel Im clX est de dimension finie. Plus précisément +l’application canonique Im clX ⊗QQℓ → Hℓ(X) est injective. +Remarque 3.19. Cette conjecture est une conséquence de la conjecture hom = +num car l’équivalence numérique commute à l’extension des scalaires. En carac- +téristique zéro elle est connue : on utilise les théorèmes de comparaison pour se +reporter à la cohomologie singulière. +Motifs de dimension finie. +Conjecture 3.20. (Dimension finie.) Tout motif de Chow M admet une décom- +position (non unique) +M = M+ ⊕ M− +vérifiant ΛNM+ = 0 et SymN M− = 0 pour un naturel N assez grand. +Remarque 3.21. +(1) (Lien avec les autres conjectures.) Supposons que M = +h(X) admet une décomposition de Chow–Künneth (Conjecture 3.1). Alors +M+ = +� +h2n(X) +et +M− = +� +h2n+1(X) +devraient vérifier la conjecture ci-dessus. En effet la conservativité (Conjec- +ture 3.3) prédit qu’il suffit de vérifier ces relations en cohomologie, or la réa- +lisation de M+ est un espace vectoriel gradué de dimension finie concentré +en degrés pairs, par conséquence toute puissance N-ème extérieure l’annule +dès que N est plus grand que la dimension totale. +Le raisonnement est analogue pour M−. On remarque que la réalisation +est concentrée en degré impair et que, par la règle des signes de Koszul, une +puissance symétrique sur un motif ou une variété devient une puissance +extérieure sur les groupes de cohomologie. +(2) (Applications.) Au delà d’être une conjecture naturelle, la notion de di- +mension finie s’est avéré être utile : elle est stable par plusieurs opérations, +dont le produit tensoriel et le passage à un facteur direct, elle est vérifiée au +moins par les courbes, et elle permet de déduire la conservativité pour tous +les foncteurs tensoriels quotient (notamment π et π′ de (0.2)). Cela a permis +à Kimura de déduire la conjecture de Bloch pour les surfaces dominées par + +MÉMOIRE HDR +23 +un produit de courbes [Kim05], voir le Théorème 0.1(1). En appliquant ces +propriétés de conservativité au Frobenius, Kahn a déduit que l’application +classe de cycle est injective pour les produits de courbes elliptiques sur un +corps fini [Kah03], voir le Théorème 0.1(2). +4. Exemples +Dans cette section on discute des exemples de motifs. Ces exemples sont organisés +dans trois sous-sections. Dans la première on présente des cas classiques où les +conjectures de la section précédente sont vérifiées. La deuxième sous-section montre +des subtilités (assez amusantes!) entre les différentes réalisations. La dernière partie +étudie les motifs de variétés abéliennes CM. Tous les exemples de cette section sont +repris à plusieurs endroits dans le texte. +On continue à travailler avec le diagramme (0.2) : +CHM(k) +π +�� +R +�▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +▲ +Mot(k)� � +I +� +π′ +�� +GrVect. +NUM(k) +et h(X) indiquera le motif d’une variété X dans les différentes catégories, on préci- +sera la catégorie concernée si cela est important. +Quelques évidences des conjectures de la Section 3. +Proposition 4.1. Soient h(X) ∈ Mot(k) un motif homologique, f un endomor- +phisme de h(X) et pn(f) le polynôme caractéristique de f agissant sur Hn(X). +Supposons que pn(f) et pm(f) soient premiers entre eux pour tous les n ̸= m. +Alors h(X) admet la décomposition de Künneth +h(X) = +� +n=0 +hn(X), +voir la Conjecture 3.9. De plus les projecteurs de cette décomposition appartiennent +à l’algèbre Q[f]. +Remarque 4.2. +(1) La preuve est élémentaire : on applique l’identité de Be- +zout entre pn(f) et � +n̸=m pm(f), voir [KM74]. +(2) On peut appliquer cette proposition à toute variété projective et lisse définie +sur un corps fini et à f leur Frobenius : les polynômes pn(f) vont bien entre +premiers être eux par les conjectures de Weil. +La proposition s’applique aussi aux variétés abéliennes sur un corps quel- +conque. Dans ce cas f est la multiplication par un entier N. +(3) Ces résultats ne s’étendent pas automatiquement à une décomposition de +Chow–Künneth (Conjecture 3.1), notamment pour les variétés sur un corps + +24 +GIUSEPPE ANCONA +fini c’est une question ouverte. On dispose d’une décomposition de Chow– +Kunneth pour les variétés abéliennes [DM91], mais pour cela il faut utiliser +la transformée de Fourier, voir Section 8. +Proposition 4.3. Soient M un motif homologique et f un endomorphisme de M. +Alors f est inversible si et seulement si son action en cohomologie l’est. +Remarque 4.4. +(1) La preuve est élémentaire : on applique Cayley–Hamilton. +(2) Cette question donne une réponse affirmative à la Conjecture 3.3 de conser- +vativité pour le foncteur +I : Mot(k) −→ GrVect +pour les endomorphismes. On peut partiellement étendre le résultat aux +morphismes quelconques de la façon suivante. +Proposition 4.5. Soient M et N deux motifs homologiques et considérons deux +applications f : M → N et g : N → M. Supposons que I(f) et I(g) soient des +isomorphismes en cohomologie, alors f et g sont des isomorphismes. +Remarque 4.6. +(1) C’est une conséquence de la Proposition 4.3 appliquée +aux endomorphismes fg et gf. +(2) On remarquera que ce n’est pas une solution complète de la conservativité +pour I : étant donné un morphisme f il faut être capable d’en construire +un dans l’autre sens. Cette proposition est tout de même utile comme on +peut le voir dans la proposition ci-dessous, ainsi que dans la Section 5. +Proposition 4.7. (Kleiman [Kle68]) Soit A une variété abélienne de dimension g. +Alors son motif homologique admet une décomposition de Künneth +h(A) = +2g +� +n=0 +hn(A) +et un isomorphisme de motifs en algèbres de Hopf graduées +2g +� +n=0 +hn(A) = +2g +� +n=0 +Symn h1(A) +qui donne en cohomologie l’isomorphisme classique 10 H∗(A) = Λ∗H1(A). De plus +A vérifie la conjecture standard de type Lefschetz (Conjecture 3.5 et Conjecture +3.11) +h2g−n(A)(g) = hn(A)∨ ∼= hn(A)(n). +Démonstration. La décomposition de Künneth a déjà été discutée dans la Remarque +4.2(3). +On considère l’inclusion diagonale ∆ : A ֒→ An. Elle induit une application +h(∆) : h(A)⊗n → h(A) qui donne une application Symn h1(A) → hn(A). De façon +analogue on construit une application dans l’autre sens en partant du morphisme +de somme s : An → A. +10. Le changement de puissance symétrique en alternée est le même que celui dans la Remarque +3.21(1). + +MÉMOIRE HDR +25 +On peut maintenant appliquer la Proposition 4.5 et déduire �2g +n=0 hn(A) ∼= +�2g +n=0 Symn h1(A). Les applications construites ci-dessus respectent les structures +d’algèbre de Hopf par définition de ces dernières, ce sont d’ailleurs les structures +d’algèbre de Hopf qui forcent ces applications à être des isomorphismes en cohomo- +logie. +L’égalité h2g−n(A)(g) = hn(A)∨ est donnée par la dualité de Poincaré, voir la +Remarque 3.10. Pour l’autodualité hn(A)∨ ∼= hn(A)(n) on se réduit au cas n = 1, +grâce à l’égalité Symn h1(A) = hn(A). +Pour h1(A)∨ ∼= h1(A)(1) on applique encore la Proposition 4.5. Une applica- +tion est donnée par l’opérateur de Lefschetz Lg−1 : h1(A) → h2g−1(A)(g − 1) = +h1(A)∨(−1). Pour l’autre sens, fixons une isogénie entre A et sa duale A∨, ce qui +donne un isomorphisme h1(A) ∼= h1(A∨). Il s’agit alors de construire une applica- +tion de h1(A)∨(−1) vers h1(A∨). C’est la donné d’un diviseur sur A×A∨, le diviseur +de Poincaré convient. +□ +Remarque 4.8. Les résultats de la proposition ci-dessus sont valables même dans +CHM(k) mais leur preuve est plus délicate et repose sur la transformée de Fourier +pour les anneaux de Chow de variétés abéliennes, voir aussi la Section 8. +Motifs et théorèmes de comparaison. +Exemple 4.9. (Réduction supersingulière et Q-structures.) Soit S une surface K3 +complexe de rang de Picard maximal et choisissons un premier p de bonne réduction +tel que la réduction Sp soit supersingulière. La surface quartique de Fermat avec +p ≡ −1 (4) est un tel exemple [SK79]. Dans ce cas le motif homologique vérifie +h(S) = +1 ⊕ h2(S0) ⊕ +1(−2) +et +h2(S0) = +1(−1)⊕20 ⊕ h2,tr(S). +Le motif h2,tr(S) est appelé motif transcendant. Sa réalisation singulière est une +structure de Hodge de dimension 2 et de type (0, 2). Sa réduction modulo p est +isomorphe à +1(−1)⊕2. +Notons M le motif h2,tr(S)(1), VB sa réalisation singulière, Mp sa réduction +modulo p, et Zp les classes algébriques en caractéristique p, c’est-à-dire Zp = +Hom(1, Mp). Remarquons que VB et Zp sont des Q-espaces vectoriels de dimen- +sion 2. +Considérons les identifications +VB ⊗Q Qℓ = Rℓ(M0) = Rℓ(Mp) = Zp ⊗Q Qℓ, +(4.1) +où la première est donnée par le théorème de comparaison d’Artin entre cohomologie +singulière et ℓ-adique, la deuxième suit du théorème de changement de base propre +et lisse et la dernière vient de la propriété de supersingularité. On peut imaginer +l’identification VB ⊗Q Qℓ = Zp ⊗Q Qℓ comme un isomorphisme de périodes. C’est +ce point de vue qui a inspiré le travail présenté dans la Section 7. +Il est naturel de se demander si les deux Q-structures VB et Zp sont respectées +sous cette identification. Expliquons pourquoi la réponse est non. Si on considère +le cup-produit sur VB et Zp on déduit deux Q-formes quadratiques qB et qZ. Si les +espaces étaient égaux on aurait, en particulier, que ces deux formes quadratiques + +26 +GIUSEPPE ANCONA +seraient isomorphes. Or qB est définie positive par les relations de Hodge–Riemann +et qZ est définie négative par le théorème de l’indice de Hodge 11. +Remarque 4.10. Gardons les notations de l’exemple ci-dessus. C’est intéressant de +regarder les Q-formes quadratiques qB et qZ non seulement à la place à l’infini mais +aussi aux autres completions de Q. L’identification (4.1) montre l’égalité qB ⊗Qℓ = +qZ ⊗ Qℓ pour tout ℓ ̸= p. La seule place qui reste à déterminer est en p. Or elle +est déterminée par les autres places 12 et en particulier qB ⊗ Qp ̸= qZ ⊗ Qp puisque +qB ⊗ R ̸= qZ ⊗ R. +Une remarque amusante : observons que la forme quadratique qZ reconnait le +nombre premier p, c’est en effet le seul nombre premier pour lequel qB ⊗ Qp ̸= +qZ ⊗ Qp. En particulier si on fait varier p parmi tous les premiers à réduction +supersingulière on trouvera des Q-formes quadratiques toujours différentes. +Plus sérieusement, remarquons que les Qp-espaces VB ⊗QQp et Zp⊗QQp ont une +interprétation cohomologique : VB ⊗Q Qp est isomorphe à la réalisation p-adique +de M, encore par le théorème de comparaison d’Artin, et Zp ⊗Q Qp est la partie +Frobenius invariante de la réalisation crystalline de Mp. Il est naturel alors de se +demander si la relation qB⊗Qp ̸= qZ⊗Qp peut s’obtenir par des méthodes purement +p-adiques, par exemple via la théorie de Hodge p-adique. La réponse est oui, elle +sera esquissé dans la Section 6. Ces techniques permettront par ailleurs de montrer +des cas de la Conjecture de positivité 3.7, voir encore la Section 6. +Exemple 4.11. (Périodes.) Considérons la catégorie CHM(Q) des motifs définis +sur le corps Q. Elle est munie de deux foncteurs vers les Q-espaces vectoriels gradués +RB, RdR : CHM(Q) −→ GrVectQ +la réalisation singulière, ou de Betti, et la réalisation de de Rham algébrique. Les +théorèmes de comparaison fournissent une identification +RB ⊗Q C = RdR ⊗Q C +essentiellement induite par l’integration des formes différentielles algébriques sur +des simplexes topologiques. Les coefficients complexes qui apparaissent dans cette +identification sont appelés périodes et sont attendus être aussi transcendants que +possible. +Le formalisme tannakien montre l’existence d’isomorphismes +RB ⊗Q Q ∼= RdR ⊗Q Q +de foncteurs monoïdaux (mais on ne sait pas en construire un explicitement). Mon- +trons qu’en revanche il ne peut pas y avoir d’isomorphisme monoïdal entre RB et +RdR et même que l’on a +RB ⊗Q R ̸∼= RdR ⊗Q R. +Pour le montrer il suffit de trouver un motif M muni d’une application +q : Sym2 M −→ +1 +11. Voir respectivement la Définition 2.6 et la Conjecture 3.15, qui est connue pour les divi- +seurs. Les formes quadratiques qu’y apparaissent sont obtenues à partir du cup-produit par un +changement de signe. +12. Cela suit de la formule du produit sur les symboles de Hilbert, l’argument sera détaillé dans +la Section 6. + +MÉMOIRE HDR +27 +telle que les réalisations RB(q) et RdR(q) sont deux formes quadratiques avec si- +gnature différente. +Par exemple on peut prendre M tel que RB(M) soit une structure de Hodge +de poids 0 et types (−1, 1) et (0, 0) et q tel que RB(q) soit une polarisation de +RB(M), voir Définition 2.6. Alors la signature de la forme quadratique RB(q) est +(dim RB(M)0,0, 2 · dim RB(M)−1,1). +D’autre part la réalisation RdR(q) respecte la filtration de de Rham et donc +l’espace Fil1RdR(q) est isotrope ce qui force la signature (s+, s−) de RdR(q) à +vérifier s+ ≥ dim Fil1RdR(q). Or dim Fil1RdR(q) = dim RB(M)−1,1, il suffit donc +de trouver un exemple où l’on a dim RB(M)−1,1 > dim RB(M)0,0 pour conclure. +De tels exemples se trouvent dans la catégorie engendrée par les courbes d’équa- +tion Cn : y2 = xn−1, voir [Sch15]. Plus précisément M sera 13 la partie G-invariante +du motif h2(CN +n )(1) pour certains entiers n, N et pour un groupe fini G convenable +agissant sur CN +n et défini sur Q. +Notons par ailleurs que, par le théorème de comparaison de de Rham, RB ⊗Q R +est isomorphe à la cohomologie de de Rham classique du lieu complexe sous-jacent, +calculée à l’aide des formes différentielles C∞ à coefficients réels. En particulier on +vient de montrer que, pour les variétés algébriques définies sur R, il ne peut pas +y avoir d’isomorphisme naturel entre la cohomologie de de Rham algébrique de la +variété et la cohomologie de de Rham classique du lieu complexe. +Motifs de variétés abéliennes CM. La Proposition 4.7 montre que le motif +d’une variété abélienne A est contrôlé par son h1(A). Si A est une variété abélienne +CM on peut utiliser l’action CM pour décomposer h1(A) et donc le motif de A +tout entier. Cette décomposition est bien utile : Clozel l’avait déjà utilisée pour +un résultat sur la conjecture hom = num (voir le Théorème 5.4) et on l’utilisera +également pour les Conjectures de conservativité et de positivité, voir les Section 5 +et 6. +D’autre part, bien qu’élémentaire, cette décomposition n’est pas digeste à la +première lecture, on encourage à y revenir au fur et à mesure de ses applications. +Exemple 4.12. Soit A une variété abélienne simple de dimension g définie sur un +corps fini. Fixons une polarisation et considérons l’involution de Rosati induite sur +End(A) ⊗ Q. Par Honda–Tate il existe un corps CM de degré 2g et une inclusion +F ⊂ End(A) ⊗ Q tels que l’involution laisse stable F et agit comme la conjugaison +complexe sur F. +Considérons l’action de F sur le motif h1(A) dans la catégorie CHM(k)� +F des +motifs de Chow à coefficients dans une clôture galoisienne �F de F. Elle décompose +le motif +h1(A) = L1 ⊕ . . . ⊕ L2g +(4.2) +en une somme de 2g facteurs échangés par l’action du groupe de Galois Gal( �F/Q) +et la réalisation de chaque facteur est une droite propre pour l’action de F. De plus +le choix d’une polarisation induit un morphisme dans CHM(k) +q1 : h1(A) ⊗ h1(A) −→ +1(−1). +13. Remarquons que le premier candidat que l’on pourrait imaginer, à savoir M = h2(S)(1) +avec S une surface, ne peut pas fonctionner. En effet pour les surface on a l’inégalité opposée : +h1,1 > h0,2, voir [Sch15, Proposition 22]. + +28 +GIUSEPPE ANCONA +Par rapport à cet accouplement, une droite propre est orthogonale à toutes les +autres hormis sa conjuguée complexe. +À l’aide de la formule hn(A) = Symn h1(A) on déduit une décomposition de +hn(A) dans CHM(k)� +F en somme de facteurs dont la réalisation a dimension un : +chaque facteur correspond au produit tensoriel de n différents Li. De plus l’accou- +plement qn = Symn q1 rend la réalisation d’une telle droite orthogonale à toutes les +autres hormis sa conjuguée complexe. +Remarque 4.13. Gardons les notations de l’exemple ci-dessus. +(1) Soient α1, . . . , α2g les valeurs propres de l’action du Frobenius sur h1(A) +comptées avec multiplicité. Quitte à les renuméroter on a αi · α2g−i = q, +où q est le cardinal du corps de base. Cette symétrie des valeurs propres +provient de l’accouplement parfait q1. +Les valeurs propres de l’action du Frobenius sur hn(A) sont données par +tous les produits possibles de n distincts αi. La dimension de l’espace des +classes Galois-invariantes dans H2n(A)(n) est alors donnée par le nombre +de collections {αi1, . . . , αi2n} vérifiant +αi1 · . . . · αi2n = qn. +(4.3) +La conjecture de Tate prédit que chaque droite propre de H2n(A)(n) cor- +respondant à une telle collection contient une classe algébrique. +Cette conjecture est connue pour les diviseurs. Une droite propre contient +donc une intersection de diviseurs si et seulement si la collection αi1, . . . , αi2n +vérifie +αij · αi2n−j = q, +∀j, +(4.4) +quitte à renuméroter. +(2) Soit F0 le plus grand sous-corps totalement réel de �F. Le groupe de Galois +Gal( �F /F0) est d’ordre deux, engendré par la conjugaison complexe. Son +action recolle la décomposition de l’exemple ci-dessus en une décomposition +de hn(A) dans la catégorie CHM(k)F0 des motifs de Chow à coefficients dans +F0. Cette décomposition est orthogonale par rapport à l’accouplement qn. +Les facteurs obtenus sont de rang un ou deux. +(3) L’action du groupe de Galois Gal( �F /Q) recolle la décomposition de l’exemple +ci-dessus en une décomposition de hn(A) dans la catégorie CHM(k) des mo- +tifs de Chow à coefficients rationnels. Cette décomposition est orthogonale +par rapport à l’accouplement qn. Le rang des facteurs obtenus varie et vaut +au plus 2g. +Par la description du point (1) on remarque que chaque facteur de +h2n(A)(n) rentre dans une des trois catégories suivantes : +(a) La réalisation du facteur est engendrée par des classes qui sont toutes +intersections de diviseurs, +(b) La réalisation du facteur est Frobenius invariante mais ne contient au- +cune intersection de diviseurs, +(c) Le Frobenius agissant sur la réalisation du facteur n’a aucun vecteur +fixe. + +MÉMOIRE HDR +29 +Pour les questions de cycles algébriques c’est surtout la classe (b) qui est +intéressante. Si A est de dimension quatre les facteurs de ce type ont tou- +jours rang deux. Pour le montrer il s’agit d’étudier les quadruplets vérifiant +(4.3) mais qui ne vérifient pas (4.4). C’est une étude élémentaire mais dont +la combinatoire est délicate, voir [Anc21, §7] pour les détails. +5. Autodualité et conservativité +Cette section concerne les Conjectures 3.3 et 3.5 de conservativité et d’autodua- +lité et les résultats que l’on peut obtenir pour les variétés abéliennes. On notera +CHM(k)ab, +Mot(k)ab +et +NUM(k)ab +les catégories de motifs engendrées par les motifs de variétés abéliennes. +On commence par rappeler les théorèmes fondamentaux de semisimplicité de +Jannsen et de nilpotence de Kimura, puis on en déduit les Conjectures 3.3, et 3.5 +pour CHM(k)ab, avec k = C. +Dans une deuxième partie on explique le contenu de [Anc22] qui étudie ces +conjectures pour k = Fq. Il faudra combiner les théorèmes de Jannsen et Kimura +avec les décompositions de l’Exemple 4.12 induites par la multiplication complexe. +Cette méthode est inspirée par un travail de Clozel [Clo99] que nous rappellons +également. +Théorème 5.1. (Jannsen [Jan07]) La catégorie NUM(k) des motifs numériques +est semisimple. +Théorème 5.2. (Kimura–O’Sullivan [Kim05, O’S05]) Le noyau du foncteur de +projection +πnum : CHM(k)ab −→ NUM(k)ab +est nilpotent. En particulier le foncteur πnum est conservatif et toute décomposition +dans NUM(k)ab se relève en une décomposition dans CHM(k)ab. L’énoncé reste +valable si on remplace NUM(k)ab par Mot(k)ab (ou n’importe quelle catégorie ten- +sorielle quotiente). +Proposition 5.3. La Conjecture d’autodualité 3.5 et la conjecture hom = num +sont vraies pour tout motif dans CHM(C)ab. De plus le foncteur de réalisation +singulière +R : CHM(C)ab −→ GrVectQ +est conservatif (cf. Conjecture 3.3). +Démonstration. La conjecture standard de type Lefschetz est vraie pour les variétés +abéliennes (Proposition 4.7). En caractéristique zéro, on peut en déduire hom = +num (Remarque 3.17). +De plus, on peut munir hn(A) d’un accouplement +hn(A) ⊗ hn(A) → +1(−n) +dont la réalisation singulière est une polarisation (Proposition 3.14). On en déduit +que pour tout facteur direct M du motif homologique hn(A) la restriction de l’ac- +couplement à M induit un isomorphisme M ∼= M ∨(−n). (Ce fait est impliqué par +la Proposition 2.13 et c’est le point crucial où on l’utilise que les motifs sont définis + +30 +GIUSEPPE ANCONA +sur C. ) L’autodualité pour les motifs homologiques se relève aussi dans CHM(C)ab +par le Théorème 5.2. +Passons maintenant à la conservativité et considérons le diagramme (0.2). Encore +par le Théorème 5.2, il suffira de démontrer la conservativité de la réalisation des +motifs homologiques I : Mot(C)ab → GrVectQ. D’autre part, la conjecture hom = +num déduite au début de la preuve dit que les catégories Mot(C)ab et NUM(C)ab +coincident. En particulier, par le Théorème 5.1, I est un foncteur entre catégories +semisimples, il est donc conservatif. +□ +Dans la suite de la section on travaille sur un corps fini k = Fq. Ce qui rem- +placera l’utilisation de la polarisation dans la preuve ci-dessus est la multiplication +complexe, via les décompositions de l’Exemple 4.12. +Théorème 5.4. (Clozel [Clo99]) Soit A une variété abélienne sur un corps fini. +Alors il existe une infinité de nombres premiers ℓ tels que l’équivalence numérique +coïncide avec l’équivalence homologique pour la cohomologie ℓ-adique. +Démonstration. On se ramène au cas où A est simple. Soient Q ⊂ F0 ⊂ �F les corps +de nombres introduits dans l’Exemple 4.12 et la Remarque 4.13(2). (Ces corps +dépendent de A et plus précisément du choix d’un corps CM dans ses endomor- +phismes.) +Fixons un nombre premier ℓ tel qu’il existe une place λ de F0 au-dessus de ℓ telle +que la complétion (F0)λ ne contienne pas �F. On va montrer qu’un tel ℓ convient. +Remarquons qu’il y a une infinité de tels ℓ et que l’on peut estimer leur densité +avec Chebotareff. +Considérons les classes algébriques de codimension n que l’on voit comme classes +dans h2n(A)(n) et soit q2n l’accouplement construit dans l’Exemple 4.12. Puisqu’il +est non dégénéré il suffit de voir que pour chaque classe algébrique non nulle γ il +existe une classe algébrique δ telle que q2n(γ, δ) ̸= 0. On vérifie que la question est +stable par changement de coefficients et on travaille avec les cycles à coefficients +dans F0 et la réalisation λ-adique Rλ à valeurs dans les (F0)λ-espaces vectoriels. +Dans ce cas on utilise la décomposition de la Remarque 4.13(2) en plans et droites. +Il suffit alors de travailler avec un seul de ces facteurs M et supposer que γ vit +dans M. Dans ce cas, si la forme quadratique q2n est sans vecteur isotrope sur M +le choix δ = γ convient. +Si M a dimension un cela suit du fait que q2n est non dégénérée sur chaque +facteur de la décomposition et donc sur M. Si M a dimension deux alors il admet +au plus deux droites isotropes. Ces deux droites existent au moins sur M ⊗F0 �F : il +s’agit de la décomposition en droites de l’Exemple 4.12. Montrons que ces droites +ne sont pas contenues dans le (F0)λ-espace vectoriel Rλ(M). Pour cela il suffira de +construire un endomorphisme f : M → M dans la catégorie CHM(k)F0 dont ces +droites sont des droites propres et de valeurs propres appartenant à �F − F0. Cela +donnera la conclusion voulue puisqu’on a que (F0)λ ne contient pas �F. +La construction de ce f procède ainsi. Considérons la décomposition (4.2). Quitte +à changer la numérotation, le motif M est de la forme +M = (L1 ⊗ . . . ⊗ L2n) ⊕ (¯L1 ⊗ . . . ⊗ ¯L2n), +où ¯· est la conjugaison complexe. Fixons un ordre sur les Li, ceci permet de réaliser +M comme facteur direct de h1(A)⊗2n dans la catégorie CHM(k)F0. On peut alors + +MÉMOIRE HDR +31 +définir f par l’action induite par un générateur de F ⊂ End(A) ⊗ Q sur le premier +terme du produit tensoriel et l’identité sur les autres 2n − 1. +□ +Proposition 5.5. La conjecture d’autodualité 3.5 est vraie pour les motifs de +CHM(Fq)ab de poids pair et dont la réalisation a dimension un. +Démonstration. Soient X un tel motif et n son poids (pair). On a une variété +abélienne A, telle que X est facteur direct de hn(A). +On veut montrer que X ∼= X∨(−n). Par le Théorème 5.2 il suffit de montrer +πnum(X) ∼= πnum(X)∨(−n). Par la semisimplicité de Jannsen, il suffit alors de +montrer que Hom(πnum(X), πnum(X)∨(−n)) ̸= 0. Cet énoncé peut se démontrer +après extension des scalaires. On étend les scalaires au corps F0 de la Remarque +4.13(2). +On utilise la décomposition de cette même remarque. Par semisimplicité on peut +supposer que X soit un facteur direct d’un des facteurs M de cette décomposition. +Rappelons que l’on dispose d’un accouplement qn non-dégénéré sur M et que M a +dimension un ou deux. +Si M a dimension un alors X = M et on a terminé. Si M a dimension deux, +remontons aux motifs homologiques, via le Théorème 5.2. On pourra alors utiliser +la réalisation et il suffira de montrer que l’accouplement restreint à la droite qui +est la réalisation de X reste non-dégénéré, autrement dit que la droite n’est pas +isotrope. (C’est ici que l’on utilisera que le poids est pair. Remarquons notamment +que si le poids est impair l’accouplement sur M est alterné et donc toute droite est +isotrope.) +La subtilité est que la catégorie des motifs homologiques dépend a priori de +la cohomologie choisie mais la bonne nouvelle est qu’il suffit d’étudier une seule +cohomologie bien choisie. On utilise alors la cohomologie λ-adique comme dans +la preuve du Théorème 5.4 avec le même choix de λ : on y avait montré que la +réalisation de M est sans vecteurs isotropes. +□ +Théorème 5.6. Les foncteurs de réalisation ℓ-adique +Rℓ : CHM(Fq)ab −→ GrVectQℓ +sont conservatifs. +Démonstration. Soit f : X → Y une application dans CHM(Fq)ab telle que Rℓ(f) : +Rℓ(X) → Rℓ(Y ) soit un isomorphisme. On veut montrer que f est un isomorphisme +également. +Par le Théorème 5.2 il suffira de travailler avec l’équivalence homologique. En +utilisant la décomposition de Künneth, il suffira de supposer que Rℓ(X) et Rℓ(Y ) +sont concentrés en un même degré cohomologique. +Supposons d’abord que Rℓ(X) et Rℓ(Y ) aient dimension un. On dispose d’ap- +plications +1 −→ Y ⊗ X∨ ∼= Y ∨ ⊗ X −→ +1, +où la première et la dernière application sont obtenues par adjonction à partir de +f et l’isomorphisme central vient de la Proposition 5.5. +Dans ce cas les réalisations des applications ci-dessus sont des isomorphismes et +par la Proposition 4.5 on a +1 ∼= Y ⊗ X∨ donc X ∼= Y . + +32 +GIUSEPPE ANCONA +Travaillons maintenant dans le cas général : soit d la dimension de Rℓ(X) et +Rℓ(Y ). Supposons que leur degré cohomologique soit pair (sinon il faudra remplacer +des produits extérieurs par des produits symétriques dans la suite). L’application +Λdf : ΛdX −→ ΛdY +retombe dans le cas particulier de la dimension un traité au-dessus. C’est donc un +isomorphisme et on dispose de l’application (Λdf)−1. +On peut maintenant construire une application g : Y → X via +Y ∼= ΛdY ⊗ (Λd−1Y )∨ −→ ΛdX ⊗ (Λd−1X)∨ ∼= X +où le premier et le dernier isomorphisme viennent du lemme ci-dessous et l’applica- +tion centrale est (Λdf)−1⊗(Λd−1f)∨. Par construction, Rℓ(g) est un isomorphisme, +on conclut alors par la Proposition 4.5. +□ +Lemme 5.7. ([O’S05, Lemma 3.2]) Soit M un motif homologique dont la réalisa- +tion est concentrée en un degré pair et de dimension d, alors +M ∼= ΛdM ⊗ (Λd−1M)∨ +Démonstration. Le motif ΛdM est un facteur direct de M ⊗ Λd−1M. Ceci fournit +deux applications entre M et ΛdM ⊗ (Λd−1M)∨ dans les deux directions. Leur +réalisation est un isomorphisme, on conclut par la Proposition 4.5. +□ +6. Positivité en caracteristique positive +Dans cette section nous étudions la Conjecture de positivité 3.7. Le résultat +principal dit que la conjecture est vérifiée pour les motifs de dimension 2 et à +réduction supersingulière (Théorème 6.1). Nous expliquons ensuite comment ap- +pliquer ce résultat pour déduire la Conjecture standard de type Hodge 3.15 pour +certaines variétés, par exemples les variétés abéliennes de dimension quatre. Puis +nous discutons le rôle de l’hypothèse de dimension 2. +Théorème 6.1. Soient M un motif homologique sur un corps k de caractéristique +p et q : Sym2 M → +1 un morphisme dans Mot(k). Supposons que q soit la réduction +modulo p d’une application �q : Sym2 � +M → +1 définie en caractéristique zéro. +Définissons qZ comme la restriction de q à toutes les classes algébriques Z(M) = +Hom(1, M) de M et qB comme la réalisation singulière de �q. +Supposons que qB soit une polarisation et supposons avoir M ∼= +1⊕2. Alors qZ +est définie positive. +Remarque 6.2. (Sur les hypothèses : relevabilité et rang 2.) Il n’est pas rare d’avoir +des motifs qui se relèvent à la caractéristique zéro. Par exemple il est attendu que +tout motif sur un corps fini se relève, car la conjecture de Tate prédit qu’un tel +motif serait de type abélien. En général, même si un motif se relève, ses classes +algébriques ne se relèveront pas à la caractéristique zéro, ce qui rend les résultats de +positivité difficiles, puisqu’ils ne peuvent pas se déduire des propriétés de positivité +des polarisation, voir Définition 2.6. +L’hypothèse restrictive dans le théorème ci-dessus est la dimension deux. La +façon d’utiliser ce résultat pour déduire la conjecture standard de type Hodge pour +certaines variétés est la suivante. On décompose le motif d’une variété donnée autant +que possible. Certains facteurs ne posséderont pas de classes algébriques, d’autres + +MÉMOIRE HDR +33 +en posséderont uniquement certaines pour lesquels la conjecture standard de type +Hodge peut se déduire des cas connus : par exemple ce sont des classes qui se +relèvent à la caractéristique zéro ou qui sont construites à partir de diviseurs. Enfin, +ils resteront parfois des facteurs qui possèdent des classes algébriques qui ne se +ramènent pas à des cas connus. Le point est alors de trouver des variétés pour +lesquels ces derniers facteurs sont de dimension deux. Un exemple est donné dans +le corollaire ci-dessous. +Corollaire 6.3. Soit A une variété abélienne de dimension quatre définie sur un +corps de caractéristique p. Alors +(1) La Conjecture standard de type Hodge 3.15 est vraie pour A, +(2) Le produit d’intersection +CH2(A)/num × CH2(A)/num −→ Q +est de signature (ρ2 − ρ1 + 1; ρ1 − 1), où ρn = dimQ(CH2(A)/num), +(3) Il y a une infinité de nombres premiers ℓ ̸= p pour lesquels l’équivalence nu- +mérique sur A coïncide avec l’équivalence homologique pour la cohomologie +ℓ-adique. +Démonstration. Par un argument de spécialisation on peut supposer que le corps de +définition est fini. On peut alors utiliser la décomposition de la Remarque 4.13(3). +Les facteurs qui sont a priori mystérieux pour la Conjecture standard de type Hodge +3.15 sont ceux de type (b), dans la notation de la même remarque. Il se trouve que +tous ces facteurs de toutes les variétés abéliennes de dimension quatre sont bien de +dimension deux. Ce fait est un petit miracle combinatoire, voir [Anc21, §7] pour +les détails ou la Remarque 4.13(3) pour un aperçu. +Les points (1) et (2) sont en fait équivalents. Cette équivalence n’utilise pas le +fait que A est une variété abélienne mais uniquement la dimension quatre. Elle se +déduit de la décomposition en parties primitives. +Si le corps de définition est fini le point (3) est un cas particulier du Théorème 5.4 +de Clozel. Pour se ramener aux corps finis on spécialise et on utilise la Proposition +3.14. +□ +Remarque 6.4. +(1) (Dimension supérieure.) Pour les variétés abéliennes de +dimension quelconque on pourra encore utiliser la décomposition de la Re- +marque 4.13(3). En général les facteurs (b) de la remarque auront dimension +plus grande que deux. On peut tout de même trouver des exemples spo- +radiques pour lesquels ces facteurs de type (b) ont rang 2 et déduire la +conjecture standard de type Hodge à l’aide du Théorème 6.1. Ceci a été +récemment étudié par Koshikawa [Kos22]. +(2) (Supersingularité vs Frobenius invariant.) On remarquera un petit décalage +entre l’hypothèse de supersingularité M ∼= +1⊕2 du Théorème 6.1 et la +caractérisation des facteurs (b). Tout d’abord remarquons que ces deux +descriptions sont équivalentes sous la conjecture de Tate. +Inconditionnellement, a priori, parmi les facteurs de type (b) certains +pourraient ne pas posséder de classe algébrique : ces facteurs pourront être +négligés à l’étude de la conjecture standard de type Hodge. Pour les autres +on a besoin de montrer que dès qu’un facteur a une classe algébrique il est + +34 +GIUSEPPE ANCONA +engendré par des classes algébriques. Ceci se montre en utilisant l’action +CM mais l’argument nécessite de travailler avec l’équivalence numérique. +Si on travaillait avec l’équivalence homologique on se trouverait devant +des problèmes similaires à ceux qui empêchent l’argument de Clozel du +Théorème 5.4 de fonctionner pour tout nombre premier ℓ. +Définition 6.5. (Symbole de Hilbert.) Soit q une Q-forme quadratique de dimen- +sion deux et ν = 2, 3, 5, . . ., ∞ une place de Q. On définit le symbole de Hilbert +εν(q) de q en ν comme étant +1 si q(x, y) = z2 a une solution non-nulle dans la +complétion Qν et −1 sinon. +Proposition 6.6. Gardons les notations du Théorème 6.1 et soit n l’unique entier +tel que la réalisation singulière de � +M soit une structure de Hodge de type (−n, +n). +Alors qZ est définie positive si et seulement si +εp(qZ) = (−1)nεp(qB), +(6.1) +ce qui est encore équivalent au fait que qZ ⊗ Qp est isomorphe à qB ⊗ Qp si et +seulement si n est pair. +Remarque 6.7. L’idée de la preuve de cette proposition a déjà été introduite dans +l’Exemple 4.9 et la remarque qui le suit. Cette proposition est par ailleurs le point +crucial où l’hypothèse de la dimension deux est nécessaire. +Démonstration. Tout d’abord on remarque que, pour tout nombre premier ℓ ̸= p, +on a qB ⊗ Qℓ ∼= qZ ⊗ Qℓ, c’est la combinaison du théorème de comparaison d’Artin +et du changement de base propre et lisse en cohomologie ℓ-adique. Cela implique +en particulier que εℓ(qZ) = εℓ(qB), mais aussi que le discriminant de qB et qZ +coïncident dans Q∗/(Q∗)2, car un nombre rationnel est un carré s’il l’est dans +presque toute complétion. +D’autre part, en suivant la Définition 2.6, on a que qB est (−1)n-définie positive. +Cela implique en particulier que ε∞(qB) = (−1)n, et que le discriminant de qB +est positif. On en déduit que qZ a discriminant positif et donc, puisqu’on est en +rang deux, que qZ est définie positive ou définie négative. La positivité de qZ est +équivalente alors à ε∞(qZ) = +1. +A l’aide de la formule du produit sur les symboles de Hilbert on a +� +ν +εν(qZ) = 1 = +� +ν +εν(qB). +(6.2) +En simplifiant les facteurs ℓ-adiques on obtient ε∞(qZ)·εp(qZ) = ε∞(qB)·εp(qB) et +donc ε∞(qZ)·εp(qZ) = (−1)n ·εp(qB). On conclut que ε∞(qZ) = +1 si et seulement +si la formule (6.1) est satisfaite. +La dernière équivalence de la proposition suit du fait que deux Qp-formes qua- +dratiques non-dégénérées de même rang sont isomorphes si et seulement si elles ont +le même discriminant et le même symbol de Hilbert. +□ +Démonstration du Théorème 6.1. Par la proposition précédente on est ramené à un +problème purement p-adique. Ce dernier a en plus une interprétation cohomologique +qui permet de le traduire en une question de théorie de Hodge p-adique. Pour +expliquer cette traduction définissons VB,p comme la réalisation étale p-adique de +� +M et VZ,p comme la partie Frobenius invariante de la réalisation cristalline de M. + +MÉMOIRE HDR +35 +Chacun de ces deux Qp-espaces vectoriels de dimension deux est muni d’une forme +quadratique induite respectivement par �q et q. Ces deux formes quadratiques ne sont +rien d’autre que qB ⊗Qp et qZ ⊗Qp. La première identification suit du théorème de +comparaison d’Artin. La deuxième vient du fait que la partie Frobenius invariante +contient toujours l’espace engendré par les classes algébriques et dans ce cas cette +inclusion est une égalité par dimension. +Le théorème de comparaison p-adique montré par Faltings fournit un isomor- +phisme +VB,p ⊗ Bcris = VZ,p ⊗ Bcris +(6.3) +fonctoriel, compatible à toutes les structures que l’on pourrait imaginer et en par- +ticulier avec les formes quadratiques qB et qZ. On conclut à l’aide des deux phéno- +mènes suivants. +(a) La matrice de changement de base est calculable dans Mat2×2(Bcris). Elle +ne dépend que de l’entier n de la filtration de Hodge et de l’algèbre End(VB,p) des +endomorphismes de VB,p comme représentation galoisienne. +(b) La description de cette matrice est suffisante pour déduire que qB ⊗ Qp et +qZ ⊗ Qp sont isomorphes si et seulement si n est pair. +□ +On passe maintenant à la description des deux phénomènes (a) et (b) de la fin +de la preuve ci-dessus. Si le deuxième est élémentaire le premier est une propriété +remarquable de la théorie de Hodge p-adique qui la différencie de la théorie de +Hodge classique. +Exemple 6.8. Supposons avoir deux R-formes quadratiques définies et de dimen- +sion deux q1(x, y) = a1x2 + b1y2 et q2(x, y) = a2x2 + b2y2 et une identification +q1 ⊗R C = q2 ⊗R C. Supposons savoir que la matrice de changement de base de +l’identification est +�i +0 +0 +i +� +. +On pourra alors en déduire que q1 et q2 ne sont pas isomorphes (une est définie +négative et l’autre est définie positive). Les arguments au point (b) dans la preuve +du Théorème 6.1 sont tout aussi élémentaires et ressemblent à cet exemple, avec R +et C qui sont remplacés par Qp et Bcris. +Remarque 6.9. (Théorème de comparaison p-adique vs classique.) Le théorème +de comparaison p-adique +Rp ⊗ Bcris = Rcris ⊗ Bcris, +(6.4) +dont (6.3) en est une instance, est souvent considéré comme l’analogue p-adique du +théorème de comparaison entre cohomologie singulière et cohomologie de de Rham +algébrique +RB ⊗ C = RdR ⊗ C +(6.5) +que l’on a discuté dans l’Exemple 4.11. En fait le théorème de comparaison p- +adique a des avantages par rapport à sa version classique, que l’on liste ci-dessous. +(C’est grâce à ces propriétés que l’on peut notamment calculer certaines matrices +de changement de base et déterminer des relations entre leurs entrées, cf. le point +(a) de la preuve du Théorème 6.1.) + +36 +GIUSEPPE ANCONA +Rappelons que les réalisations classiques d’un motif M possèdent des structures +supplémentaires. En particulier, RdR(M) est munie d’une filtration, RB(M) est +munie d’une structure de Hodge, Rp(M) est munie de l’action du groupe de Galois +d’un corps p-adique et Rcris(M) est un ϕ-module filtré, i.e. elle est munie d’une +filtration et d’une action du Frobenius absolu ϕ. +Rappelons aussi que les coefficients des matrices qui apparaissent dans les com- +paraisons (6.4) ou (6.5) sont appelés périodes. +(1) L’anneau Bcris est muni des actions du Frobenius absolu ϕ et du groupe +de Galois absolu de Qp ainsi que d’une filtration 14. La comparaison (6.4) +respecte ces trois structures. Quand on dispose d’objets cohomologiques +suffisamment concrets, comme ceux de (6.3), on peut explicitement décrire +ces structures sur les périodes. +Le corps C en revanche n’est pas muni de structures qui imiterait la filtra- +tion ou la structure de Hodge. Connaître la structure de Hodge sous-jacente +à RB(M) n’aide pas à avoir des informations sur les périodes complexes du +motif M. +(2) Dans [Fal89], Faltings montre une équivalence de catégories entre certains +ϕ-modules filtrés, dits admissibles, et certaines représentations de groupes +de Galois de corps p-adiques, dites cristallines. La condition d’admissibi- +lité est toujours vérifiée par les modules d’origine géométrique, i.e. par les +réalisations de motifs. +Cette équivalence est en plus compatible au théorème de comparaison +(6.4). En particulier, des périodes dans Bcris qui ont un certain comporte- +ment par rapport à ϕ et à la filtration doivent correspondre à un unique +ϕ-module filtré et donc aussi à une unique représentation de Galois. Autre- +ment dit, les périodes p-adique associées à un motif donné sont caractérisées +par leur comportement par rapport à deux structures : Frobenius et filtra- +tion. +Une structure de Hodge est beaucoup plus riche qu’une filtration. Par +exemple, pour deux variétés d’une même famille, les espaces vectoriels filtrés +correspondant seront isomorphes alors que les structures de Hodge ne le +seront pas, en général 15. On ne peut pas avoir une équivalence de catégories +entre ces deux structures cohomologiques. +(De façon informelle, le passage du cas complexe (6.5) au cas p-adique +(6.4) correspond à enrichir la structure de de Rham et à réduire celle de +Hodge suffisamment pour avoir deux structures équivalentes. En effet, la +réalisation cristalline hérite la filtration de de Rham mais elle possède en +plus l’action du Frobenius absolu ϕ. D’autre part, d’après la conjecture +de Mumford-Tate, une structure de Hodge est grosso-modo équivalente à +une représentation de Galois d’un corps de nombres, or sur Rp(M) on ne +regarde que l’action d’un certain de ses sous-groupes de décomposition.) +14. Certaines structures sont définies dans un plus gros anneau noté BdR. On ignore ici ce point +pour simplifier l’exposition. +15. Si on considère la courbe elliptique Et : y2 = x(x − 1)(x − t) pour t ∈ Q − {0, 1} et le motif +Mt = h1(E) alors RdR(Mt) est la donnée d’un Q-espace vectoriel de dimension 2 muni d’une +droite à l’intérieur. En revanche la structure de Hodge RB(Mt) détermine Et à isogénie près, +notamment il y aura des structures de Hodge CM et d’autres qui ne le sont pas. + +MÉMOIRE HDR +37 +(3) La condition d’admissibilité, discutée au point précédent, se trouve être re- +lativement élémentaire à vérifier, grâce à [CF00]. Ceci permet de construire +facilement des ϕ-modules filtrés admissibles et donc des matrices de pé- +riodes avec action de Frobenius et filtration prescrites. +On ne dispose pas de méthode élémentaire de construction de périodes +complexes. Il s’agit d’intégrales de formes différentielles algébriques qui sont +souvent difficiles à calculer. Leurs relations sont prédites par la conjectures +des périodes de Grothendieck et restent mystérieuses. +En résumant, le point (a) de la preuve du Théorème 6.1 est le calcul de la matrice +de périodes associée à (6.3). Ce calcul procède comme suit : on décrit l’action de +ϕ et la filtration sur ces pédiodes (point (1) ci-dessus), puis on montre que cette +description caractérise les périodes en question (point (2)), enfin on construit de +telles périodes (point (3)). +Exemple 6.10. (Un calcul de périodes.) Soit M un motif comme dans le Théorème +6.1 pour lequel on veut montrer la relation (6.1). Cela passe par le calcul de la ma- +trice de périodes de (6.3). Ce calcul dépend de l’entier n (défini dans la Proposition +6.6) et de l’algèbre End(VB,p) des endomorphismes de VB,p comme représentation +galoisienne. +Supposons n = 1 : on doit alors montrer que qZ ⊗ Qp et qB ⊗ Qp ne sont pas +isomorphes, voir la Proposition 6.6. Supposons également que End(VB,p) soit le +corps Qp2, l’unique extension non ramifiée de degré 2 de Qp. Comme Bcris contient +toutes les extensions non ramifiées, on peut alors écrire (6.3) comme +(VB,p ⊗Qp Qp2) ⊗Qp2 Bcris = (VZ,p ⊗Qp Qp2) ⊗Qp2 Bcris. +(6.6) +L’action de End(VB,p) = Qp2 sur VB,p décompose VB,p ⊗Qp Qp2 en deux droites +échangées par le groupe de Galois Gal(Qp2/Qp). En particulier on peut choisir +deux vecteurs vB et wB échangés par le groupe de Galois et appartenant à ces +droites. +L’algèbre Qp2 agit également sur VZ,p, grâce à l’équivalence de catégories de +Faltings, voir aussi le point (2) de la Remarque 6.9. On peut alors construire vZ et +wZ de façon analogue. De plus, comme (6.3) est compatible à cette action, il existe +deux périodes α, β ∈ Bcris telles que +αvB = vZ +et +βvB = vZ. +Je prétends qu’elles satisfont aux relations +ϕ(α) = β +et +ϕ(β) = α +(6.7) +ainsi que +α ∈ Fil1 − Fil2 +et +β ∈ Fil−1 − Fil0. +(6.8) +En effet Gal(Qp2/Qp) est engendré par le Frobenius et d’autre part l’action du +Frobenius sur VB,p et VZ,p est triviale par définition. Quant à la relation sur la +filtration, on la déduit du fait que les droites propres que l’on a construites doivent +être isotropes, or la droite définie par la filtration sur la réalisation cristalline de M +doit aussi l’être, en particulier elle doit coïncider avec une de ces droites propres +(après extension des scalaires). +Les relations (6.7) et (6.8) sont un exemple du principe (1) expliqué dans la +Remarque 6.9. + +38 +GIUSEPPE ANCONA +Considérons maintenant le Qp2 espace vectoriel +P = {γ ∈ Bcris, +ϕ2(γ) = γ, +γ ∈ Fil1 − Fil2 +et +ϕ(γ) ∈ Fil−1 − Fil0}. +Je prétends qu’il a dimension 1, autrement dit que n’importe quel élément de P est +en fait une période de M construite ci-dessus. Ceci est un exemple du principe (2) +expliqué dans la Remarque 6.9. +Pour le montrer remarquons d’abord qu’une période α de M est inversible dans +Bcris. On peut alors considérer l’espace P/α ⊂ Bcris, il correspondra à +α−1P = {λ ∈ Bcris, +ϕ2(λ) = λ, +λ ∈ Fil0 − Fil−1 +et +ϕ(λ) ∈ Fil0 − Fil−1}. +Or cet espace est Qp2 ⊂ Bcris par [Fon94, Théorème 5.3.7]. +Pour conclure construisons une période t avec les propriétés +t, ϕ(t) ∈ B∗ +cris, +ϕ2(t) = p · t, +t ∈ Fil1 − Fil2 +et +ϕ(t) ∈ Fil0 − Fil1. +De ces propriétés on déduit que α = t/ϕ(t) ∈ P, autrement dit α est une période +du motif M, et α · ϕ(α) = 1/p. Cette dernière relation implique que qZ ⊗ Qp et +qB ⊗ Qp ne sont pas isomorphes par un calcul élémentaire qui est analogue à celui +de l’Exemple 6.8. +La construction de t suit le principe (3) expliqué dans la Remarque 6.9. Consi- +dérons le ϕ-module filtré N = Q2 +p muni du Frobenius +ϕ = +� +0 +1/p +1 +0 +� +et de la filtration +Fil−1 = N, Fil0 = Qp · e2, Fil1 = 0. +On vérifie que c’est un ϕ-module admissible, donc il existe une représentation de +Galois V qui lui correspond par l’équivalence de catégorie de Faltings et qui donne +une comparaison V ⊗ Bcris = N ⊗ Bcris = B2 +cris. Puisque cette identification est +compatible à toutes les structures, on voit que les vecteurs de V ⊂ B2 +cris sont +exactement de la forme (t, ϕ(t)) où t satisfait aux propriétés voulues. +Remarque 6.11. (Généralisations possibles.) La partie p-adique de l’argument +présenté se généralise aux motifs de dimension plus grande : les principes généraux +expliqués dans la Remarque 6.9 restent valables, les calculs de l’Exemple 6.10 de- +viennent plus compliqués mais peuvent être traités. Dans un travail avec Adriano +Marmora [AM22] nous avons pu en déduire une généralisation de la formule (6.1). +En suivant l’argument de la Proposition 6.6 cela donne la positivité du symbole de +Hilbert à l’infini ε∞(qZ) = +1. Malheureusement cette information ne suffit pas à +déduire que qZ est défini positive, c’est le point crucial où l’on utilisait l’hypothèse +de dimension 2. +Pour passer à la dimension supérieure il faudrait trouver un invariant défini en +toute place, tel que la place à l’infini soit contrôlée par toutes les places finies et +d’autre part tel que l’invariant à l’infini détermine toute la signature. Une tenta- +tive pourrait passer par la cohomologie galoisienne : les k-formes quadratiques non +dégénérées et de rang donné sont en bijection avec H1(k, O) où O est un k-groupe + +MÉMOIRE HDR +39 +orthogonal de rang convenable. L’application +H1(Q, O) −→ +� +ν +H1(Qν, O). +est injective, c’est le théorème d’Hasse–Minkowski. Son défaut de surjectivité est +justement contrôlé par la formule du produit des symboles de Hilbert. Le problème +déjà soulevé se reformule alors ainsi : l’application +H1(Q, O) −→ +� +ν̸=∞ +H1(Qν, O) +n’est plus injective. +On peut alors essayer d’exploiter plus d’informations géométriques de notre si- +tuation et faire surgir des groupes plus petits. Par exemple les motifs qui appa- +raissent dans le problème sont munis non seulement d’une forme quadratique mais +aussi de l’action d’un corps CM. Ajouter cette donnée au problème correspond à +étudier la cohomologie galoisienne d’un tore maximal T du groupe orthogonal O. +On dispose encore d’un principe local-global : l’application +H1(Q, T ) −→ +� +ν +H1(Qν, T ). +est injective. Le point crucial serait alors d’avoir l’injectivité aussi de l’application +H1(Q, T ) −→ +� +ν̸=∞ +H1(Qν, T ). +Malheureusement elle n’est pas injective : on peut calculer son noyau à l’aide de la +suite exacte de Poitou–Tate. +7. Périodes p-adiques à la André +Dans cette section nous présentons un travail en collaboration avec Dragos Fra- +tila. Les motivations sont d’origine géométrique - l’étude des classes algébriques en +caractéristique p - mais le résultat final est plutôt arithmétique : on construit une +algèbre de périodes p-adiques ainsi qu’un cadre tannakien pour l’étudier. +Des telles périodes devraient avoir des analogies avec les périodes complexes que +l’on a rencontré dans l’Exemple 4.11. Les périodes p-adiques de Fontaine possèdent +des propriétés cohomologiques analogues à celles des périodes complexes et même +plus fortes (Remarque 6.9). Par contre les propriétés arithmétiques des périodes +complexes, comme leur transcendence ou leur lien avec les fonctions spéciales, n’ont +pas de bon analogue dans les périodes p-adiques de Fontaine [And90]. +Pendant que notre travail avançait nous avons découvert qu’André avait tissé +des liens similaires [And95, And03]. Son travail nous a été utile pour raffiner notre +étude et notamment pour formuler la condition de ramification (Définition 7.10). +Motivation. Le point de départ vient d’une remarque de Tate : la conjecture de +Tate prédit l’existence de classes algébriques mais elle ne prédit pas quelle classe +cohomologique est algébrique [Mil07, Aside 6.5]. Une façon d’interpréter cette re- +marque est que l’on a une description du Qℓ-espace vectoriel engendré par les classes +algébrique mais on n’a pas de description du Q-espace vectoriel engendré par ces +dernières. C’est un point délicat qui est présent dès le travail de Tate sur la conjec- +ture de Tate pour les diviseurs sur les variétés abéliennes sur un corps fini [Tat66], + +40 +GIUSEPPE ANCONA +et plus récemment dans le travail de Charles sur la conjecture de Tate pour les +diviseurs sur les surfaces K3 [Cha13]. +Un exemple élémentaire qui illustre cette subtilité est le suivant : il existe des +Qℓ-droites dans la cohomologie ℓ-adique d’une variété X, disons définie sur un corps +fini, qui sont Galois invariantes et pourtant elles ne contiennent pas d’élément du +Q-espace vectoriel Im clX. Pour construire de tels exemples prenons X une surface +abélienne, ou une K3, et fixons deux classes de diviseurs α et β linéairement indé- +pendantes. Prenons maintenant une constante c ∈ Qℓ−Q. Alors la droite engendrée +par α + cβ convient. Pour le montrer on peut utiliser le produit d’intersection et le +fait qu’il est défini à coefficients rationnels. +Un échec : cas ℓ-adique. Faute de savoir décrire le Q-espace vectoriel des classes +algébriques, un premier pas est de le comparer à un autre Q-espace vectoriel. C’est +notamment ce que l’on a fait dans l’Exemple 4.9 et la remarque qui le suit. +Prenons une variété Xp définie sur Fp et supposons qu’elle se relève à une va- +riété X définie sur Q. Au moyen d’un plongement σ : Q ֒→ C on dispose d’une +identification +H∗ +B(X(C), Q) ⊗ Qℓ = H∗ +ℓ (Xp) +(7.1) +induite par le théorème de comparaison d’Artin et le changement de base propre et +lisse. Sous cette identification on peut étudier la position du Q-espace vectoriel Zp +des classes algébriques sur Xp par rapport à H∗ +B(X(C), Q). Le premier fait que l’on +remarque est que l’intersection Zp ∩H∗ +B(X(C), Q) contient le Q-espace vectoriel Z0 +des classes algébriques sur X. Inspiré par les différentes versions de la conjectures +des périodes de Grothendieck on peut se demander si cette inclusion est en fait une +égalité. +Après avoir montré que cette question a réponse affirmative dans certains cas, +nous avons compris que la réponse est négative en général. Les cas affirmatifs sont +les surfaces à rang de Picard maximal - par une méthode similaire celle présentée +dans l’Exemple 4.9 - et les variétés abéliennes CM [AF22, §10]. Il est possible de +construire des contre-exemples avec le carré d’une courbe elliptique non CM. L’ar- +gument suit en fait la technique qui a permis à André de montrer que l’analogue +de la conjecture des périodes de Grothendieck est fausse pour le théorème de com- +paraison p-adique. En effet (7.1) dépend du choix de σ. On peut faire varier σ en +utilisant le groupe de Galois absolu du corps de nombre sur lequel X est défini. Dans +certains cas on sait que l’action de ce groupe de Galois sur H∗ +ℓ (Xp) est hautement +non triviale [Ser72], ce qui permet faire varier le Q-espace vectoriel H∗ +B(X(C), Q) +et notamment de le faire rencontrer Zp de façon inattendue. +Cas p-adique. Fixons une fois pour toutes un plongement Q ⊂ Qp. Prenons +comme auparavant une variété X définie sur Q à bonne réduction et notons Xp +sa réduction. On dispose de la comparaison entre cohomologie de de Rham et co- +homologie cristalline +H∗ +dR(X, Q) ⊗ Qp = H∗ +cris(Xp, Qp) +due à Berthelot. +Notons par Zp le Q-espace vectoriel des classes algébriques sur Xp et par Z0 le +Q-espace vectoriel des classes algébriques sur X. On a comme dans le cas ℓ-adique +l’inclusion +Z0 ⊂ Zp ∩ H∗ +dR(X, Q) + +MÉMOIRE HDR +41 +et on peut encore une fois se demander si cette inclusion est en fait une égalité. +Conjecture 7.1. (pGPCw : Analogue p-adique de la version faible de la conjecture +des périodes de Grothendieck.) +Est-ce que l’inclusion Z0 ⊂ Zp ∩ H∗ +dR(X, Q) est une égalité ? +Pour rendre cette question raisonnable il est nécessaire d’imposer une condition +de ramification que l’on discutera plus tard et que l’on ignore pour l’instant (Défini- +tion 7.10). Cette question apparaît comme l’analogue p-adique de la version faible +de la conjecture des périodes de Grothendieck (appelée parfois conjecture de de +Rham–Betti [And04, §7]). La remarque ci dessous fait le lien entre cette conjecture +et différentes conjectures classiques sur les cycles algébriques. +Tout comme son pendant classique, cette conjecture prédit de la transcendence. +En effet elle prédit qu’une classe algébrique en caractéristique p qui n’est pas re- +levable à la caractéristique zéro ne peut pas être dans H∗ +dR(X, Q), autrement dit, +au moins une de ses coordonnées par rapport à une base de H∗ +dR(X, Q) doit être +transcendante. La version forte de la conjecture de Grothendieck p-adique prédira +de façon précise le degré de transcendance de toutes ces coordonnées (Conjecture +7.13). +Remarque 7.2. (pGPCw vs conjectures classiques.) Comparons maintenant la +question qui a été soulevée au paragraphe précédent, notée (pGPCw), avec trois +conjectures classiques que nous rappelons de façon informelle (voir [And04, §7] pour +plus de détails). Ces trois conjectures sont la conjecture de Hodge (HC), la version +faible de la conjecture des périodes de Grothendieck (GPCw) et la conjecture de +Hodge variationnelle p-adique de Fontaine et Messing (pHC). +(HC) Une classe rationnelle en cohomologie singulière est algébrique si et seule- +ment si elle appartient au bon degré de la filtration de de Rham. +(GPCw) Une classe rationnelle en cohomologie de de Rham est algébrique si et +seulement si elle est rationnelle pour la cohomologie singulière +(pHC) Une classe algébrique en cohomologie cristalline se relève à la caractéris- +tique zéro si et seulement si elle appartient au bon degré de la filtration de +de Rham. +De façon informelle, on peut voir (pGPCw) comme « le produit fibré de (pHC) et +(GPCw) au-dessus de (HC) ». +(pGPCw) +→ +(GPCw) +↓ +↓ +(pHC) +→ +(HC) +La partie droite du diagramme concerne la caractéristique zéro, celle de gauche la +caractéristique mixte. La conjectures du bas comparent une structure rationnelle +et une filtration, celles du haut comparent deux structures rationnelles. +Définition 7.3. (Périodes p-adiques à la André.) Soit M ∈ Mot(Q) un motif +homologique 16 à bonne réduction et soit Mp ∈ Mot(Fp) sa réduction. Notons leurs +16. Tout comme dans le cas classique on travaillera qu’avec des motifs vérifiant hom = num, +par exemple les motifs issues de produits de courbes elliptiques : ceci est nécessaire pour avoir des +catégories tannakiennes. + +42 +GIUSEPPE ANCONA +classes algébriques par +Z0(M) = HomMot(Q)(1, M) +et +Zp(M) = HomMot(Fp)(1, Mp). +Considérons le théorème de comparaison de Berthelot +RdR(M) ⊗Q Qp = Rcris(Mp). +(7.2) +Pour tout choix de base B de Zp(M) et B′ de RdR(M) définissons MatB,B′(M) +comme la matrice ayant comme vecteurs colonnes les coordonnées de B par rapport +à B′. Nous appelons les coefficients de cette matrice les périodes p-adique d’André +de M et définissons Pp(M) ⊂ Qp comme la Q-algèbre engendrée par ces périodes. +Remarque 7.4. (Périodes classiques vs périodes p-adiques à la André.) +(1) Si un élément de B est une classe algébrique qui n’est pas relevable à la +caractéristique zéro au moins une de ses périodes devrait être transcendante +par la Conjecture 7.1. +(2) La matrice MatB,B′(M) dépend bien du choix des bases B et B′, par contre +l’algèbre Pp(M) n’en dépend pas. +(3) Pour que l’espace Zp(M) ne soit pas réduit à zéro il faut que M contienne +des facteurs directs de poids zéro. Il faut typiquement imaginer M = +h2n(X)(n) pour une variété X à bonne réduction. De plus, pour avoir des +périodes intéressantes, il faut que M admette des classes algébriques mo- +dulo p qui ne sont pas relevables, sinon toutes les périodes p-adiques seraient +algébriques. +(4) Contrairement au cas classique, la matrice de périodes p-adique n’est pas +carrée, en effet l’inégalité 17 #B′ ≤ #B est stricte en général. +(5) Il est possible de définir les périodes p-adiques pour les motifs mixtes. Il +faut dans ce cas considérer uniquement les motifs mixtes vérifiant (7.2) - +cette relation n’est automatique pour les variétés ouverte. +Exemple 7.5. +(1) (Courbes elliptiques CM et valeurs Gamma.) Considérons +E une courbe elliptique CM. Les périodes complexes de son h1(E) sont un +produit de certaines valeurs spéciales de la fonction gamma ΓC en certains +rationnels explicites dépendant uniquement du corps CM. Il n’y a pas de pé- +riode p-adique associée à h1(E), puisqu’il n’y a pas de classe algébrique dans +le h1(E), par contre on peut considérer le motif M = h1(E) ⊗ h1(E)∨. Ses +classes algébriques correspondent aux endomorphismes de E. Pour avoir des +périodes p-adiques intéressantes considérons un premier p à réduction su- +persingulière, autrement tous les endomorphismes se relèveraient et toutes +les périodes seraient algébriques. Il s’agit de décrire l’action de ces endo- +morphismes par rapport à une base de H1 +dR(E, Q). Ce genre de calculs +a été traités par Coleman et Ogus [Col90, Ogu90]. On y voit apparaître +des produits de valeurs spéciales de la fonction gamma p-adique Γp en des +rationnels. +17. Point technique : cette inégalité a encore besoin de l’hypothèse hom = num. On utilise le +fait que l’équivalence numérique commute à l’extension des scalaires, voir la Conjecture 3.18 et la +remarque qui la suit. + +MÉMOIRE HDR +43 +(2) (Motifs de Kummer et logarithme.) Considérons le motif de Tate mixte de +type Kummer Ka = h1(Gm, {1, a})∨, avec a ∈ Q. Ce motif s’insère dans +une suite exacte +0 −→ +1(+1) −→ Ka −→ +1 −→ 0 +qui est non scindée pour a ̸= 0, 1, −1. En particulier ce motif n’a pas de +classe algébrique non nulle, qui est la raison d’avoir considéré Ka et non +pas son dual h1(Gm, {1, a}). Sa matrice de périodes complexes est +�2πi +log(a) +0 +1 +� +. +Fixons un nombre premier p. Pour a ̸≡ 0, 1 [p], le motif Ka a bonne +réduction. De plus les motifs de Tate sur un corps fini forment une catégorie +semisimple, en particulier la suite exacte ci-dessus se scinde modulo p. On +en déduit que le motif possède une classe algébrique non nulle modulo p +qui est donc non relevable. Sa matrice de périodes p-adiques est +�logp(a) +1 +� +dont l’analogie avec son pendant complexe est encore une fois frappante. +Cette matrice s’obtient à partir du calcul de la matrice du Frobenius agis- +sant sur RdR(Ka) qui est dû à Deligne [Del89, §2.9]. On y voit apparaître le +logp(a1−p) et on trouve curieux que le passage de la matrice de Frobenius à +la matrice de périodes corrige cet exposant. (La correction de logp(a1−p) à +logp(a) aurait pu s’obtenir en changeant de base, or pour ces motifs on dis- +pose de bases canoniques et toutes les matrices décrites ci-dessus utilisent +uniquement ces bases). +(3) (Fonctions hypergéométriques.) Soient M et N deux motifs non isomorphes +mais dont les réductions modulo p le sont. Alors le motif M ⊗ N ∨ a une +classe algébrique modulo p non relevable qui est justement associée à cet +isomorphisme. Ses périodes p-adiques sont les coordonées de RdR(N) par +rapport à RdR(M). +Par exemple on peut considérer M = h1(E) et N = h1(E′) où E et +E′ sont deux relèvements non isogènes d’une courbe elliptique ordinaire +sur un corps fini. On peut notamment choisir E comme le relèvement ca- +nonique de Serre–Tate et E′ comme une courbe elliptique non CM. Les +périodes p-adiques qui apparaissent dans ce cas là sont décrites par Katz +[Kat80]. On y voit notamment apparaître des valeurs spéciales de fonctions +hypergéométriques. +(4) (Matrice du Frobenius.) Soit f ∈ End(Mp) un endomorphisme modulo p. +On peut considérer la matrice de son action par rapport à une base de +RdR(M). Ses coefficients pourront s’interpréter comme périodes p-adiques +à la André en regardant f comme une classe algébrique modulo p du motif +M ⊗ M ∨. D’intérêt particulier est le cas où f est le Frobenius : certains +auteurs [Fur07, Bro17] ont définis les périodes p-adique associés à M comme +ses coefficients. + +44 +GIUSEPPE ANCONA +Le point de vue des périodes p-adiques à la André est meilleure pour +plusieurs raisons. Entre autres, il donne des bornes plus fine à la trans- +cendance ainsi qu’une interprétation motivique de certaines relations natu- +relles, comme celles provenant du polynôme caractéristique du Frobenius, +voir [AF22, Remark 9.7]. +Transcendance. Comme expliqué dans la Remarque 7.4, la Conjecture 7.1 prédit +la transcendance de certaines périodes p-adiques. Le prochain but est de donner +une borne au degré de transcendance de ces périodes (Théorème 7.6) ainsi qu’une +conjecture qui prédira ce degré (Conjecture 7.13). On montrera que cette dernière +conjecture implique en fait la Conjecture 7.1, voir la Proposition 7.12. +Gardons les notations de la Définition 7.3. Considérons les catégories tanna- +kiennes ⟨M⟩ et ⟨Mp⟩ engendrées par M et Mp ainsi que les groupes tannakiens +GdR(M) et Gcris(Mp) associés aux foncteurs fibres RdR et Rcris. +Dans le cadre des périodes classiques, Grothendieck démontre que leur degré de +transcendence est borné par la dimension de GdR(M). Notre résultat principal en +est l’analogue p-adique. +Théorème 7.6. Le degré de transcendance des périodes p-adiques vérifie l’inégalité +degtrPp(M) ≤ dim GdR(M) − dim Gcris(Mp). +Remarque 7.7. (Transcendance classique vs transcendance p-adique.) +(1) Cette inégalité pourrait sembler plus forte que celle du cas complexe mais +la matrice rectangulaire des périodes p-adiques est en général plus petite +que celle des périodes complexes. Elles ont la même taille uniquement dans +le cas de réduction supersingulière ce qui revient à Gcris(Mp) = {1}. +(2) Dans le cadre complexe, le point crucial est d’interpréter les périodes comme +les coordonnées d’un C-point du foncteur T (M) = Isom⊗ +⟨M⟩(RB, RdR). Par +la théorie tannakienne ce foncteur est représentable par une variété affine +sur Q. L’action naturelle de GdR(M) = Aut⊗ +⟨M⟩(RdR) sur T (M) rend cette +variété un torseur. +Dans le cas p-adique le foncteur Zp n’est pas un foncteur fibre, pour des +questions de dimension, ce qui le fait sortir du cadre tannakien. C’est tout +de même un foncteur lax-monoïdal (ce qui revient à dire que le produit de +classes algébriques est une classe algébrique). +Le foncteur Isom⊗ +⟨M⟩(Zp, RdR) est vide en général, encore pour des rai- +sons de dimension. On peut en revanche considérer les transformations na- +turelles tensorielles ou les plongements. +Théorème 7.8. L’inclusion de foncteurs Emb⊗ +⟨M⟩(Zp, RdR) ⊆ Nat⊗ +⟨M⟩(Zp, RdR) +est une égalité. Ces foncteurs sont représentables par une variété H(M) affine sur +Q. L’action naturelle de GdR(M) sur H(M) est transitive. De plus on a un iso- +morphisme H(M)Qp = GdR(M)Qp/Gcris(Mp). + +MÉMOIRE HDR +45 +Remarque 7.9. (Théorème 7.6 implique Théorème 7.8.) En analogie avec le cas +classique on peut interpréter +Zp ⊗ Qp ֒→ Rcris = RdR ⊗ Qp +(7.3) +comme un Qp-point de H(M). L’évaluation en ce point donne un morphisme d’al- +gèbres +eval : O(H(M)) −→ Qp. +(7.4) +Par construction, l’image de ce morphisme est l’algèbre Pp(⟨M⟩) engendrée par +toutes les périodes p-adiques de tous les motifs appartenant à la catégorie ⟨M⟩. +Contrairement au cas classique, ces périodes contiennent strictement celles de M, +en général. Cela vient du fait que l’inclusion Zp(M)⊗n ⊂ Zp(M ⊗n) est stricte en +général : c’est le défaut d’une formule de Künneth pour les classes algébriques. +Le Théorème 7.8 et les relations +Pp(M) ⊂ Pp(⟨M⟩) = Im eval ⊂ Qp. +(7.5) +prouvent le Théorème 7.6 et même l’inégalité plus forte +degtrPp(M) ≤ degtrPp(⟨M⟩) ≤ dim H(M) = GdR(M) − Gcris(Mp). +(7.6) +La définition suivante est inspirée de travaux d’André [And95, And03]. +Définition 7.10. (Condition de ramification.) On dit qu’un motif N est CM si +End(N) est un corps de nombres tel que dimQ End(N) = dim RdR(N). +On dit que le nombre premier p ne ramifie pas dans ⟨M⟩ si, pour tout N dans +⟨M⟩ qui est CM, le nombre premier p ne ramifie pas dans le corps de nombre +End(N). +Exemple 7.11. Soit A une variété abélienne. Dans le cas où N = h1(A), imposer +que End(N) soit un corps de nombres tel que dimQ End(N) = dim RdR(N) revient +à demander que A soit simple et CM. Dans ce cas demander que p ne ramifie pas +dans End(N) correspond à demander que p ne ramifie pas dans son corps CM. +Proposition 7.12. Pour un motif M ′ fixé, les nombres premiers qui ramifient +dans ⟨M ′⟩ sont en nombre fini. +Conjecture 7.13. (p-GPCs : Analogue p-adique de la version forte de la conjecture +des périodes de Grothendieck.) +Si p ne ramifie pas dans ⟨M⟩ alors l’application d’évaluation (7.4) est injective. +De façon équivalente, l’espace homogène H(M) est connexe et le Qp-point de H(M) +induit par (7.3) vit au-dessus du point générique de H(M) (ou encore l’inégalité +degtrPp(⟨M⟩) ≤ dim H(M) = GdR(M) − Gcris(Mp) est en fait une égalité.) +Proposition 7.14. Si M vérifie la Conjecture 7.13 alors pour tout N dans ⟨M⟩ +on a une +Z0(N) = Zp(N) ∩ RdR(N), +c’est-à-dire la Conjecture 7.1 (pGPCw) a réponse affirmative pour N. + +46 +GIUSEPPE ANCONA +Remarque 7.15. +(1) Les preuves des résultats de cette section utilisent des +techniques tannakiennes. Comme déjà mentionné, notamment dans la Re- +marque 7.7(2), on ne peut pas utiliser les résultats classiques tel quels mais +il faut plutôt adapter leurs preuves. +(2) La condition de ramification (Définition 7.10) est inspirée d’une condition +qu’André a imposée dans l’étude de ce genre de questions pour les variétés +abéliennes à réduction supersingulière. Il avait remarqué que ces variétés +pouvaient avoir des périodes p-adiques vérifiant des relations algébriques +non motiviques. Comme mentionné dans l’Exemple 7.5(1), les périodes qui +apparaissent pour de tels motifs sont liées aux valeurs spéciales de la fonc- +tion Gamma p-adique. Ces dernières se trouvent être plus souvent algé- +briques que leurs analogues complexes. +(3) Assez peu est connu sur la conjecture classique des périodes de Grothen- +dieck. La version forte a été démontrée pour les courbes elliptiques CM par +Chudnovsky [Chu80]. Le cas particulier de la courbe elliptique de Fermat +implique notamment la transcendance de ΓC(1/3). La version faible (voir +(GPCw) de la Remarque 7.2) a été montrée pour les diviseurs sur les varié- +tés abéliennes et sur les surfaces K3 par Bost et Charles [BC16], en utilisant +entre autre le théorème du sous-groupe analytique de Wüstholtz [Wüs89]. +Ce sont des résultats difficiles et on peut s’attendre à ce que leurs analogues +p-adiques le soient aussi. +Le seul cas où la Conjecture 7.13 est vérifiée est pour le motif de Kum- +mer Ka de l’Exemple 7.5(2). Cela suit de la transcendence des valeurs +spéciales du logarithme p-adique [Ber77]. Le premier cas ouvert intéressant +serait celui des courbes elliptiques à réduction supersingulière, ce qui don- +nerait notamment la transcendence de Γp(1/3), pour p ≡ 2[3]. La version +faible semble aussi difficile. Une petit résultat dans cette direction a été +donné dans le cas des courbes elliptique non CM à réduction supersingu- +lière [AF22, Proposition 3.5]. +8. Motifs des schémas en groupes commutatifs +Cette section résume deux travaux en collaboration avec Stephan Enright-Ward, +Annette Huber et Simon Pepin Lehalleur [AEWH15, AHPL16]. Ils portent sur +l’anneau de Chow et le motif d’un schéma en groupes commutatifs et généralisent +les théorèmes de Beauville [Bea86] et Deninger–Murre [DM91] qui traitent le cas +des schémas abéliens. Nous expliquons quelles sont les subtilités qui apparaissent +quand on quitte le cadre des schémas en groupes projectifs. Les motifs de Voevodsky +deviennent essentiels, non seulement leur existence mais aussi la nature de leur +construction : c’est un exemple du principe expliqué au §1.3. +Dans ce qui suit S est une variété de type fini et lisse sur un corps k qui jouera +le rôle d’une base fixée. Tout S-schéma en groupe que l’on considérera sera lisse et +de type fini à fibres connexes. Certains énoncés sont valables dans des meilleures +généralités. Pour un S-schéma lisse f : X → S, CH(X) indique l’anneau de Chow +de l’espace total X. On continue à travailler avec les coefficients rationnels. + +MÉMOIRE HDR +47 +Théorème 8.1. (Beauville [Bea86], Deninger–Murre [DM91]) Soient A un S- +schéma abélien de dimension relative g et nA : A → A le morphisme de multi- +plication par n. Alors on a une décomposition +CHi(A) = +g+i +� +r=i +CHi +(r)(A) +où CHi +(r)(A) = {α ∈ CHi(A), n∗ +Aα = nrα, ∀n ∈ Z}. +Remarque 8.2. +(1) Ce théorème a été démontré par Beauville dans le cas +S = Spec(k) et Deninger–Murre dans le cas général. Les deux résultats +utilisent de façon cruciale la transformée de Fourier que l’on rappelle plus +loin. +(2) Quand S = Spec(k), on a l’inclusion +ker cli +A ⊇ +g+i +� +r̸=2i +CHi +(r)(A) +(8.1) +qui vient du fait que n∗ +A agit sur la cohomologie de degré s comme ns · Id . +(3) La décomposition du théorème induit une bigraduation sur l’anneau CH∗ +(•)(A). +Quand S = Spec(k), la nouvelle graduation scinde la filtration de Bloch– +Beilinson qui est conjecturée avoir certaine propriétés par rapport à l’appli- +cation classe de cycle, notamment l’inclusion (8.1) devrait être une égalité. +La filtration de Bloch–Beilinson est conjecturée exister pour tous les an- +neaux de Chow de toutes les variétés projectives et lisses sur k, mais en +général cette filtration ne se scinde pas. +Démonstration. Soient A∨ le schéma abélien dual, P ∈ CH1(A × A∨) le diviseur +associé au fibré de Poincaré et π1, π2 les projections de A×A∨ sur les deux facteurs. +On définit la transformée de Fourier +FA : CH∗(A) −→ CH∗(A∨), +α �→ (π2)∗(exp(P) · π∗ +1α). +On vérifie que c’est un isomorphisme, dont l’inverse est essentiellement FA∨. On en +déduit la décomposition +CHi(A) = +� +s +{α ∈ CHi(A), +FA(α) ∈ CHs(A)}. +(8.2) +Ensuite un calcul direct permet de voir comment n∗ +A agit sur chaque facteur de la +décomposition (8.2). On retrouve ainsi la décomposition de l’énoncé et l’identifica- +tion CHi +(r)(A) = {α ∈ CHi(A), +FA(α) ∈ CHg+i−r(A)}. +□ +Nous montrons la généralisation suivante. +Théorème 8.3. Soient G un S-schéma en groupes commutatifs de dimension re- +lative d et nG : G → G le morphisme de multiplication par n. Alors on a une +décomposition +CH∗(G) = +2d +� +r=0 +CH∗ +(r)(G) +où CH∗ +(r)(G) = {α ∈ CH∗(G), n∗ +Gα = nrα, ∀n ∈ Z}. + +48 +GIUSEPPE ANCONA +Remarque 8.4. +(1) On ne sait pas définir une transformée de Fourier pour +un tel G : l’application π2 n’est pas propre, donc (π2)∗ n’existe pas, et de +plus le dual d’un tel G est un 1-motif en général et non pas une variété. +(2) Le premier cas non trivial pour les groupes non projectifs est donné par +S = Spec(k) et G qui admet une suite exacte +0 −→ Gm −→ G −→ A −→ 0 +où A est une variété abélienne. Dans ce cas le Théorème 8.3 est facile pour +Gm et connu pour A mais on ne peut pas le déduire directement pour G. +Le problème est que cet énoncé se comporte bien pour les sommes directes +mais mal pour les suites exactes. L’énoncé qui suivra sera plus adapté à ce +genre de dévissage. +Définition 8.5. Soient Sm/S la catégorie des S-schémas lisses de type fini et +PSh(S) la catégorie des préfaisceaux sur Sm/S à valeur dans les Q-espaces vec- +toriels. Soient Q(G) ∈ PSh(S) le préfaisceau qui associe à chaque Y ∈ Sm/S le +Q-espace vectoriel ayant comme base l’ensemble HomS(Y, G) et G ∈ PSh(S) celui +qui associe à chaque Y ∈ Sm/S le Q-espace vectoriel HomS(Y, G) ⊗Z Q. La loi de +groupe de G induit une transformation naturelle +sG : Q(G) −→ G. +Par construction de DM(S), la catégorie des motifs relatifs 18, les préfaisceaux ci- +dessus induisent des motifs et la transformation naturelle un morphisme entre eux +que l’on notera +αG/S : M(G/S) −→ M1(G/S). +Le motif M(G/S) est appelé le motif de G et le motif M1(G/S) est appelé le +1-motif de G. +Théorème 8.6. Gardons les notations de la définition ci-dessus. Alors le motif +Symr M1(G/S) est nul pour r assez grand et le morphisme αG/S se prolonge en un +unique morphisme de motifs en algèbres de Hopf +ϕG/S : M(G/S) −→ +� +r=0 +Symr M1(G/S) +(8.3) +qui est de plus un isomorphisme. +Remarque 8.7. +(1) (Motifs vs faisceaux.) Ce théorème n’est pas une consé- +quence formelle d’un énoncé sur les préfaisceaux. Remarquons par exemple +que le faisceau Symr G n’est pas nul, puisque G est un faisceau en espaces +vectoriels. +Un phénomène plus subtile est le suivant : le théorème montre en parti- +culier l’existence d’applications non nulles de M1(G/S) vers M(G/S). En +revanche, il n’y a pas d’application non nulle du faisceau G vers Q(G). +18. On considère ici uniquement la version stable de cette catégorie, c’est-à-dire la catégorie +obtenue après ⊗-inversion du motif de Lefschetz. Il y a plusieurs descriptions de la catégorie +stable DM(S). Il se trouve qu’elles sont équivalentes sous des hypothèses assez générales qui sont +notamment satisfaites pour les bases S que l’on considère. La version qui est adaptée à la Définition +8.5 est celle des motifs étales étudiés par Ayoub. Cette catégorie DM(S) est obtenue à partir de +D(PSh(S)) en localisant pour imposer la descente étale et l’invariance par A1-homotopie, puis en +stabilisant. + +MÉMOIRE HDR +49 +Esquissons l’argument. Soient α : G → Q(G) une telle transformation na- +turelle et idG ∈ G(G) l’application identité. La naturalité de α implique +qu’il suffit de voir que α(idG) = 0. Posons α(idG) = ai · fi où ai sont des +nombres rationnels et les fi sont des endomorphismes de G. +Pour montrer ai · fi = 0 considérons la naturalité par rapport aux mor- +phismes nG de multiplication par n : +idG +❴ +� +✤ +� � ai · fi +❴ +� +G +nG +� +α � Q(G)(G) +nG +� +G +α � Q(G)(G) +nG = n · idG ✤ +� � nai · fi +� ai · (nG ◦ fi). +On prétend que l’égalité � nai · fi = � ai · (nG ◦ fi) force tous les fi à +être nuls. tout d’abord supposons par l’absurde qu’il y avait un fi, disons +f1, qui n’était pas de torsion. Alors la liste des morphismes nG ◦ f1 serait +infinie et on pourrait choisir un n tel que nG ◦ f1 n’apparaisse pas dans la +liste des fi. Ceci contredirait l’égalité � nai · fi = � ai · (nG ◦ fi). +On peut alors supposer que les fi soient tous de torsion et on peut donc +choisir un n tel que les nG ◦ fi soient tous nuls. L’égalité � nai · fi = +� ai · (nG ◦ fi) = 0 implique alors � ai · fi = 0. +(2) (Décomposition de Chow–Künneth.) On continue à noter par nG : G → G +le morphisme de multiplication par n. Remarquons que son action sur +M1(G/S) vaut n · Id. En particulier l’isomorphisme (8.3) donne une dé- +composition de M(G/S) en espaces propres par rapport à l’action de nG. +On en déduit par ailleurs que cette décomposition est une décomposition +de Chow–Künneth relative (voir la Conjecture 3.1 pour le cas absolu). +(3) (Décomposition de l’anneau de Chow.) La décomposition du point (2) +donne une décomposition de +HomDM(S)(M(G/S), +1(p)[q]) +(8.4) +en espaces propres par rapport à l’action de nG. Comme ces Hom cal- +culent 19 les groupes de Chow supérieurs [CD19, Corollary 14.2.14] on dé- +duit le Théorème 8.3. +(4) (Décomposition de motifs vs décomposition d’anneaux de Chow.) Pour les +schémas abéliens G = A les décompositions des anneaux de Chow au point +(3) permettent de retrouver celle du motif au point (2) comme l’ont remar- +qué Deninger et Murre. Le point est de considérer A ×S A comme schéma +abélien sur A et d’appliquer le Théorème 8.1 à ce schéma abélien : la dia- +gonale se décomposera alors en somme de vecteurs propres. D’autre part la +19. Les constructions de Grothendieck et de Voevodsky ont une convention de covariance diffé- +rente, notamment les motifs de Chow se plongent dans les motifs de Voevodsky par un foncteur +contravariant. C’est la raison pour laquelle l’objet +1(p)[q] apparaît à droite dans la formule (8.4). + +50 +GIUSEPPE ANCONA +diagonale s’interprète comme l’identité du motif M(A/S) et on peut veri- +fier que cette décomposition de Id ∈ End(M(A/S)) est une décomposition +en somme de projecteurs orthogonaux. +Pour un G général, la formule +EndDM(S)(M(G/S)) = HomDM(S)(M(G/S) ⊗ M(G/S)∨, +1) +ne permet pas de relier ce groupe à l’anneau de Chow de G ×S G. En effet +la dualité de Poincaré identifie, à un twist et shift près, M(G/S)∨ avec +Mc(G/S) qui n’est pas, en général, M(G/S). Dans ce cas la décomposition +des anneaux de Chow du Théorème 8.3 ne permet pas de retrouver celle +des motifs au point (2). +(L’argument ci-dessus montre qu’en général les endomorphismes d’un +motif sont reliés aux groupes de Chow uniquement dans le cas propre et +lisse. Cela a déjà été signalé au §1.1 et c’est le point qui limite la construction +classique de Grothendieck au cadre propre et lisse.) +(5) (Voevodsky vs Chow.) Dans le cas G = A d’un schéma abélien, une formule +M(A/S) ∼= +� +r=0 +Symr M1(A/S) +(8.5) +a été montré par Künnemann encore à l’aide de la transformée de Fourier +[Kün94]. Une version plus faible, valable pour les motifs homologiques, a +été discutée dans la Proposition 4.7. +A l’époque de [Kün94] on ne disposait pas des motifs de Voevodsky et +le travail a été fait dans les motifs de Chow. Dans ce cas l’existence du +motif M1(A/S) présent dans la formule (8.5) n’est pas du tout triviale : +il faut construire un projecteur convenable de End(M(A/S)). Ce motif est +en revanche facile à définir dans le cadre de Voevodsky (Définition 8.5). Un +des avantage des motifs de Voevodsky sur les motifs de Chow est notam- +ment cette possibilité de disposer facilement de motifs par des constructions +faisceautiques : c’est le principe que nous avons mentionné au §1.3. +Démonstration. La preuve se base sur deux dévissages qui font chacun l’objet d’un +article. Un premier dévissage sert à se réduire au cas d’un corps algébriquement clos +S = Spec(K), [AHPL16]. Le deuxième [AEWH15] est une réduction aux cas des +variétés abéliennes, ce qui nous ramène essentiellement au résultat de Künnemann +[Kün94]. +Réduction à S = Spec(K). Pour le premier dévissage on utilise le théorème +suivant d’Ayoub [Ayo14, Proposition 3.24]. Si f : M → N est un morphisme de +motifs dans DM(S) alors pour voir si f est un isomorphisme il suffit de voir si son +tiré en arrière en tout point géométrique l’est. +Soit i : Spec(K) → S un point géométrique. Pour compléter le premier dévissage +il suffit alors de montrer que +i∗ϕG/S = ϕG×SSpec(K)/ Spec(K). +(8.6) +La formule (8.6) est en fait le point technique du travail. Pour tout morphisme +g : T → S, le foncteur g∗ est caractérisé par la propriété que, pour tout S-schéma +lisse X, on ait le changement de base +g∗M(X/S) = M(X ×S T/T ) +(8.7) + +MÉMOIRE HDR +51 +comme pour la cohomologie à support compact. Or parmi les deux motifs qui +interviennent dans le morphisme ϕG/S, seulement M(G/S) est de cette forme. Le +point est alors de trouver une résolution de M1(G) par un complexe dont tous les +termes sont des sommes d’objets de la forme M(X/S) en tous les degrés et toutes +les flèches de connections sont induites par des morphismes de S-schémas. +Nous construisons une telle résolution inspirée par des construction de [Bre70]. +On utilise uniquement des X qui sont des puissances de G. Avec les notations de +la Définition 8.5 on peut écrire son début sous la forme +· · · −→ Q(G × G) +tG +−→ Q(G) +sG +−→ G, +où tG([g1, g2]) = [g1] + [g2] − [g1 + g2]. +Réduction à G = A. Pour le deuxième dévissage fixons un corps algébriquement +clos K et considérerons S = Spec(K). Dans la suite on allégera la notation en +enlevant les /S. +Le théorème de Chevalley nous dit qu’un groupe algébrique sur un corps K +s’insère dans une suite exacte +0 −→ L −→ G −→ A −→ 0, +où A est une variété abélienne et L est un groupe linéaire. On peut alors raisonner +par récurrence sur la dimension de L et se ramener à +0 −→ Ga −→ G −→ H −→ 0 +(8.8) +ou +0 −→ Gm −→ G −→ H −→ 0, +(8.9) +où dans les deux cas H est un groupe pour lequel l’énoncé est connu. Le cas (8.8) +est facile : par homotopie l’énoncé pour G est équivalent à l’énoncé pour H. Pour +le cas (8.9) l’argument est plus délicat. +Premièrement, la suite exacte (8.9) donne une suite exacte au niveau des fonc- +teurs des points. Si on applique Symn a cette dernière on obtient un triangle +Symr−1 M1(H) ⊗ M1(Gm) −→ Symr M1(G) −→ Symr M1(H), +où on a utilisé Sym2 M1(Gm) = 0 [Voe00, Corollary 2.1.5]. Cette suite exacte permet +de déduire par récurrence que Symr M1(G) s’annule pour r assez grand. +Deuxièmement, on complète le Gm-fibré G −→ H en un A1-fibré ¯G −→ H avec +une section nulle. Le triangle de localisation par rapport à la section donne +M(G) −→ M( ¯G) −→ M(H)(1)[2], +or par homotopie M( ¯G) = M(H). On utilise maintenant l’hypothèse de récurrence +et on en déduit le diagramme +M(G) +� +ψ +�✤ +✤ +✤ +✤ +✤ +✤ +M(H) +� +ϕH +� +M(H)(1)[2] +ϕH(1)[2] +� +� +r Symr M1(G) +� � +r Symr M1(H) +� � +r Symr M1(H)(1)[2]. +(8.10) + +52 +GIUSEPPE ANCONA +L’existence de ψ suit de la commutativité du carré de droite. Cette commutativité +n’est pas gratuite, elle utilise la définition de l’application ϕH. +Par hypothèse de récurrence on déduit que l’application ψ est un isomorphisme. +Malheureusement la récurrence n’est pas terminée puisqu’on ne sait pas lier ψ à +ϕG. L’existence de ψ a tout de même une conséquence importante : M(G) est un +motif de dimension finie, voir la Remarque 3.21(2). +On montre que la réalisation de ϕG est un isomorphisme. On peut alors appliquer +la Proposition 4.5 à ϕG et ψ−1 pour déduire que ϕG est un isomorphisme après +passage au quotient par l’équivalence homologique. On peut ensuite utiliser les +propriétés des motifs de dimension finie pour conclure que ϕG est un isomorphisme +même avant passage au quotient (voir la Remarque 3.21(2), cf. le Théorème 5.2 +dans le cas pur). +□ +Remarque 8.8. +(1) La notion de dimension finie dans les motifs a toujours +été appliquée pour les motifs purs : elle ne se comporte pas bien par suite +exacte et on connait des exemples de motifs mixtes qui ne sont pas de +dimension finie. À notre connaissance cette preuve est le premier exemple +d’application de ces idées aux motifs mixtes. +(2) Le Théorème 8.6 a permis à Huber et Kings de construire le polylogarithme +pour tous les schémas en groupes commutatifs. +(3) Le Théorème 8.6 donne une description du motif d’un groupe algébrique +commutatif G en terme d’un motif assez simple, M1(G). On pourrait espérer +que cela puisse aider à une meilleure compréhension des anneaux de Chow +de G. +9. Construction de classes algébriques +Cette section concerne la conjecture standard de type Lefschetz, introduite dans +la Conjecture 3.11. Dans un travail en cours en collaboration avec Mattia Cavicchi, +Robert Laterveer et Giulia Saccà, nous étudions cette conjecture pour les variétés +hyper-kähler qui admettent une fibration lagrangienne. Le point de départ est une +idée récente de Voisin [Voi22] que l’on regarde dans la perspective du théorème de +décomposition. +Nous nous concentrerons dans la suite sur les variétés complexes, la conjecture +standard de type Lefschetz est alors une instance particulière de la conjecture de +Hodge. De façon assez surprenante elle en est aussi le pilier principal : André dé- +montre que sous la conjecture standard de type Lefschetz le transport parallèle de +classes algébriques est encore algébrique [And96]. Il en déduit que cette conjecture +impliquerait la conjecture de Hodge pour les variétés abéliennes. +La conjecture standard de type Lefschetz est connue pour les variétés abéliennes, +voir la Proposition 4.7, et on sait en déduire le cas des surfaces. Plus récemment +Charles et Markmann l’ont montrée pour les variétés hyper-kähler de type KS[n], +[CM13]. +La dimension d’une variété X sera notée dX et la dimension de la fibre générique +d’un morphisme f sera notée df. Même en présence de faisceaux pervers on utilisera +la convention classique pour les degrés cohomologiques. +9.1. Une approche naïve. Considérons une variété projective et lisse X et sup- +posons qu’elle admette un morphisme f : X → B vers une base B qui est aussi + +MÉMOIRE HDR +53 +projective et lisse. Nous nous demandons jusqu’à quel point connaître la conjecture +standard de type Lefschetz pour B et pour les fibres lisses de f peut aider pour +montrer la conjecture standard de type Lefschetz pour X. +Le théorème de décomposition implique en particulier une décomposition 20 de +structures de Hodge +Hn(X) = +� +a+b=n +Ha(B,pRbf∗Q). +(9.1) +Soient η un diviseur sur X qui soit relativement ample, LB un diviseur ample sur +B et β son tiré en arrière sur X. Le théorème de décomposition fournit aussi les +isomorphismes +∪ηb : Ha(B,pRdf−bf∗Q) +∼ +−−→ Ha(B,pRdf+bf∗Q), +(9.2) +∪βa : HdB−a(B,pRbf∗Q) +∼ +−−→ HdB+a(B,pRbf∗Q). +(9.3) +La combination de (9.1), (9.2) et (9.3) peut suggérer que la conjecture standard +de type Lefschetz se ramène à montrer que les inverses de ces isomorphismes sont +algébriques et à première vue on pourrait penser que ces derniers se ramènent +uniquement à l’étude des fibres de f et de la base B. Mais il faut en fait faire +attention à un certain nombre de subtilités. +(1) Les faisceaux pervers pRbf∗Q dépendent aussi des fibres singulières du mor- +phisme f. +(2) Inverser l’action de LB sur la cohomologie de B ne suffit pas à inverser β. +En effet f ∗ H(B) ⊂ H(X) ne représente que le facteur H∗(B,pR0f∗Q) de la +décomposition (9.1). +(3) Même si on était capables de construire des correspondances algébriques Ληb +et Λβa qui agiraient comme les inverses de (9.2) et (9.3) il n’est pas clair de +pouvoir les mettre ensemble pour déduire une correspondance algébrique +Λn : HdX+n(X) +∼ +−−→ HdX−n(X). Il faudrait par exemple contrôler comment +Ληb agit sur les degrés pervers différents de b. C’est délicat, notamment +parce que les opérateurs η et β ne sont pas bigradués en général. +Dans le cas des variétés hyper-kähler qui admettent une fibration lagrangienne on +peut espérer contourner ces problèmes : le théorème du support de Ngô donne une +description explicite des faisceaux pervers pRbf∗Q, un argument de Voisin permet +grosso-modo de construire une deuxième fibration lagrangienne qui inverse le rôle +de η et β, enfin Shen et Yin ont montré que les opérateurs η et β sont en fait +bigradués pour les fibrations lagrangiennes. +Proposition 9.1. Soit f : X → B une application entre variétés projectives, lisses +et connexes et soit U l’ouvert de B sur lequel l’application est lisse. +(1) Supposons que la fibre générique vérifie la conjecture standard de type Lef- +schetz et que +pRbf∗Q = IC((Rbf∗Q)|U). +(9.4) +20. Une telle décomposition n’est pas unique en général, seulement la filtration perverse associée +l’est. Deligne, puis De Cataldo, ont proposé des décompositions qui se comportent mieux que les +autres [Del94b, dC13]. + +54 +GIUSEPPE ANCONA +Alors il existe des correspondances dans X ×B X dont l’action sur la coho- +mologie relative induit des isomorphismes pRdf+bf∗Q +∼ +−−→ pRdf−bf∗Q. +(2) Supposons qu’il existe des correspondances dans X×BX dont l’action sur la +cohomologie relative induit des isomorphismes pRdf+bf∗Q +∼ +−−→ pRdf−bf∗Q. +Alors il existe une décomposition du motif +h(X) = +� +b +h(B,pRbf∗Q) +(9.5) +où R(h(B,pRbf∗Q)) = H∗(B,pRbf∗Q). De plus chaque facteur de la décom- +position est autodual à un twist près. +Remarque 9.2. (Autour de la preuve.) La preuve de (1) est facile. Soit Y la fibre +générique de f. L’hypothèse sur Y dans (1) fournit des correspondances dans Y ×Y , +dont les adhérences donnent des correspondances dans X ×B X. Leur action sur +la cohomologie relative IC((Rbf∗Q)|U) est contrôlée par leur action sur le système +local (Rbf∗Q)|U et donc par l’action sur la cohomologie de Y . +Pour (2) on suit l’argument classique qui montre que Lefschetz implique Kün- +neth [Kle68]. Par rapport au cas absolu on prendra garde au fait que la décom- +position (9.5) n’est pas unique. L’action des correspondances relatives ne respecte +pas cette graduation mais seulement la filtration associée. Par ailleurs on ne sait +pas démontrer que toute décomposition cohomologique (9.1) est la réalisation d’une +décomposition motivique (9.5) mais seulement qu’il en existe au moins une qui est +d’origine motivique. +Corollaire 9.3. Soit f : X → P1 une fibration de Lefschetz dont la fibre générique +vérifie la conjecture standard de type Lefschetz (par exemple f est une fibration en +surfaces). Alors il existe une décomposition du motif +h(X) = +� +b +h(P1,pRbf∗Q) +(9.6) +où R(h(P1,pRbf∗Q)) = H∗(P1,pRbf∗Q). De plus chaque facteur de la décomposition +est autodual à un twist près. +Remarque 9.4. (Du relatif à l’absolu.) Toute variété de dimension trois admet +une fibration de Lefschetz après éclatement le long d’une courbe C. D’autre part +la formule de l’éclatement +h(BlC(X)) = h(X) ⊕ h(C)(−1), +due à Manin [Man68], montre que la conjecture standard de type Lefschetz pour +X se ramène à celle pour BlC(X). Si l’on veut étudier la conjecture standard de +type Lefschetz pour une variété X de dimension trois on peut donc supposer que X +admet une telle fibration. Ce qui manque au corollaire ci-dessus pour déduire cette +conjecture est l’existence d’une décomposition +h(P1,pRbf∗Q) = +2 +� +a=0 +ha(P1,pRbf∗Q), +où R(ha(P1,pRbf∗Q)) = Ha(P1,pRbf∗Q), telle que chaque facteur de la décomposi- +tion soit autodual à un twist près. + +MÉMOIRE HDR +55 +Pour construire cette décomposition on pourrait vouloir utiliser l’isomorphisme +(9.3) et essayer de construire une correspondance algébrique qui induise l’inverse. +C’est une question qui ressemble au problème original de construction de l’inverse +de l’opérateur de Lefschetz. On ne sait pas si c’est un problème plus simple, voir +aussi la Remarque 9.8. +Théorème 9.5. Soit X une variété hyper-kähler de dimension 2n et f : X → Pn +une fibration lagrangienne dont toutes les fibres sont irréductibles. Supposons que +le schéma en groupes Aut0(f) soit polarisable (au sens de Ngô). Alors +h(X) = +2n +� +b=0 +h(Pn,pRbf∗Q) +(9.7) +où R(h(Pn,pRbf∗Q)) = H∗(Pn,pRbf∗Q). De plus chaque facteur de la décomposition +est autodual à un twist près. +Démonstration. On utilise le théorème du support de Ngô, [Ngô10] : quand les +fibres sont toutes irréductibles on a bien l’hypothèse (9.4). D’autre part la fibre +générique est une variété abélienne, donc elle vérifie la conjecture standard de type +Lefschetz. On peut alors utiliser la Proposition 9.1 pour conclure. +□ +Corollaire 9.6. Soit X une variété hyper-kähler de dimension 2n de rang de Pi- +card 2 qui admet une fibration lagrangienne. Supposons que pour toute fibration +lagrangienne g de n’importe quel variétés hyper-kähler birationnelle à X, les fibres +de g soient irréductibles et le schéma en groupes Aut0(g) soit polarisable. Alors la +conjecture standard de type Lefschetz est vraie pour X. +Démonstration. Fixons f : X → Pn une fibration lagrangienne et soit η et β les +diviseurs introduits dans §9.1. Un argument de Voisin [Voi22] montre grosso-modo +l’existence d’une fibration lagrangienne g : X → Pn où le rôle de η et β est inversé. +En fait g existe seulement sur une variété hyper-kähler qui est birationnelle à X, +mais elles ont le même motif [Rie16]. +On peut alors appliquer le théorème ci-dessus à f et g pour déduire deux décom- +positions du motif h(X) dont tous les facteurs sont autoduaux. Il n’est pas clair que +ces deux décompositions soient compatibles, voir aussi le point (3) dans §9.1. Mais +un théorème de Shen et Yin montre que les opérateurs η et β sont bigradués pour +les fibrations lagrangiennes : on peut voir que ceci force les deux décompositions +à être compatibles. La bidécomposition que l’on en déduit est plus fine que (9.1) +et a tous les facteurs autoduaux, ce qui implique la conjecture standard de type +Lefschetz. +□ +Le corollaire ci-dessus s’applique par exemple aux variétés hyper-kähler construites +par Laza–Saccà–Voisin [LSV17]. +Remarque 9.7. (Généralisations.) On souhaite se débarrasser de l’hypothèse d’ir- +réductibilité des fibres dans le théorème ci-dessus, ce qui permettrait de l’enlever +aussi dans son corollaire et de pouvoir l’appliquer à beaucoup plus de variétés. +Si les fibres ne sont pas irréductible les faisceaux pervers pRbf∗Q ne vérifient plus +(9.4). Cependant le théorème du support de Ngô décrit aussi la nature des faisceaux +pervers supportés sur les sous-variétés strictes de la base. Il montre l’existence + +56 +GIUSEPPE ANCONA +de certaines variétés abéliennes contenues dans certaines fibres singulières dont la +cohomologie contrôle les faisceaux pervers. On pense qu’une variante stratifiée de +la Proposition 9.1 devrait s’appliquer à ce contexte général. On utilisera encore que +les variétés abéliennes vérifient la conjecture standard de type Lefschetz. +Remarque 9.8. (Du relatif à l’absolu, suite.) Reprenons les notations de la Re- +marque 9.4. On souhaiterait montrer l’existence d’une décomposition +h(P1,pRbf∗Q) = +2 +� +a=0 +ha(P1,pRbf∗Q), +(9.8) +où R(ha(P1,pRbf∗Q)) = Ha(P1,pRbf∗Q), telle que chaque facteur de la décomposi- +tion soit autodual à un twist près : ceci montrerait la conjecture standard de type +Lefschetz pour les variétés de dimension trois. +Pour construire cette décomposition on pourrait vouloir utiliser l’isomorphisme +(9.3) induit par la classe β et essayer de construire une correspondance algébrique +qui induise l’inverse. Inspirés par l’idée de Voisin esquissé dans la preuve du co- +rollaire ci-dessus on peut essayer de construire une nouvelle fibration g : X → P2 +telle que β soit relativement ample (ces constructions sont toujours possibles après +éclatement). Une telle fibration induit une décomposition +h(X) = +2 +� +a=0 +h(P2,pRag∗Q) +(9.9) +où ∪β : h(P2,pR0g∗Q) +∼ +−−→ h(P2,pR2g∗Q) admet un inverse algébrique. Le problème +est qu’en général la décomposition (9.9) n’induit pas la décomposition (9.8). C’est +encore lié au fait que les actions de η et β ne sont pas bigraduées en général. + +MÉMOIRE HDR +57 +Références +[ACLS22] +Giuseppe Ancona, Mattia Cavicchi, Robert Laterveer, and Giulia Saccà. Standard +conjectures for some lagrangian fibrations. prépublication, 2022. +[AEWH15] Giuseppe Ancona, Stephen Enright-Ward, and Annette Huber. On the motive of a +commutative algebraic group. Documenta Math., 20 :807–858, 2015. +[AF22] +Giuseppe Ancona and Dragos Fratila. Algebraic classes in mixed characteristic and +André’s p-adic periods. prépublication, 2022. +[AHPL16] +Giuseppe Ancona, Annette Huber, and Simon Pepin Lehalleur. On the relative motive +of a commutative group scheme. Algebr. Geom., 3(2) :150–178, 2016. +[AM22] +Giuseppe Ancona and Adriano Marmora. The Hilbert symbol in the Hodge standard +conjecture. prépublication, 2022. +[Anc21] +Giuseppe +Ancona. +Standard +conjectures +for +abelian +fourfolds. +Invent. +Math., +223(1) :149–212, 2021. +[Anc22] +Giuseppe Ancona. Conservativity of realization functors on motives of abelian type +over finite fields. Proceedings of "Motives and Complex Multiplication", Fresán, Jet- +chev, Jossen and Pink editors, 2022. +[And90] +Yves André. p-adic Betti lattices. In p-adic analysis (Trento, 1989), volume 1454 of +Lecture Notes in Math., pages 23–63. Springer, Berlin, 1990. +[And95] +Yves André. Théorie des motifs et interprétation géométrique des valeurs p-adiques +de G-functions (une introduction). In Number theory (Paris, 1992–1993), volume +215 of London Math. Soc. Lecture Note Ser., pages 37–60. Cambridge Univ. Press, +Cambridge, 1995. +[And96] +Yves André. Pour une théorie inconditionnelle des motifs. Inst. Hautes Études Sci. +Publ. Math., (83) :5–49, 1996. +[And03] +Yves André. Period mappings and differential equations. From C to Cp. Tôhoku- +Hokkaidô lectures in arithmetic geometry. With appendices : A : Rapid course in +p-adic analysis by F. Kato, B : An overview of the theory of p-adic unifomization +by F. Kato, C : p-adic symmetric domains and Totaro’s theorem by N. Tsuzuki, +volume 12. Tokyo : Mathematical Society of Japan, 2003. +[And04] +Yves André. Une introduction aux motifs (motifs purs, motifs mixtes, périodes), vo- +lume 17 of Panoramas et Synthèses [Panoramas and Syntheses]. Société Mathéma- +tique de France, Paris, 2004. +[Ayo07] +Joseph Ayoub. Les six opérations de Grothendieck et le formalisme des cycles évanes- +cents dans le monde motivique. I. Astérisque, (314) :x+466 pp. (2008), 2007. +[Ayo10] +Joseph Ayoub. Note sur les opérations de Grothendieck et la réalisation de Betti. J. +Inst. Math. Jussieu, 9(2) :225–263, 2010. +[Ayo14] +Joseph Ayoub. La réalisation étale et les opérations de Grothendieck. Ann. Sci. École +Norm. Sup., 47 :1–141, 2014. +[Ayo15] +Joseph Ayoub. Une version relative de la conjecture des périodes de Kontsevich-Zagier. +Ann. of Math. (2), 181(3) :905–992, 2015. +[BC16] +Jean-Benoît Bost and François Charles. Some remarks concerning the Grothendieck +period conjecture. J. Reine Angew. Math., 714 :175–208, 2016. +[Bea86] +Arnaud Beauville. Sur l’anneau de Chow d’une variété abélienne. Math. Ann., +273(4) :647–651, 1986. +[Ber77] +Daniel Bertrand. Sous-groupes à un paramètre p-adique de variétés de groupe. Invent. +Math., 40(2) :171–193, 1977. +[Bon10] +Mikhail V. Bondarko. Weight structures vs. t-structures ; weight filtrations, spectral +sequences, and complexes (for motives and in general). J. K-Theory, 6(3) :387–504, +2010. +[Bre70] +Lawrence Breen. Extensions of abelian sheaves and Eilenberg-MacLane algebras. +Invent. Math., 9 :15–44, 1969/1970. +[Bro12] +Francis Brown. Mixed Tate motives over Z. Ann. of Math. (2), 175(2) :949–976, 2012. + +58 +GIUSEPPE ANCONA +[Bro17] +Francis +Brown. +Notes +on +motivic +periods. +Commun. +Number +Theory +Phys., +11(3) :557–655, 2017. +[CD19] +Denis-Charles Cisinski and Frédéric Déglise. Triangulated categories of mixed motives. +Springer Monographs in Mathematics. Springer, Cham, 2019. +[CDN22] +Mattia Cavicchi, Frédéric Déglise, and Jan Nagel. Motivic decompositions of families +with tate fibers : smooth and singular cases. prépublication, 3(2), 2022. +[CF00] +Pierre Colmez and Jean-Marc Fontaine. Construction des représentations p-adiques +semi-stables. Invent. Math., 140(1) :1–43, 2000. +[Cha13] +François Charles. The Tate conjecture for K3 surfaces over finite fields. Invent. Math., +194(1) :119–145, 2013. +[Chu80] +G. V. Chudnovsky. Algebraic independence of values of exponential and elliptic func- +tions. In Proceedings of the International Congress of Mathematicians (Helsinki, +1978), pages 339–350. Acad. Sci. Fennica, Helsinki, 1980. +[Clo99] +Laurent Clozel. Equivalence numérique et équivalence cohomologique pour les variétés +abéliennes sur les corps finis. Ann. of Math. (2), 150(1) :151–163, 1999. +[CM13] +François Charles and Eyal Markman. The standard conjectures for holomorphic sym- +plectic varieties deformation equivalent to Hilbert schemes of K3 surfaces. Compos. +Math., 149(3) :481–494, 2013. +[Col90] +Robert F. Coleman. On the Frobenius matrices of Fermat curves. In p-adic analysis +(Trento, 1989), volume 1454 of Lecture Notes in Math., pages 173–193. Springer, +Berlin, 1990. +[dC13] +Mark Andrea A. de Cataldo. Hodge-theoretic splitting mechanisms for projective +maps. J. Singul., 7 :134–156, 2013. With an appendix containing a letter from P. +Deligne. +[Del89] +P. Deligne. Le groupe fondamental de la droite projective moins trois points. In Galois +groups over Q (Berkeley, CA, 1987), volume 16 of Math. Sci. Res. Inst. Publ., pages +79–297. Springer, New York, 1989. +[Del94a] +Pierre Deligne. à quoi servent les motifs ? In Motives (Seattle, WA, 1991), volume 55 +of Proc. Sympos. Pure Math., pages 143–161. Amer. Math. Soc., Providence, RI, 1994. +[Del94b] +Pierre Deligne. Décompositions dans la catégorie dérivée. In Motives (Seattle, WA, +1991), volume 55 of Proc. Sympos. Pure Math., pages 115–128. Amer. Math. Soc., +Providence, RI, 1994. +[Del13] +Pierre Deligne. Multizêtas, d’après Francis Brown. Number 352, pages Exp. No. 1048, +viii, 161–185. 2013. Séminaire Bourbaki. Vol. 2011/2012. Exposés 1043–1058. +[DM91] +Christopher Deninger and Jacob Murre. Motivic decomposition of abelian schemes +and the Fourier transform. J. Reine Angew. Math., 422 :201–219, 1991. +[Fal89] +Gerd Faltings. Crystalline cohomology and p-adic Galois-representations. In Algebraic +analysis, geometry, and number theory (Baltimore, MD, 1988), pages 25–80. Johns +Hopkins Univ. Press, Baltimore, MD, 1989. +[Fon94] +Jean-Marc Fontaine. Le corps des périodes p-adiques. Astérisque, (223) :59–111, 1994. +With an appendix by Pierre Colmez, Périodes p-adiques (Bures-sur-Yvette, 1988). +[Fur07] +Hidekazu Furusho. p-adic multiple zeta values. II. Tannakian interpretations. Amer. +J. Math., 129(4) :1105–1144, 2007. +[Jan07] +Uwe Jannsen. On finite-dimensional motives and Murre’s conjecture. In Algebraic +cycles and motives. Vol. 2, volume 344 of London Math. Soc. Lecture Note Ser., +pages 112–142. Cambridge Univ. Press, Cambridge, 2007. +[Kah03] +Bruno Kahn. Équivalences rationnelle et numérique sur certaines variétés de type +abélien sur un corps fini. Ann. Sci. École Norm. Sup. (4), 36(6) :977–1002 (2004), +2003. +[Kat80] +Nicholas M. Katz. p-adic L-functions, Serre-Tate local moduli, and ratios of solutions +of differential equations. In Proceedings of the International Congress of Mathemati- +cians (Helsinki, 1978), pages 365–371. Acad. Sci. Fennica, Helsinki, 1980. + +MÉMOIRE HDR +59 +[Kim05] +Shun-Ichi Kimura. Chow groups are finite dimensional, in some sense. Math. Ann., +331(1) :173–201, 2005. +[Kle68] +Steven Kleiman. Algebraic cycles and the Weil conjectures. In Dix esposés sur la +cohomologie des schémas, pages 359–386. North-Holland, Amsterdam, 1968. +[KM74] +Nicholas M. Katz and William Messing. Some consequences of the Riemann hypothesis +for varieties over finite fields. Invent. Math., 23 :73–77, 1974. +[Kos22] +Teruhisa Koshikawa. The numerical Hodge standard conjecture for the square of a +simple abelian variety of prime dimension. prépublication, 3(2), 2022. +[Kün94] +Klaus Künnemann. On the Chow motive of an abelian scheme. In Motives, volume +55.1 of Proceedings of Symposia in pure mathematics, pages 189–205. American ma- +thematical society, 1994. +[LSV17] +Radu Laza, Giulia Saccà, and Claire Voisin. A hyper-Kähler compactification of the +intermediate Jacobian fibration associated with a cubic 4-fold. Acta Math., 218(1) :55– +135, 2017. +[Man68] +Ju. I. Manin. Correspondences, motifs and monoidal transformations. Mat. Sb. (N.S.), +77 (119) :475–507, 1968. +[Mil07] +J. Milne. The tate conjecture over finite fields. pages 1–27. 2007. Available on the +author’s webpage. +[Mum68] +D. Mumford. Rational equivalence of 0-cycles on surfaces. J. Math. Kyoto Univ., +9 :195–204, 1968. +[Ngô10] +Bao Châu Ngô. Le lemme fondamental pour les algèbres de Lie. Publ. Math. Inst. +Hautes Études Sci., (111) :1–169, 2010. +[Ogu90] +Arthur Ogus. A p-adic analogue of the Chowla-Selberg formula. In p-adic analysis +(Trento, 1989), volume 1454 of Lecture Notes in Math., pages 319–341. Springer, +Berlin, 1990. +[O’S05] +Peter O’Sullivan. The structure of certain rigid tensor categories. C. R. Math. Acad. +Sci. Paris, 340(8) :557–562, 2005. +[Rie16] +Ulrike Rieß. On Beauville’s conjectural weak splitting property. Int. Math. Res. Not. +IMRN, (20) :6133–6150, 2016. +[Sch15] +Stefan Schreieder. On the construction problem for Hodge numbers. Geom. Topol., +19(1) :295–342, 2015. +[Ser72] +Jean-Pierre Serre. Propriétés galoisiennes des points d’ordre fini des courbes ellip- +tiques. Invent. Math., 15(4) :259–331, 1972. +[SK79] +Tetsuji Shioda and Toshiyuki Katsura. On Fermat varieties. Tôhoku Math. J. (2), +31(1) :97–115, 1979. +[Tat66] +John Tate. Endomorphisms of abelian varieties over finite fields. Invent. Math., 2 :134– +144, 1966. +[Voe00] +Vladimir Voevodsky. Triangulated categories of motives over a field. In Cycles, trans- +fers, and motivic homology theories, volume 143 of Ann. of Math. Stud., pages 188– +238. Princeton Univ. Press, Princeton, NJ, 2000. +[Voi22] +Claire Voisin. On the Lefschetz standard conjecture for Lagrangian covered hyper- +Kähler varieties. Adv. Math., 396 :Paper No. 108108, 29, 2022. +[Wil17] +Jörg Wildeshaus. Intermediate extension of Chow motives of Abelian type. Adv. +Math., 305 :515–600, 2017. +[Wüs89] +G. Wüstholz. Algebraische Punkte auf analytischen Untergruppen algebraischer Grup- +pen. Ann. of Math. (2), 129(3) :501–517, 1989. +IRMA, Université de Strasbourg 7, rue René Descartes, 67000 Strasbourg +Email address: ancona@math.unistra.fr +URL: http://irma.math.unistra.fr/~ancona/ + diff --git a/99E0T4oBgHgl3EQfgAA0/content/tmp_files/load_file.txt b/99E0T4oBgHgl3EQfgAA0/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..5055f41eec88bbc548259f0f2cfa2855d97847a1 --- /dev/null +++ b/99E0T4oBgHgl3EQfgAA0/content/tmp_files/load_file.txt @@ -0,0 +1,2146 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf,len=2145 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='02411v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='AG] 6 Jan 2023 QUELQUES ASPECTS ARITHMÉTIQUES ET GÉOMÉTRIQUES DES CYCLES ALGÉBRIQUES ET DES MOTIFS GIUSEPPE ANCONA Habilitation à diriger les recherches Soutenue le 17 Novembre 2022 devant le jury composé de : Yves André, examinateur Joseph Ayoub, rapporteur Jean-Benoît Bost, examinateur Anna Cadoret, rapporteur François Charles, rapporteur Carlo Gasbarri, examinateur Rutger Noot, garant Table des matières Remerciements 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Introduction 7 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A quoi servent les motifs ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 10 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cohomologies de Weil 12 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjectures standard et moins standard 18 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemples 23 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Autodualité et conservativité 29 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Positivité en caracteristique positive 32 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Périodes p-adiques à la André 39 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs des schémas en groupes commutatifs 46 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Construction de classes algébriques 52 Références 57 1 MÉMOIRE HDR 3 Remerciements On m’avait souvent décrit l’Habilitation comme une formalité par laquelle il fallait passer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je me rends compte qu’en fait d’une part j’ai pris plaisir à rédiger ce mémoire et d’autre part c’est avec un peu d’émotion que je m’approche à cette soutenance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je serai entouré par des collègues et des amis qui me sont chers et j’aimerais par ces quelques lignes les remercier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Merci aux trois rapporteurs d’avoir accepté de consacrer leur temps à la lecture de mon mémoire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Joseph Ayoub a été mon encadrant de postdoc et c’est une figure centrale de la théorie des motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’ai bénéficié de nombreuses discussions avec lui et c’est un grand honneur pour moi l’attention qu’il a porté à mes travaux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’ai rencontré Anna Cadoret pendant ma thèse à l’occasion d’un groupe de travail organisé par son ANR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’ai le souvenir d’une atmosphère détendue et ouverte vers les jeunes, comme il arrive rarement dans notre métier (atmosphère que l’on a essayé de reproduire avec la RéGA, puis le JAVA).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’ai toujours aimé l’originalité des questions qu’elle sait se poser et je me réjouis de ses appréciations à ce mémoire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' François Charles était 2 ans avant moi à l’ENS, il a représenté pour plusieurs d’entre nous une référence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’ai toujours été impressionné par ses fortes intuitions et je le remercie pour les échanges que l’on a eu et pour son intérêt pour mes résultats.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Rutger Noot a été rapporteur de ma thèse et m’a beaucoup encouragé après : je lui en suis reconnaissant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Depuis que je suis à Strasbourg, j’ai pu profiter de sa culture mathématique à plusieurs reprises.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’était pour moi le choix naturel de garant et je le remercie d’avoir accepté.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il n’y a probablement aucune section de ce mémoire où le nom d’Yves André n’apparaisse pas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Bien que j’essaie constamment d’élargir mes intérêts et m’ouvrir à de nouveaux domaines, je finis toujours par y découvrir un théorème fondamental qui lui est dû.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sa présence dans mon jury est un honneur pour moi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’ai suivi un cours de Jean-Benoît Bost en M2 : sa clarté et sa vision sont devenues une référence pour moi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je le remercie pour son soutien depuis cette époque ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' c’est un plaisir pour moi de l’avoir dans mon jury d’Habilitation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les déjeuners à la cantine sont souvent enrichis par les digressions mathématiques (toujours passionnées) de Carlo Gasbarri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je le remercie pour celles que l’on a eu, celles que l’on aura et d’avoir accepté de faire partie du jury.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je remercie Emiliano Ambrosi, Pierre Baumann, Mattia Cavicchi, Frédéric Cha- poton, Dragos Fratila, Florence Lecomte et Rutger Noot pour leurs relectures aux premières versions de ce texte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Certains des résultats présentés ici sont issus de collaborations avec des ma- thématiciens qui m’ont beaucoup appris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je les remercie tous ici, à partir de ma première collaboratrice, Annette Huber, avec laquelle nous continuons à beaucoup échanger, jusqu’au plus jeune, Mattia Cavicchi, mathématicien doué à qui je sou- haite tout le meilleur pour ses débuts de carrière.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mes recherches mathématiques n’auraient pas pu avoir lieu sans les échanges, les conseils et les remarques de mes chers amis mathématiciens Olivier Benoist, Yohan Brunebarbe, Javier Fresán et Marco Maculan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 4 GIUSEPPE ANCONA Quand je repense à mon passé mathématique je dois remercier mon directeur de thèse Jörg Wildeshaus, ainsi que Frédéric Déglise qui était chercheur à Paris 13 à l’époque de ma thèse et qui m’a beaucoup appris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pendant ces années à l’IRMA j’ai eu le plaisir de discuter de mathématiques avec de nombreux collègues, tant dans mon équipe que dans les autres.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je pense en particulier à Emilano Ambrosi, Dragos Fratila, Robert Laterveer, Yohann Le Floch, Adriano Marmora, Pierre Py et Ana Rechtman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’ambiance au laboratoire est bonne, j’ai échangé avec à peu près tout le monde à l’occasion d’enseignements, conseils ou autre et j’ai souvent trouvé de l’empathie et de la bienveillance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Chers collègues qui lisez ces lignes, je vous en remercie !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le travail d’enseignant-chercheur a besoin d’un soutien administratif et technique considérable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous sommes gâté de ce point de vue-là à Strasbourg : je remercie en particulier Alexandra Carminati, Sandrine Cerdan, Pascale Igot, Delphine Karles- kind, Jessica Maurer-Spoerk, Alexis Palaticky et Alain Sartout.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je suis ému par la présence de plusieurs amis et collègues qui n’habitent pas à Strasbourg et qui ont fait de la route pour venir voir ma soutenance : Javier, Mattia, Olivier, Quentin et Yohan, merci !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je suis également touché par la présence d’amis qui ont décidé de passer une heure de leur vie à me voir délirer devant des tableaux bleus : Joanne, Cédric, François.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' cela me touche beaucoup.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Enfin, merci à tout ce qui est non-mathématique dans ma vie et qui la rend si heureuse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je pense aux amis du tango de la belle association Hermosa : merci pour la bonne ambiance des milongas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je pense aux amis du foot du terrain 5 : merci pour les matchs intenses et pour les après-matchs encore plus intenses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je pense aux jogging avec Gianluca (accompagnés toujours de questionnements et conseils réciproques).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je pense aux jeunes parents avec lesquels on partage les apéros où l’on ne peut jamais terminer nos phrases ainsi que les fêtes regrettées le lendemain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Et évidemment je pense à ma famille pétillante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Merci à toutes les générations de la Nonna Michelina qui me chantait l’opéra pendant qu’elle préparait ses pâtes jusqu’à mes filles qui sautent toujours à mon cou.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Merci à Anna qui a toujours envie de me voir donner un exposé et toujours la joie de le célébrer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Merci à ma mère qui nous a préparé un super pot (vous verrez.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='), à mon père qui me parlait de sciences en voiture pendant qu’il m’accompagnait au judo, à mes frères et à notre chambre à trois où on pratiquait tout sport possible, à la famille de Maglie qui nous a toujours accompagné avec amour et à ma belle famille avec qui on partage voyages, bonnes bouffes et discussions sur le monde.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A Anna, compagna di viaggio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A Camilla e Gaia, il nostro viaggio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Introduction Le protagoniste de ce mémoire est un morphisme d’anneaux gradués 1 appelé application classe de cycle clX : CH(X) −→ H(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ici X est une variété projective et lisse définie sur un corps k, H est une cohomologie de Weil 2 et CH(X) est l’anneau de Chow à coefficients rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il faut penser à CH(X) comme un invariant de X de nature algébrique et à H(X) comme un invariant de nature topologique : l’application clX compare ces différentes natures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les questions autour de clX peuvent se diviser grossièrement en trois classes : (1) Décrire l’image de clX, (2) Décrire le noyau de clX, (3) Décrire la structure multiplicative de clX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple, des conjectures qui entrent dans la première classe sont Hodge, Tate, la conjecture des périodes de Grothendieck, ou encore les conjectures stan- dard de type Künneth et de type Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la deuxième classe on trouve la conjecture de Bloch–Beilinson ou la conjecture de nilpotence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Certaines conjectures sont à cheval entre la première et la deuxième classe, par exemple la conjecture de conservativité ou la conjecture sur la dimension finie de Kimura.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Rentrent dans la troisième classe les conjectures standard de type Hodge et de type « hom = num ».' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour décomposer l’étude dans ces trois classes on peut d’abord factoriser clX (comme morphisme d’anneaux) et obtenir le diagramme suivant : CH(X) p �� clX �❖ ❖ ❖ ❖ ❖ ❖ ❖ ❖ ❖ ❖ ❖ CH(X)/ ker clX � � i � H(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’anneau quotient CH(X)/ ker clX est également noté CH(X)/ hom et appelé an- neaux des cycles modulo l’équivalence homologique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les questions appartenant à la classe (1) ci-dessus reviennent alors à l’étude de l’injection i et celles de la classe (2) à la projection p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour exprimer la classe (3) on considère l’équivalence numérique 3 et on complète le diagramme ainsi : 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La graduation dans CH(X) est induite par la codimension des sous-variétés et la multipli- cation est appelée produit d’intersection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple si k = C on pourra choisir la cohomologie singulière ou pour k de caractéristique p ≥ 0 et ℓ un nombre premier tel que ℓ ̸= p on pourra choisir la cohomologie ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette équivalence rend tous les points de la variété équivalents et les cycles de codimension complémentaire sont mis en dualité par le produit d’intersection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 8 GIUSEPPE ANCONA CH(X) p �� clX �▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ CH(X)/ hom� � i � q �� H(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' CH(X)/ num (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) La compréhension de l’algèbre CH(X)/ num et de la projection q encode une bonne partie des questions de la classe (3) ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les conjectures principales sur les cycles algébriques s’expriment mieux si on passe aux motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela revient, grosso-modo, à considérer le diagramme (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) pour toutes les variétés X à la fois : CHM(k) π �� R �▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ Mot(k)� � I � π′ �� GrVect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' NUM(k) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) Ici R indique la réalisation des motifs de Chow vers les espaces vectoriels gra- dués : c’est une collection d’applications classe de cycle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On a de plus des foncteurs pleins π, π′ de projection vers les motifs homologiques et les motifs numériques et un foncteur fidèle I des motifs homologiques vers les espaces vectoriels gradués, également appelé réalisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le théorie des motifs est utile non seulement à la formulation de questions sur les cycles algébriques mais aussi à la démonstration de résultats sur ces derniers, voir la Section 1 pour une discussion de ce point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En guise d’exemple, voici une petite liste de résultats - certains issus de mes travaux - où l’on remarquera que les motifs n’apparaissent pas dans les énoncés et pourtant sont cruciaux dans leurs preuves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Kahn [Kah03]) L’application classe de cycle est injective pour les produits de courbes elliptiques sur un corps fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Kimura [Kim05]) Soit S un surface complexe projective et lisse dominée par un produit de courbes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si l’application classe de cycle est surjective alors elle est injective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) [Anc21] Soit A une variété abélienne de dimension quatre, alors le produit d’intersection CH2(A)/num × CH2(A)/num −→ Q MÉMOIRE HDR 9 est de signature (ρ2 − ρ1 + 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ρ1 − 1), où ρn = dimQ(CHn(A)/num).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (4) [AHPL16] Soient S un schéma de base régulier et G un S-schéma en groupes commutatifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors l’action du morphisme nG : G −→ G de multiplication par n décompose l’espace CH(G) en une somme finie de sous-espaces propres (de plus les valeurs propres sont des puissances expli- cites de n et la décomposition ne dépend pas de l’entier n ≥ 2 choisi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par sa nature même, la théorie des motifs se mélange aux différentes cohomo- logies de Weil que l’on peut considérer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ainsi, en fonction du corps de base k, on se retrouve à utiliser la théorie de Hodge (k = C), les représentations galoisiennes (k corps de nombres), l’arithmétique des nombres de Weil (k fini) ou encore de la théorie de Hodge p-adique (en caractéristique mixte).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela dégage les aspects arithmétiques et géométriques de la théorie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces différentes théories cohomologiques ont des analogies, que l’on retrouve dans les motifs par des théorèmes ou des conjectures, ainsi que des différences, que la théorie des motifs vise à réparer, voir la Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut distinguer les différents résultats dans le domaine des motifs d’une part par la partie du diagramme 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 que l’on étudie et d’autre part par la nature géométrique ou arithmétique du corps de base k qui est concerné.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour aider la lecture du texte qui suivra, voici une répartition des travaux présentés.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Arithmétique Géométrie CHM(k) §5 [Anc22] §8 [AEWH15, AHPL16] Mot(k) §7 [AF22] §9 [ACLS22] NUM(k) §6 [Anc21] Organisation du texte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les premières sections du texte fournissent une intro- duction partielle à la théorie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous avons déjà mentionné que les motifs sont utiles à l’étude des cycles algébrique et donné le Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 comme exemple, mais nous n’avons pas dit pourquoi ils sont utiles : c’est le but de la Section 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Sec- tion 2 présente la théorie de Hodge et ses pendants arithmétiques ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' on insiste sur les analogies mais surtout sur les différences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Section 3 donne des conjectures sur les cycles algébriques et la Section 4 des exemples de motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces questions et ces exemples sont repris dans les sections successives où l’on présente différents résultats organisés selon le diagramme ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la Section 5 on démontre la conjecture de conservativité pour les motifs provenant de variétés abéliennes défi- nies sur un corps fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Section §6 concerne la conjecture standard de type Hodge et montre le Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1(3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la Section §7 on introduit une nouvelle classe de périodes p-adiques qui surgit dans l’étude des classes algébriques en caractéristique mixte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la Section §8 on montre le Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1(4) qui nécessite l’utilisation de techniques motiviques modernes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Section §9 étudie les classes algébriques de certaines variétés hyper-kähler qui admettent une fibration lagrangienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 10 GIUSEPPE ANCONA 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A quoi servent les motifs ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans un article du même titre [Del94a], Deligne expliquait que les motifs n’ont qu’une utilité essentiellement philosophique permettant de transférer des idées d’une cohomologie à l’autre, grosso-modo en appliquant le diagramme 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 à différentes cohomologies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Aujourd’hui on comprend que les motifs sont aussi un vrai outil tech- nique, comme l’avait envisagé Grothendieck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Deligne même revoit sa position dans son Bourbaki sur les multizétas et explique que les travaux de Brown sont « un des cas où la philosophie des motifs est non seulement un guide précieux, mais permet des démonstrations » [Del13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J’aimerais expliquer ici l’utilité des motifs notamment dans la théorie des cycles algébriques : une liste d’énoncés où leur utilisation est cruciale a déjà été donnée dans le Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Plusieurs avertissements sont tout de même nécessaires.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pre- mièrement, je ne prétends pas que cet outil soit l’unique possible, beaucoup de résultats intéressants sur les anneaux de Chow ont été obtenus par d’autres mé- thodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Deuxièmement, à l’heure actuelle les applications les plus impressionnantes des motifs apparaissent plutôt dans la théorie des périodes [Bro12, Ayo15] - on peut espérer que les applications majeures de la théorie aux anneaux de Chow sont encore à venir.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans sa construction de la théorie des motifs purs (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' pour les variétés propres et lisses) Grothendieck avait imaginé un pont entre les cycles algébriques et, par exemple, la cohomologie ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Son idée était que la compréhension des premiers aurait impliqué ainsi des résultats sur la deuxième, par exemple les conjectures standard auraient impliqué les conjectures de Weil.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En un sens cela semblait la direction raisonnable : les cycles étaient là depuis plus longtemps (on pourrait dire depuis le théorème de Bézout) et leur définition pouvait les faire paraître comme plus accessibles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On comprend aujourd’hui qu’ils sont plus mystérieux que ce que l’on aurait pu imaginer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci est devenu flagrant probablement avec le théorème de Mumford [Mum68] qui montre que le groupe des 0-cycles ne peut pas, en général, être para- métré par une variété de type fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En revanche beaucoup de progrès ont été faits sur la cohomologie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce pont maintient donc toute son utilité mais il faut plutôt le parcourir dans l’autre direction : on essaiera d’exploiter des informations cohomo- logiques et de les transposer sur les cycles via les motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La question naturelle qui se pose est alors pourquoi l’application classe de cycle ne serait pas elle-même suffisante pour un tel pont entre anneaux de Chow et cohomologie?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une réponse courte est que la cohomologie a des propriétés agréables (Künneth, Poincaré,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') que les anneaux de Chow n’ont pas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les motifs sont une façon de réorganiser les applications classe de cycle de sorte à ce que l’on ait encore ces propriétés agréables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs purs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour expliquer l’idée derrière la construction, prenons la situa- tion suivante (inspirée par le formalisme tannakien).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit φ : H → G un morphisme de groupes et étudions le foncteur induit f = φ∗ : RepF (G) −→ RepF (H) (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) MÉMOIRE HDR 11 sur les représentations F-linéaires pour F un corps fixé.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons avoir à notre disposition pour cette étude uniquement la collection d’applications cV : V G −→ V H ⊂ f(V ) (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) pour chaque V ∈ Rep(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La donnée de cette collection est certainement moins agréable que la donnée de f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut tout de même retrouver f en remarquant que son action sur les morphismes est donnée par cW⊗V ∨ : HomG(V, W) = (W ⊗ V ∨)G −→ (W ⊗ V ∨)H = HomH(V, W).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) Cette idée de passer de l’invariant à l’équivariant est l’étape essentielle dans la construction de Grothendieck des motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas pour chaque variété X, V sera son motif h(X) (objet abstrait de la catégorie en construction), V G sera l’anneau de Chow CH(X), f(V ) sera la cohomologie singulière (ou ℓ-adique,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') H(X), V H seront les classes de Hodge Hdg(X) (ou les classes Galois invariantes, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') et cV sera l’application classe de cycle clX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En résumant : V ⇝ h(X), V G ⇝ CH(X), f(V ) ⇝ H(X), V H ⇝ Hdg(X), cV ⇝ clX .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) Pour que la construction dans (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) soit applicable il faut donner un sens au dual d’un motif et au produit tensoriel de deux motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est ici qu’il est nécessaire de considérer des variétés propres et lisses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet la formule de Künneth et la dualité de Poincaré suggèrent h(X) ⊗ h(Y ) = h(X × Y ) et h(X)∨ = h(X) (à un twist de Tate près).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’espace Hom(h(X), h(Y )) sera alors contrôlé 4 par CH(X × Y ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une fois que la construction de cette catégorie est faite on pourra imaginer - et essayer de démontrer - des analogies entre H(X) et h(X) qui ne seraient pas raisonnables avec CH(X) pour ensuite déduire des informations sur les anneaux de Chow en passant aux Hom dans la catégorie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple on pourra imaginer que H(X) et h(X) ont « la même dimension », ce qui ne peut pas avoir lieu avec CH(X) - voir l’analogie (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette idée est à la base de la notion de dimension dans les motifs due à Kimura et O’Sullivan qui est l’ingrédient essentiel dans la preuve du Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1(1) et joue également un rôle dans les parties (2) et (4) du même théorème.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs mixtes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La théorie a beaucoup évoluée depuis ses fondations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Comme la théorie de Hodge, qui a évolué d’abord avec les structures de Hodge mixtes as- sociées à des variétés qui ne sont pas forcément projectives ou lisses, pour arriver jusqu’aux modules de Hodge mixtes qui visent à étudier des familles de variétés sur des bases générales, également la théorie des motifs a eu une accélération si- gnificative avec la catégorie triangulée des motifs mixtes de Voevodsky jusqu’aux motifs relatifs sur une base générale [Ayo07, CD19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces catégories sont liées par le formalisme des six foncteurs, tout comme les modules de Hodge mixtes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus, certains Hom dans ces catégories permettent de retrouver les anneaux de Chow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour les variétés générales qui ne sont pas projectives et lisses l’argument ci-dessus suggère qu’il n’est pas raisonnable d’espérer un lien entre Hom(h(X), h(Y )) et des anneaux de Chow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ef- fectivement dans les motifs de Voevodsky ces Hom n’ont pas d’interprétation en terme d’invariants classiques, voir aussi la Remarque 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7(4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 12 GIUSEPPE ANCONA On dispose également de foncteurs de réalisation, par exemple vers les faisceaux constructibles, qui permettent de retrouver les applications classes de cycles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tout comme dans le cas des variétés projectives et lisses expliqué plus haut, ces catégories ont l’avantage d’avoir des analogies avec les catégories cohomologiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple on dispose d’une filtration de poids sur les motifs tout comme en théo- rie de Hodge mixte [Bon10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Une telle structure n’a pas de bons analogues dans les anneaux de Chow : par exemple un ouvert d’un espace affine a un anneau de Chow trivial alors que la filtration de poids en cohomologie peut être non triviale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Un autre exemple est le résultat suivant d’Ayoub [Ayo14, Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='24] : une application f entre motifs au-dessus d’une base S est un isomorphisme si et seule- ment si la restriction de f en tout point de S l’est.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cet énoncé est bien entendu inspiré de son pendant pour les faisceaux constructibles ou étales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces nouvelles catégories présentent un deuxième avantage : on dispose main- tenant de beaucoup plus de flexibilité, analogue à celle permise par les modules de Hodge mixtes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple, si on veut étudier l’anneau de Chow d’une variété projective et lisse X, on pourrait avoir envie de stratifier X et d’étudier chaque strate, ou de fibrer X au-dessus d’une base et d’étudier comment les fibres varient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce genre de construction mène très souvent à des variétés qui ne sont pas lisses et pour lesquelles les anneaux de Chow et leur fonctorialité ne sont pas définis : ces catégories de motifs permettent, entre autre, de contourner ce problème.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces techniques ont permis la construction de certains cycles « concrets », par exemple certains cycles prédits par la conjecture de Hodge joint au théorème de décomposition, qui étaient inaccessible par des méthodes directes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notamment cela a été appliqué aux variétés de Shimura [Wil17], aux fibrés en quadriques [CDN22] et aux variétés hyper-kähler qui admettent une fibration lagrangienne [ACLS22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Complexes motiviques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une troisième raison pour laquelle les motifs sont utiles à l’étude des cycles algébriques vient de leur définition moderne (depuis Voe- vodsky).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la théorie de Grothendieck les motifs sont des symboles formels et leur lien avec les cycles algébriques a lieu par construction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans les catégories mo- dernes les motifs sont des complexes de faisceaux et il est possible d’en construire un certain nombre explicitement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Leur lien avec les cycles algébriques est loin d’être une tautologie et c’est en fait un des résultats plus profond de la théorie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors espérer que certaines questions délicates sur les cycles algébriques puissent devenir concrètes dans leur pendant faisceautique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est ce qui se passe notam- ment dans la construction de la réalisation de Betti [Ayo10] ou dans la preuve du Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1(4), voir Section 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cohomologies de Weil Dans cette section on rappelle la définition classique de structure de Hodge (Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) ainsi qu’une formulation équivalente qui se prête à mieux décrire les propriétés de positivité (Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le but principal de la section est de rappeler ces propriétés de positivité ainsi que des propriétés d’autodualité des structures de Hodge puis de montrer que leurs analogues en cohomologie ℓ-adique sont faux en général (Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='16 et Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La définition de polarisation est cruciale, on essaie de la justifier dans la Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 13 D’autres différences entre la théorie de Hodge et ses analogues arithmétiques sont éparpillées un peu partout dans le texte, voir par exemple la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='18 ou la Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Structure de Hodge, définition classique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Une structure de Hodge pure de poids n ∈ Z est la donnée d’un Q-espace vectoriel de dimension finie V muni d’une décomposition V ⊗Q C = � p+q=n p,q∈Z V p,q (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) vérifiant V p,q = V q,p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si on considère les espaces V {p,q} = V ⊗QR∩(V p,q +V q,p) on trouve la définition équivalente suivante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Structure de Hodge, définition équivalente.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Une structure de Hodge pure de poids n ∈ Z est un triplet formé d’un Q-espace vectoriel de dimension finie V , d’une décomposition V ⊗Q R = � p≤q p+q=n p,q∈Z V {p,q} (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) en sous-espace réels et d’une structure complexe sur les espaces V {p,q} pour p ̸= q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les paires (p, q) qui apparaissent dans la décomposition sont appelées les types de V .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Lien entre les définitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Les deux présentations ci-dessus sont bien équivalentes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On remarque que, pour p < q, l’ espace V {p,q} ⊗ C est muni de deux structures complexes, par conséquent on a une décomposition V {p,q} ⊗ C = V p,q ⊕ V q,p, où V p,q est l’espace où les structures coïncident et V q,p est l’espace où elles sont conjuguées 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 est plus pratique pour exprimer des propriétés de positivité 6, voir par exemple Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Cohomologie singulière.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Si X est une variété projective et lisse sur les nombres complexes sa cohomologie singulière est munie d’une 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On pourrait inverser le rôle de ces deux espaces, dans ce cas il faudrait changer les signes dans la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9 pour avoir les mêmes conventions de signe classiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En revanche la structure tensorielle s’exprime mieux avec la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1, par la règle (V ⊗ W )p,q = � a+c=p b+d=q V a,b ⊗ W c,d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans le cas réel on remarque que V {a,b} ⊗ W {c,d} a deux structures complexes quand a < b et c < d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci induit une décomposition en deux espaces comme précédemment, celui où les structures coïncident contribue à (V ⊗ W )a+c,b+d et l’autre à (V ⊗ W )m,M où m et M sont le minimum et le maximum de la paire {a + d, b + c}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 14 GIUSEPPE ANCONA structure de Hodge fonctorielle en X, plus précisément Hn(X) est pure de poids n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tous les théorèmes que l’on peut imaginer (Poincaré, Künneth, Lefschetz,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') respectent cette structure supplémentaire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Variétés abéliennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Dans le cas particulier d’une variété abélienne com- plexe A dont la variété analytique sous-jacente est Cg/Λ on a H1(A)⊗R = Λ⊗R = Cg, ce qui fait apparaître explicitement la structure complexe dans la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 de l’espace H1(A) ⊗ R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) (Twist de Tate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') En poids 2n il existe une seule structure de Hodge de dimension 1 à isomorphisme non unique près.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On fixe une telle structure de Hodge et on la note Q(−n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour une structure de Hodge V on note V (−n) = V ⊗ Q(−n) et on l’appelle twist de Tate ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' on choisit les Q(−n) de sorte à avoir l’identification Q(a) ⊗ Q(b) = Q(a + b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Traditionnellement on choisit Q(n) = (2iπ)nQ ⊂ C, puisque c’est la structure de Hodge qui apparaît naturellement dans la cohomologie sin- gulière de degré maximale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce choix particulier ne joue pas de rôle dans les constructions qui suivront, sauf pour la notion de polarisation, voir Re- marque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7(1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Autodualité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Une structure de Hodge n’est pas, en général, au- toduale dans le sens où V et V ∨ ne sont pas en général isomorphes, même à un twist près (le seul twist pour lequel cette autodualité est raisonnable étant le poids).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple prenons une structure de Hodge de poids 0 et supposons que la décomposition (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) ait deux facteurs non nuls V ⊗Q R = V {0,0} ⊕ V {−1,+1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sup- posons de plus que V {0,0} soit compatible avec la structure rationnelle et V {−1,+1} ne le soit pas, c’est-à-dire que l’on ait l’égalité dimQ(V {0,0}∩V ) = dimR V {0,0} mais dimQ(V {−1,+1} ∩ V ) < dimR V {−1,+1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si V était autodual on aurait en particulier une forme bilinéaire sur V pour laquelle V {0,0} et V {−1,+1} seraient l’orthogonal l’un de l’autre.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or comme une telle forme bilinéaire serait définie sur Q, l’espace V {−1,+1} serait forcé à être également compatible avec la structure rationnelle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il se trouve que toutes les structures de Hodge provenant de la géométrie algé- brique sont autoduales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En fait même plus, elles sont munies d’une forme bilinéaire « aussi définie que possible », c’est l’objet de la définition de polarisation, rappelée ci-dessous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Polarisation - poids pair.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Une polarisation sur une structure de Hodge V pure de poids 2n est une forme bilinéaire symétrique b sur le Q-espace vectoriel V telle que : (1) Les espaces V {p,q} sont orthogonaux deux à deux par rapport à b, (2) L’adjointe par rapport à b de la structure complexe sur V {p,q} est sa conju- guée, (3) La restriction de b aux facteurs V {n−a,n+a} est définie (−1)a-positive 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est-à-dire définie positive si a est pair et définie négative si a est impair.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 15 Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) Les deux premières propriétés de la définition sont équi- valentes à l’existence d’un morphisme de structures de Hodge b : V ⊗ V −→ Q(−2n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inversement, étant donné un tel b il est nécessaire de choisir une identi- fication Q(−2n) ⊗Q R ∼= R pour pouvoir exprimer la dernière propriété (de positivité).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4(3) fixe un tel choix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Dans le cas V = H2n(X) on déduit qu’une polarisation est définie positive sur les classes algébriques, car elle l’est sur V {n,n}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est utile de connaître les signes sur les autres facteurs même pour la compréhension des classes algébriques, voir l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9 et la Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) Une polarisation est automatiquement non dégénérée donc les structures de Hodge qui la possèdent sont autoduales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part il y a des structures de Hodge autoduales, même simples, qui n’admettent pas de polarisation, voir l’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Autrement dit la polarisabilité est une notion plus forte que l’autodualité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Autodualité vs polarisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Construisons une structure de Hodge simple et autoduale qui n’admet pas de polarisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit A une variété abélienne complexe de dimension 4 et très générale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors la partie primitive V = H4,prim(A) est une structure de Hodge simple, de type (0, 4), (1, 3) et (2, 2) et polarisable (par le Corollaire 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définissons une nouvelle structure de Hodge W où le Q-espace vectoriel est le même que V , ainsi que la décomposition, mais on inverse le rôle de la partie (0, 4) avec la partie (1, 3), c’est-à-dire W {0,4} = V {1,3}, W {1,3} = V {0,4} et W {2,2} = V {2,2}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On prétend que W satisfait aux propriétés requises.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Premièrement remarquons qu’une polarisation sur V induit un accouplement b sur W qui rend W autoduale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part ce b ne peut pas être une polarisation car les signes de la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 ne sont pas respectés.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il reste à montrer que W n’admet pas de polarisation et pour cela il suffit de voir que b est l’unique accouplement sur W à scalaire près et donc que EndHS(W) = Q · Id .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or, par construction, on a EndHS(W) = EndHS(V ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour comprendre les endomorphismes de V , on prend le point de vue tannakien.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le groupe tannakien associé à H1(A) est GSp8 et la représentation sous-jacente à H4,prim(A) ⊂ H1(A)⊗4 est géométriquement irréductible, donc EndHS(V ) = Q · Id .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous rappelons ci-dessous la définition de polarisation dans le cas de poids im- pair, elle est un peu moins agréable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Même si on s’intéresse seulement aux classes algébriques les groupes de cohomologie de degré impair peuvent être utiles (par exemple la cohomologie d’une variété abélienne A est contrôlée par son H1) mais il est souvent suffisant de se rappeler uniquement que la définition de polarisation est stable par produit tensoriel (et donc une polarisation sur H1(A) en induira une sur Hn(A) via Hn(A) = ΛnH1(A)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Polarisation - poids impair.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Une polarisation sur une structure de Hodge V pure de poids 2n+1 est une forme bilinéaire alternée b sur le Q-espace vectoriel V telle que : (1) Les espaces V {p,q} soient orthogonaux entre eux par rapport à b, 16 GIUSEPPE ANCONA (2) L’adjointe par rapport à b de la structure complexe sur V {p,q} est sa conju- guée, (3) La forme bilinéaire symétrique b(·, i·) est définie (−1)a-positive sur les fac- teurs V {n−a,n+1+a}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) La Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7 s’applique également dans ce cas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) On pourrait vouloir travailler avec b(i·, ·) à la place de b(·, i·), dans ce cas les signes s’inverserait.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On trouve plus agréable d’avoir de la positivité sur la « partie centrale » V {n,n+1}, c’est analogue à ce qui se passe dans la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 avec la partie centrale V {n,n}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Polarisations issues de la géométrie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soit X une variété algé- brique complexe de dimension dX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’une part la dualité de Poincaré fournit une identification Hn(X) = H2dX−n(X)∨(−dX).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part, le théorème de Lefschetz difficile donne un isomorphisme Hn(X) ∼= H2dX−n(X)(dX −n), au moyen du choix d’une section hyperplane L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En combinant les deux on obtient une forme bilinéaire non-dégénérée bL : Hn(X) ⊗ Hn(X) −→ Q(−2n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Celle-ci est bien un morphisme de structure de Hodge, mais elle n’a pas les signes demandés par une polarisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour les obtenir, il faut modifier les signes de cer- tains facteurs de la décomposition de Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple écrivons la décomposition de Lefschetz en degré six H6 = H6,prim ⊕ H4,prim(−1) ⊕ H2,prim(−2) ⊕ H0,prim(−3), elle est orthogonale par rapport à l’accouplement bL ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La décomposition de Hodge de chacune de ces quatre structures de Hodge est encore orthogonale;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' la signature de bL est la suivante (où les cases vides sont pour les sous-espaces qui sont toujours réduits à zéro).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' signes bL H6,prim H4,prim(−1) H2,prim(−2) H0,prim(−3) (3, 3) + + (2, 4) + + (1, 5) + (0, 6) + Pour obtenir une polarisation il faudra donc changer le signe sur les facteurs H6,prim et H2,prim(−2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Pourquoi la définition de polarisation?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Considérons le change- ment de signe expliqué dans l’exemple ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il serait tentant, à première vue, de changer encore de signe, cette fois-ci par rapport à la décomposition en V {p,q}, de sorte à avoir une forme quadratique définie positive dans la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le problème est que cette deuxième décomposition n’est pas en général définie sur Q et donc ce changement de signes donnerait une forme bilinéaire qui n’est pas définie sur Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De manière générale, pour les structures de Hodge provenant de la géométrie, on ne peut pas espérer avoir une forme quadratique qui soit à la fois définie sur Q, compatible avec la décomposition de Hodge et définie positive, voir l’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il faut alors imaginer la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 comme la meilleure approximation d’un MÉMOIRE HDR 17 produit scalaire qui puisse exister pour les structures de Hodge issues des variétés algébriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’ailleurs la proposition ci-dessous montre que la polarisation a toutes les conséquences que l’on aimerait déduire d’un produit scalaire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient V une structure de Hodge, b une polarisation sur V et W ⊂ V une sous-structure de Hodge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors la restriction de b à W est une polarisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus l’orthogonal W ⊥ ⊂ V de W par rapport à b est une sous- structure de Hodge et on a l’égalité de structures de Hodge V = W ⊕ W ⊥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Corollaire 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit X une variété complexe projective et lisse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors tout sous- structure de Hodge de la cohomologie singulière de X est polarisable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Polarisations vs produits scalaires.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Construisons une structure de Hodge V de poids zéro et d’origine géométrique telle que HomHS(V ⊗ V, Q(0)) ne contient pas une forme quadratique définie positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit E une courbe elliptique non CM et considérons H1(E).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On a une décom- position de structures de Hodge H1(E) ⊗ H1(E)∨ = Q(0) ⊕ V , où V a dimension 3 et types (0, 0) et (−1, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On prétend que HomHS(V, V ∨) = HomHS(V ⊗V, Q(0)) est un Q-espace vectoriel de dimension 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le montrer on prend le point de vue tannakien.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut voir que le groupe tannakien associé à H1(E) est GL2 et que H1(E) est la représenta- tion standard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le groupe GL2 agit sur V et on a l’identification HomHS(V, V ∨) = HomGL2(V, V ∨).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part, la théorie classique des représentations nous dit que V est géométriquement irréductible et donc que HomGL2(V, V ∨) est un Q-espace vectoriel de dimension (au plus) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part l’espace HomHS(V ⊗V, Q(0)) contient une polarisation, par le Corol- laire 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14, et donc cet espace est formé uniquement de multiples d’une polarisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier HomHS(V ⊗ V, Q(0)) ne peut pas contenir une forme quadratique définie positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut aussi en déduire que H2(E × E) n’admet pas une forme quadratique définie positive qui respecte la structure de Hodge car V (−1) en est un facteur direct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le reste de la section insiste sur les différences entre théorie de Hodge et cohomo- logie ℓ-adique, au regard notamment de la notion de polarisation et des propriétés d’autodualité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Polarisations en cohomologie ℓ-adique?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La proposition ci-dessus dit en particulier que les structures de Hodge issues de la géométrie algébrique forment une catégorie semi-simple et que chaque objet est autodual à un twist près.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est une grande différence avec la cohomologie ℓ-adique : la semi-simplicité est seulement conjecturale et l’autodualité est fausse en général, voir l’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La notion même de polarisation n’a pas d’analogue : on ne peut même pas formuler des propriétés de positivité analogues à celles de la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 pour la simple raison que la notion de positif n’a pas de sens dans Qℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut construire des accouplements sur la cohomologie ℓ-adique de la même façon qu’en théorie de Hodge, comme dans l’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On ne connait pas de formulation, même conjecturale, qui décrirait cette Qℓ-forme quadratique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par 18 GIUSEPPE ANCONA ailleurs les invariants d’une telle forme quadratique, comme son symbole de Hilbert, ne contrôlent pas ceux des sous-formes quadratiques (contrairement à ce qui se passe avec la signature).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci suggère que même si on trouvait une propriété analogue à la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 pour les groupes de cohomologies Hn ℓ (X) elle pourrait ne pas être valable pour les facteurs directs de Hn ℓ (X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Non autodualité en cohomologie ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soient k un corps de type fini et E une courbe elliptique définie sur k telle que End(E)⊗Q soit un corps quadratique imaginaire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Prenons un nombre premier ℓ différent de la caractéristique de k et tel que End(E) ⊗ Qℓ = Qℓ ⊕ Qℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors l’action de End(E) ⊗ Qℓ donne une décomposition de représentations galoisiennes H1 ℓ (E) = V ⊕ W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le cup-produit in- duit une autodualité sur le H1 ℓ (E), elle réalise W comme dual de V par la positivité de Rosati.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part on prétend que V et W ne sont pas isomorphes comme repré- sentation de Galois, ce qui impliquera en particulier que V n’est pas autodual.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet s’ils étaient isomorphes on aurait EndGal H1 ℓ (E) = M2×2(Qℓ), or on a EndGal H1 ℓ (E) ∼= End(E) ⊗ Qℓ comme prédit par la conjecture de Tate, montrée dans ce context par Tate, Faltings et Zarhin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’exemple ci-dessus dépend du nombre premier ℓ choisi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On s’at- tend à ce que l’on ne puisse pas trouver une représentation d’origine géométrique, « indépendante de ℓ » et non autoduale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette idée est rendue précise par les motifs, voir la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 et la remarque qui la suit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjectures standard et moins standard Soient k un corps de base et H∗ une cohomologie de Weil.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons le dia- gramme (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) qui représente des catégories de motifs : CHM(k) π �� R �▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ Mot(k)� � I � π′ �� GrVect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' NUM(k) on les considère toujours à coefficients rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour chaque variété projective et lisse X, de dimension dX, on considère son motif h(X), on utilisera ce même symbole dans les différentes catégories, on précisera la catégorie concernée si cela est important.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le but de cette section est de donner une collection de conjectures qui permettent d’avoir une intuition sur les motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces conjectures sont essentiellement classiques, hormis la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 d’autodualité et la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7 de positivité qui sont nouvelles à notre connaissance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La section est organisée en trois sous-sections.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La première présente les conjec- tures principales de la théorie (qui essentiellement entrainent les autres conjectures).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La deuxième sous-section présente les conjectures dite standard de Grothendieck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La dernière porte sur la notion de dimension finie de Kimura.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 19 Avec les techniques actuelles ces conjectures sont hors de portée en toute généra- lité, mais il est possible en démontrer des cas particuliers, comme on le verra dans les sections successives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjectures principales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Chow–Künneth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Il existe une décomposition (non unique) h(X) = 2dX � n=0 hn(X) dans la catégorie CHM(k) telle que R(hn(X)) = Hn(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Non unicité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Le facteur h0(X) existe toujours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le construire il suffit de considérer une application constante de X vers X : elle sera bien l’identité sur le H0(X) et nulle sur les autres groupes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le facteur ainsi défini dépend de l’image de cette application constante, ou plus précisément de sa classe modulo équivalence rationnelle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier une telle décomposition ne peut pas être unique en général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Autodualité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') On peut conjecturer l’existence d’une décomposition de Chow– Künneth qui ait la propriété supplémentaire d’être autoduale, c’est-à-dire que si on considère la dualité de Poincaré h(X)∨ = h(X)(dX) alors le fac- teur hn(X)∨ correspond à h2dX−n(X)(dX).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’un point de vue de cycles algébriques cela veut dire que les projecteurs pn sont donnés par une collec- tion de cycles dans X×X telle que σ∗pn = p2d−n, où σ : X×X → X×X est l’inversion des deux facteurs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce n’est pas automatique de construire une décomposition autoduale à partir d’une décomposition de Chow–Künneth : même si l’on pose p2d−n = σ∗pn on pourrait avoir que les projecteurs pn et σ∗pn ne sont pas orthogonaux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Conservativité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Tous les foncteurs du diagramme (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) sont conservatifs, c’est–à-dire un morphisme entre motifs est en fait un isomorphisme si son image via un de ces foncteurs du diagramme l’est.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On dit qu’un motif est de poids n s’il est facteur direct d’un motif de la forme hn(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Autodualité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Si M est un motif de poids n (toujours à coeffi- cients rationnels) alors il existe un isomorphisme (non unique) M ∨ ∼= M(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Autodualité des motifs vs autodualité en cohomologie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Les groupes de cohomologies Hn(X) jouissent de ce type d’autodualité via la dualité de Poincaré et l’isomorphisme de Lefschetz difficile (qui dépend du choix d’une section hyperplane), mais en général les facteurs directs de Hn(X) n’ont pas cette propriété d’autodualité, voir l’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci n’est pas en contradiction avec la conjecture ci-dessus : les exemples construits ne sont pas la réalisation d’un motif à coefficients rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 20 GIUSEPPE ANCONA (2) (Lien avec les conjectures classiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Cette conjecture est nouvelle à notre connaissance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous trouvons sa formulation naturelle et elle nous a guidé dans l’étude de la conjecture de conservativité, voir la Section 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle suit par ailleurs de conjectures classiques de positivité (notamment la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15 que l’on verra plus loin), de la même manière que l’autodualité pour les structures de Hodge suit des propriétés de positivité des polarisations, voir la Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13 et la remarque qui la suit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Positivité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Supposons que le corps de base k soit de caractéris- tique p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient M un motif homologique sur k et q : Sym2 M → 1 un morphisme dans Mot(k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que q soit la réduction modulo p d’une application �q : Sym2 � M → 1 définie en caractéristique zéro.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définissons qZ comme la restriction de q à toutes les classes algébriques Z(M) = Hom(1, M) de M et qB comme la réalisation singulière de �q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que qB soit une polarisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors qZ est définie positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Sur la relevabilité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') L’hypothèse de relevabilité à la ca- ractéristique zéro est vérifiée dans des cas intéressants, par exemple les variétés abéliennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas, les classes algébriques qui se relèvent à la caractéristique zéro vérifient automatiquement la conjecture, voir la Re- marque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soulignons tout de même qu’il y a en général des classes algé- briques qui ne sont pas relevables, même dans le cas des variétés abéliennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Lien avec les conjecture classique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Cette conjecture est nouvelle et nous ne savons pas si elle peut se déduire de conjectures classiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous trouvons sa formulation naturelle et elle nous a guidé dans l’étude de la conjecture standard de type Hodge, voir la Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjectures standard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Künneth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Dans la catégorie Mot(k) des motifs homologiques, il existe une décomposition h(X) = 2dX � n=0 hn(X) telle que la réalisation de hn(X) soit Hn(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Lien avec Chow–Künneth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Cette conjecture est bien sûr une conséquence de la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons qu’une décomposition de Künneth est automatiquement unique et autoduale par définition d’équivalence homologique : il s’agit de la graduation de GrVect et de la dualité de Poincaré en cohomologie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est une différence avec la décomposition de Chow–Künneth, voir la Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soient dX la dimension de X et L une section hyperplane de X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors pour tout n ≤ dX il existe une correspondance γn dans X × X dont la réalisation en degré dX + n induit un isomorphisme R(γn) : HdX+n(X) ∼ −−→ HdX−n(X)(−n) MÉMOIRE HDR 21 qui est l’inverse du cup produit par Ln.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Lien avec l’autodualité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La Conjecture standard de type Les- chetz 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11 implique la Conjecture standard de type Künneth 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9, c’est un argument classique de Kleiman [Kle68].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inversement, si X vérifie Künneth alors la conjecture de type Lefschetz est équi- valente à l’autodualité hn(X)∨ ∼= hn(X)(n) du motif homologique hn(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette équivalence se déduit de la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle montre en particulier que la conjec- ture de type Lefschetz ne dépend pas de la section hyperplane L choisie et elle suit de la Conjecture d’autodualité 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (hom = num.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Le foncteur π′ est une équivalence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La quatrième et dernière des conjectures standard est celle de type Hodge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour la formuler il est nécessaire d’introduire la proposition suivante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Anc21, §3]) Supposons que X vérifie la conjecture stan- dard de type Lefschetz (Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Choisissons une section hyperplane L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors le motif homologique hn(X) admet une décomposition en parties primitives et il est possible de construire un accouplement qX,n,L : hn(X) ⊗ hn(X) → Q(−n) de façon analogue à la construction d’une polarisation sur la cohomologie singulière d’une variété algébrique complexe 8, voir l’Exemple 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus, si on restreint l’accouplement qX,2n,L(2n) : h2n(X)(n) ⊗ h2n(X)(n) → Q aux classes algébriques Zn(X)/ hom = HomMot(k)(1, h2n(X)(n)) on obtient une Q- forme quadratique qZ,hom : Zn(X)/ hom ⊗ Zn(X)/ hom → Q dont le noyau est formé exactement par les cycles numériquement triviaux 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier qZ,hom induit une forme quadratique sur les cycles modulo équivalence numérique qZ : Zn(X)/ num ⊗ Zn(X)/ num → Q qui est non dégénérée.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Conjecture standard de type Hodge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La forme quadratique qZ : Zn(X)/ num ⊗ Zn(X)/ num → Q introduite ci-dessus est définie positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Lien avec la conjecture de positivité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La conjecture standard de type Hodge (Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15) ne demande pas la relevabilité de X à la caracté- ristique zéro et en ce sens elle est plus générale que la Conjecture de positivité 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part pour les variétés relevables la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7 est plus générale que la 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier, si k = C et H∗ est la cohomologie singulière, R(qX,n,L) est la polarisation classique induite par L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Autrement dit, la conjecture hom = num pour X est équivalente à dire que qZ,hom est non dégénérée.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 22 GIUSEPPE ANCONA conjecture standard de type Hodge, car elle s’applique à des polarisations abstraites qui ne seraient pas forcément celles provenant d’une section hyperplane.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Positivité en caractéristique zéro.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') En caractéristique zéro, la forme quadratique introduite ci-dessus qZ : Zn(X)/ hom ⊗ Zn(X)/ hom → Q est définie positive : c’est une conséquence des propriétés de positivité d’une polari- sation (dites relations de Hodge–Riemann), voir Remarque 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier, en caractéristique zéro, la conjecture standard de type Lefschetz implique les autres conjectures standard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une autre différence entre la caractéristique zéro et la caractéristique positive se trouve dans la conjecture suivante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons l’application classe de cycle ℓ-adique clX : CH(X) → Hℓ(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors le Q-espace vectoriel Im clX est de dimension finie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Plus précisément l’application canonique Im clX ⊗QQℓ → Hℓ(X) est injective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette conjecture est une conséquence de la conjecture hom = num car l’équivalence numérique commute à l’extension des scalaires.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En carac- téristique zéro elle est connue : on utilise les théorèmes de comparaison pour se reporter à la cohomologie singulière.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs de dimension finie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Dimension finie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Tout motif de Chow M admet une décom- position (non unique) M = M+ ⊕ M− vérifiant ΛNM+ = 0 et SymN M− = 0 pour un naturel N assez grand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Lien avec les autres conjectures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Supposons que M = h(X) admet une décomposition de Chow–Künneth (Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors M+ = � h2n(X) et M− = � h2n+1(X) devraient vérifier la conjecture ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet la conservativité (Conjec- ture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) prédit qu’il suffit de vérifier ces relations en cohomologie, or la réa- lisation de M+ est un espace vectoriel gradué de dimension finie concentré en degrés pairs, par conséquence toute puissance N-ème extérieure l’annule dès que N est plus grand que la dimension totale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le raisonnement est analogue pour M−.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On remarque que la réalisation est concentrée en degré impair et que, par la règle des signes de Koszul, une puissance symétrique sur un motif ou une variété devient une puissance extérieure sur les groupes de cohomologie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Au delà d’être une conjecture naturelle, la notion de di- mension finie s’est avéré être utile : elle est stable par plusieurs opérations, dont le produit tensoriel et le passage à un facteur direct, elle est vérifiée au moins par les courbes, et elle permet de déduire la conservativité pour tous les foncteurs tensoriels quotient (notamment π et π′ de (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela a permis à Kimura de déduire la conjecture de Bloch pour les surfaces dominées par MÉMOIRE HDR 23 un produit de courbes [Kim05], voir le Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1(1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En appliquant ces propriétés de conservativité au Frobenius, Kahn a déduit que l’application classe de cycle est injective pour les produits de courbes elliptiques sur un corps fini [Kah03], voir le Théorème 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1(2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemples Dans cette section on discute des exemples de motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces exemples sont organisés dans trois sous-sections.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la première on présente des cas classiques où les conjectures de la section précédente sont vérifiées.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La deuxième sous-section montre des subtilités (assez amusantes!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') entre les différentes réalisations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La dernière partie étudie les motifs de variétés abéliennes CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tous les exemples de cette section sont repris à plusieurs endroits dans le texte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On continue à travailler avec le diagramme (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) : CHM(k) π �� R �▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ Mot(k)� � I � π′ �� GrVect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' NUM(k) et h(X) indiquera le motif d’une variété X dans les différentes catégories, on préci- sera la catégorie concernée si cela est important.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Quelques évidences des conjectures de la Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient h(X) ∈ Mot(k) un motif homologique, f un endomor- phisme de h(X) et pn(f) le polynôme caractéristique de f agissant sur Hn(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que pn(f) et pm(f) soient premiers entre eux pour tous les n ̸= m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors h(X) admet la décomposition de Künneth h(X) = � n=0 hn(X), voir la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus les projecteurs de cette décomposition appartiennent à l’algèbre Q[f].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) La preuve est élémentaire : on applique l’identité de Be- zout entre pn(f) et � n̸=m pm(f), voir [KM74].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) On peut appliquer cette proposition à toute variété projective et lisse définie sur un corps fini et à f leur Frobenius : les polynômes pn(f) vont bien entre premiers être eux par les conjectures de Weil.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La proposition s’applique aussi aux variétés abéliennes sur un corps quel- conque.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas f est la multiplication par un entier N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) Ces résultats ne s’étendent pas automatiquement à une décomposition de Chow–Künneth (Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1), notamment pour les variétés sur un corps 24 GIUSEPPE ANCONA fini c’est une question ouverte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On dispose d’une décomposition de Chow– Kunneth pour les variétés abéliennes [DM91], mais pour cela il faut utiliser la transformée de Fourier, voir Section 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient M un motif homologique et f un endomorphisme de M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors f est inversible si et seulement si son action en cohomologie l’est.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) La preuve est élémentaire : on applique Cayley–Hamilton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Cette question donne une réponse affirmative à la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3 de conser- vativité pour le foncteur I : Mot(k) −→ GrVect pour les endomorphismes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut partiellement étendre le résultat aux morphismes quelconques de la façon suivante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient M et N deux motifs homologiques et considérons deux applications f : M → N et g : N → M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que I(f) et I(g) soient des isomorphismes en cohomologie, alors f et g sont des isomorphismes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) C’est une conséquence de la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3 appliquée aux endomorphismes fg et gf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) On remarquera que ce n’est pas une solution complète de la conservativité pour I : étant donné un morphisme f il faut être capable d’en construire un dans l’autre sens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette proposition est tout de même utile comme on peut le voir dans la proposition ci-dessous, ainsi que dans la Section 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Kleiman [Kle68]) Soit A une variété abélienne de dimension g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors son motif homologique admet une décomposition de Künneth h(A) = 2g � n=0 hn(A) et un isomorphisme de motifs en algèbres de Hopf graduées 2g � n=0 hn(A) = 2g � n=0 Symn h1(A) qui donne en cohomologie l’isomorphisme classique 10 H∗(A) = Λ∗H1(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus A vérifie la conjecture standard de type Lefschetz (Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 et Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11) h2g−n(A)(g) = hn(A)∨ ∼= hn(A)(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La décomposition de Künneth a déjà été discutée dans la Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2(3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On considère l’inclusion diagonale ∆ : A ֒→ An.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle induit une application h(∆) : h(A)⊗n → h(A) qui donne une application Symn h1(A) → hn(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De façon analogue on construit une application dans l’autre sens en partant du morphisme de somme s : An → A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le changement de puissance symétrique en alternée est le même que celui dans la Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='21(1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 25 On peut maintenant appliquer la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 et déduire �2g n=0 hn(A) ∼= �2g n=0 Symn h1(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les applications construites ci-dessus respectent les structures d’algèbre de Hopf par définition de ces dernières, ce sont d’ailleurs les structures d’algèbre de Hopf qui forcent ces applications à être des isomorphismes en cohomo- logie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’égalité h2g−n(A)(g) = hn(A)∨ est donnée par la dualité de Poincaré, voir la Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour l’autodualité hn(A)∨ ∼= hn(A)(n) on se réduit au cas n = 1, grâce à l’égalité Symn h1(A) = hn(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour h1(A)∨ ∼= h1(A)(1) on applique encore la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une applica- tion est donnée par l’opérateur de Lefschetz Lg−1 : h1(A) → h2g−1(A)(g − 1) = h1(A)∨(−1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour l’autre sens, fixons une isogénie entre A et sa duale A∨, ce qui donne un isomorphisme h1(A) ∼= h1(A∨).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il s’agit alors de construire une applica- tion de h1(A)∨(−1) vers h1(A∨).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est la donné d’un diviseur sur A×A∨, le diviseur de Poincaré convient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les résultats de la proposition ci-dessus sont valables même dans CHM(k) mais leur preuve est plus délicate et repose sur la transformée de Fourier pour les anneaux de Chow de variétés abéliennes, voir aussi la Section 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs et théorèmes de comparaison.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Réduction supersingulière et Q-structures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soit S une surface K3 complexe de rang de Picard maximal et choisissons un premier p de bonne réduction tel que la réduction Sp soit supersingulière.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La surface quartique de Fermat avec p ≡ −1 (4) est un tel exemple [SK79].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas le motif homologique vérifie h(S) = 1 ⊕ h2(S0) ⊕ 1(−2) et h2(S0) = 1(−1)⊕20 ⊕ h2,tr(S).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le motif h2,tr(S) est appelé motif transcendant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sa réalisation singulière est une structure de Hodge de dimension 2 et de type (0, 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sa réduction modulo p est isomorphe à 1(−1)⊕2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notons M le motif h2,tr(S)(1), VB sa réalisation singulière, Mp sa réduction modulo p, et Zp les classes algébriques en caractéristique p, c’est-à-dire Zp = Hom(1, Mp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons que VB et Zp sont des Q-espaces vectoriels de dimen- sion 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons les identifications VB ⊗Q Qℓ = Rℓ(M0) = Rℓ(Mp) = Zp ⊗Q Qℓ, (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) où la première est donnée par le théorème de comparaison d’Artin entre cohomologie singulière et ℓ-adique, la deuxième suit du théorème de changement de base propre et lisse et la dernière vient de la propriété de supersingularité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut imaginer l’identification VB ⊗Q Qℓ = Zp ⊗Q Qℓ comme un isomorphisme de périodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est ce point de vue qui a inspiré le travail présenté dans la Section 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il est naturel de se demander si les deux Q-structures VB et Zp sont respectées sous cette identification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Expliquons pourquoi la réponse est non.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si on considère le cup-produit sur VB et Zp on déduit deux Q-formes quadratiques qB et qZ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si les espaces étaient égaux on aurait, en particulier, que ces deux formes quadratiques 26 GIUSEPPE ANCONA seraient isomorphes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or qB est définie positive par les relations de Hodge–Riemann et qZ est définie négative par le théorème de l’indice de Hodge 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Gardons les notations de l’exemple ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est intéressant de regarder les Q-formes quadratiques qB et qZ non seulement à la place à l’infini mais aussi aux autres completions de Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’identification (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) montre l’égalité qB ⊗Qℓ = qZ ⊗ Qℓ pour tout ℓ ̸= p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La seule place qui reste à déterminer est en p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or elle est déterminée par les autres places 12 et en particulier qB ⊗ Qp ̸= qZ ⊗ Qp puisque qB ⊗ R ̸= qZ ⊗ R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une remarque amusante : observons que la forme quadratique qZ reconnait le nombre premier p, c’est en effet le seul nombre premier pour lequel qB ⊗ Qp ̸= qZ ⊗ Qp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier si on fait varier p parmi tous les premiers à réduction supersingulière on trouvera des Q-formes quadratiques toujours différentes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Plus sérieusement, remarquons que les Qp-espaces VB ⊗QQp et Zp⊗QQp ont une interprétation cohomologique : VB ⊗Q Qp est isomorphe à la réalisation p-adique de M, encore par le théorème de comparaison d’Artin, et Zp ⊗Q Qp est la partie Frobenius invariante de la réalisation crystalline de Mp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il est naturel alors de se demander si la relation qB⊗Qp ̸= qZ⊗Qp peut s’obtenir par des méthodes purement p-adiques, par exemple via la théorie de Hodge p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La réponse est oui, elle sera esquissé dans la Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces techniques permettront par ailleurs de montrer des cas de la Conjecture de positivité 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7, voir encore la Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Périodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Considérons la catégorie CHM(Q) des motifs définis sur le corps Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle est munie de deux foncteurs vers les Q-espaces vectoriels gradués RB, RdR : CHM(Q) −→ GrVectQ la réalisation singulière, ou de Betti, et la réalisation de de Rham algébrique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les théorèmes de comparaison fournissent une identification RB ⊗Q C = RdR ⊗Q C essentiellement induite par l’integration des formes différentielles algébriques sur des simplexes topologiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les coefficients complexes qui apparaissent dans cette identification sont appelés périodes et sont attendus être aussi transcendants que possible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le formalisme tannakien montre l’existence d’isomorphismes RB ⊗Q Q ∼= RdR ⊗Q Q de foncteurs monoïdaux (mais on ne sait pas en construire un explicitement).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mon- trons qu’en revanche il ne peut pas y avoir d’isomorphisme monoïdal entre RB et RdR et même que l’on a RB ⊗Q R ̸∼= RdR ⊗Q R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le montrer il suffit de trouver un motif M muni d’une application q : Sym2 M −→ 1 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Voir respectivement la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 et la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15, qui est connue pour les divi- seurs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les formes quadratiques qu’y apparaissent sont obtenues à partir du cup-produit par un changement de signe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela suit de la formule du produit sur les symboles de Hilbert, l’argument sera détaillé dans la Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 27 telle que les réalisations RB(q) et RdR(q) sont deux formes quadratiques avec si- gnature différente.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple on peut prendre M tel que RB(M) soit une structure de Hodge de poids 0 et types (−1, 1) et (0, 0) et q tel que RB(q) soit une polarisation de RB(M), voir Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors la signature de la forme quadratique RB(q) est (dim RB(M)0,0, 2 · dim RB(M)−1,1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part la réalisation RdR(q) respecte la filtration de de Rham et donc l’espace Fil1RdR(q) est isotrope ce qui force la signature (s+, s−) de RdR(q) à vérifier s+ ≥ dim Fil1RdR(q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or dim Fil1RdR(q) = dim RB(M)−1,1, il suffit donc de trouver un exemple où l’on a dim RB(M)−1,1 > dim RB(M)0,0 pour conclure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De tels exemples se trouvent dans la catégorie engendrée par les courbes d’équa- tion Cn : y2 = xn−1, voir [Sch15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Plus précisément M sera 13 la partie G-invariante du motif h2(CN n )(1) pour certains entiers n, N et pour un groupe fini G convenable agissant sur CN n et défini sur Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notons par ailleurs que, par le théorème de comparaison de de Rham, RB ⊗Q R est isomorphe à la cohomologie de de Rham classique du lieu complexe sous-jacent, calculée à l’aide des formes différentielles C∞ à coefficients réels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier on vient de montrer que, pour les variétés algébriques définies sur R, il ne peut pas y avoir d’isomorphisme naturel entre la cohomologie de de Rham algébrique de la variété et la cohomologie de de Rham classique du lieu complexe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs de variétés abéliennes CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7 montre que le motif d’une variété abélienne A est contrôlé par son h1(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si A est une variété abélienne CM on peut utiliser l’action CM pour décomposer h1(A) et donc le motif de A tout entier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette décomposition est bien utile : Clozel l’avait déjà utilisée pour un résultat sur la conjecture hom = num (voir le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) et on l’utilisera également pour les Conjectures de conservativité et de positivité, voir les Section 5 et 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part, bien qu’élémentaire, cette décomposition n’est pas digeste à la première lecture, on encourage à y revenir au fur et à mesure de ses applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit A une variété abélienne simple de dimension g définie sur un corps fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fixons une polarisation et considérons l’involution de Rosati induite sur End(A) ⊗ Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par Honda–Tate il existe un corps CM de degré 2g et une inclusion F ⊂ End(A) ⊗ Q tels que l’involution laisse stable F et agit comme la conjugaison complexe sur F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons l’action de F sur le motif h1(A) dans la catégorie CHM(k)� F des motifs de Chow à coefficients dans une clôture galoisienne �F de F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle décompose le motif h1(A) = L1 ⊕ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ⊕ L2g (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) en une somme de 2g facteurs échangés par l’action du groupe de Galois Gal( �F/Q) et la réalisation de chaque facteur est une droite propre pour l’action de F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus le choix d’une polarisation induit un morphisme dans CHM(k) q1 : h1(A) ⊗ h1(A) −→ 1(−1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons que le premier candidat que l’on pourrait imaginer, à savoir M = h2(S)(1) avec S une surface, ne peut pas fonctionner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet pour les surface on a l’inégalité opposée : h1,1 > h0,2, voir [Sch15, Proposition 22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 28 GIUSEPPE ANCONA Par rapport à cet accouplement, une droite propre est orthogonale à toutes les autres hormis sa conjuguée complexe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' À l’aide de la formule hn(A) = Symn h1(A) on déduit une décomposition de hn(A) dans CHM(k)� F en somme de facteurs dont la réalisation a dimension un : chaque facteur correspond au produit tensoriel de n différents Li.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus l’accou- plement qn = Symn q1 rend la réalisation d’une telle droite orthogonale à toutes les autres hormis sa conjuguée complexe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Gardons les notations de l’exemple ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) Soient α1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' , α2g les valeurs propres de l’action du Frobenius sur h1(A) comptées avec multiplicité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Quitte à les renuméroter on a αi · α2g−i = q, où q est le cardinal du corps de base.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette symétrie des valeurs propres provient de l’accouplement parfait q1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les valeurs propres de l’action du Frobenius sur hn(A) sont données par tous les produits possibles de n distincts αi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La dimension de l’espace des classes Galois-invariantes dans H2n(A)(n) est alors donnée par le nombre de collections {αi1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' , αi2n} vérifiant αi1 · .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' · αi2n = qn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) La conjecture de Tate prédit que chaque droite propre de H2n(A)(n) cor- respondant à une telle collection contient une classe algébrique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette conjecture est connue pour les diviseurs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une droite propre contient donc une intersection de diviseurs si et seulement si la collection αi1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' , αi2n vérifie αij · αi2n−j = q, ∀j, (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) quitte à renuméroter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Soit F0 le plus grand sous-corps totalement réel de �F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le groupe de Galois Gal( �F /F0) est d’ordre deux, engendré par la conjugaison complexe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Son action recolle la décomposition de l’exemple ci-dessus en une décomposition de hn(A) dans la catégorie CHM(k)F0 des motifs de Chow à coefficients dans F0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette décomposition est orthogonale par rapport à l’accouplement qn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les facteurs obtenus sont de rang un ou deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) L’action du groupe de Galois Gal( �F /Q) recolle la décomposition de l’exemple ci-dessus en une décomposition de hn(A) dans la catégorie CHM(k) des mo- tifs de Chow à coefficients rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette décomposition est orthogonale par rapport à l’accouplement qn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le rang des facteurs obtenus varie et vaut au plus 2g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par la description du point (1) on remarque que chaque facteur de h2n(A)(n) rentre dans une des trois catégories suivantes : (a) La réalisation du facteur est engendrée par des classes qui sont toutes intersections de diviseurs, (b) La réalisation du facteur est Frobenius invariante mais ne contient au- cune intersection de diviseurs, (c) Le Frobenius agissant sur la réalisation du facteur n’a aucun vecteur fixe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 29 Pour les questions de cycles algébriques c’est surtout la classe (b) qui est intéressante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si A est de dimension quatre les facteurs de ce type ont tou- jours rang deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le montrer il s’agit d’étudier les quadruplets vérifiant (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) mais qui ne vérifient pas (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est une étude élémentaire mais dont la combinatoire est délicate, voir [Anc21, §7] pour les détails.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Autodualité et conservativité Cette section concerne les Conjectures 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3 et 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 de conservativité et d’autodua- lité et les résultats que l’on peut obtenir pour les variétés abéliennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On notera CHM(k)ab, Mot(k)ab et NUM(k)ab les catégories de motifs engendrées par les motifs de variétés abéliennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On commence par rappeler les théorèmes fondamentaux de semisimplicité de Jannsen et de nilpotence de Kimura, puis on en déduit les Conjectures 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3, et 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 pour CHM(k)ab, avec k = C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans une deuxième partie on explique le contenu de [Anc22] qui étudie ces conjectures pour k = Fq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il faudra combiner les théorèmes de Jannsen et Kimura avec les décompositions de l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12 induites par la multiplication complexe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette méthode est inspirée par un travail de Clozel [Clo99] que nous rappellons également.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Jannsen [Jan07]) La catégorie NUM(k) des motifs numériques est semisimple.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Kimura–O’Sullivan [Kim05, O’S05]) Le noyau du foncteur de projection πnum : CHM(k)ab −→ NUM(k)ab est nilpotent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier le foncteur πnum est conservatif et toute décomposition dans NUM(k)ab se relève en une décomposition dans CHM(k)ab.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’énoncé reste valable si on remplace NUM(k)ab par Mot(k)ab (ou n’importe quelle catégorie ten- sorielle quotiente).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La Conjecture d’autodualité 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 et la conjecture hom = num sont vraies pour tout motif dans CHM(C)ab.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus le foncteur de réalisation singulière R : CHM(C)ab −→ GrVectQ est conservatif (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La conjecture standard de type Lefschetz est vraie pour les variétés abéliennes (Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En caractéristique zéro, on peut en déduire hom = num (Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus, on peut munir hn(A) d’un accouplement hn(A) ⊗ hn(A) → 1(−n) dont la réalisation singulière est une polarisation (Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On en déduit que pour tout facteur direct M du motif homologique hn(A) la restriction de l’ac- couplement à M induit un isomorphisme M ∼= M ∨(−n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Ce fait est impliqué par la Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13 et c’est le point crucial où on l’utilise que les motifs sont définis 30 GIUSEPPE ANCONA sur C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ) L’autodualité pour les motifs homologiques se relève aussi dans CHM(C)ab par le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Passons maintenant à la conservativité et considérons le diagramme (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Encore par le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2, il suffira de démontrer la conservativité de la réalisation des motifs homologiques I : Mot(C)ab → GrVectQ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part, la conjecture hom = num déduite au début de la preuve dit que les catégories Mot(C)ab et NUM(C)ab coincident.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier, par le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1, I est un foncteur entre catégories semisimples, il est donc conservatif.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Dans la suite de la section on travaille sur un corps fini k = Fq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce qui rem- placera l’utilisation de la polarisation dans la preuve ci-dessus est la multiplication complexe, via les décompositions de l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Clozel [Clo99]) Soit A une variété abélienne sur un corps fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors il existe une infinité de nombres premiers ℓ tels que l’équivalence numérique coïncide avec l’équivalence homologique pour la cohomologie ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On se ramène au cas où A est simple.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient Q ⊂ F0 ⊂ �F les corps de nombres introduits dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12 et la Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13(2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Ces corps dépendent de A et plus précisément du choix d’un corps CM dans ses endomor- phismes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Fixons un nombre premier ℓ tel qu’il existe une place λ de F0 au-dessus de ℓ telle que la complétion (F0)λ ne contienne pas �F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On va montrer qu’un tel ℓ convient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons qu’il y a une infinité de tels ℓ et que l’on peut estimer leur densité avec Chebotareff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons les classes algébriques de codimension n que l’on voit comme classes dans h2n(A)(n) et soit q2n l’accouplement construit dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Puisqu’il est non dégénéré il suffit de voir que pour chaque classe algébrique non nulle γ il existe une classe algébrique δ telle que q2n(γ, δ) ̸= 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On vérifie que la question est stable par changement de coefficients et on travaille avec les cycles à coefficients dans F0 et la réalisation λ-adique Rλ à valeurs dans les (F0)λ-espaces vectoriels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas on utilise la décomposition de la Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13(2) en plans et droites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il suffit alors de travailler avec un seul de ces facteurs M et supposer que γ vit dans M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas, si la forme quadratique q2n est sans vecteur isotrope sur M le choix δ = γ convient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si M a dimension un cela suit du fait que q2n est non dégénérée sur chaque facteur de la décomposition et donc sur M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si M a dimension deux alors il admet au plus deux droites isotropes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces deux droites existent au moins sur M ⊗F0 �F : il s’agit de la décomposition en droites de l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Montrons que ces droites ne sont pas contenues dans le (F0)λ-espace vectoriel Rλ(M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour cela il suffira de construire un endomorphisme f : M → M dans la catégorie CHM(k)F0 dont ces droites sont des droites propres et de valeurs propres appartenant à �F − F0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela donnera la conclusion voulue puisqu’on a que (F0)λ ne contient pas �F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La construction de ce f procède ainsi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons la décomposition (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Quitte à changer la numérotation, le motif M est de la forme M = (L1 ⊗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ⊗ L2n) ⊕ (¯L1 ⊗ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ⊗ ¯L2n), où ¯· est la conjugaison complexe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fixons un ordre sur les Li, ceci permet de réaliser M comme facteur direct de h1(A)⊗2n dans la catégorie CHM(k)F0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors MÉMOIRE HDR 31 définir f par l’action induite par un générateur de F ⊂ End(A) ⊗ Q sur le premier terme du produit tensoriel et l’identité sur les autres 2n − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Proposition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La conjecture d’autodualité 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 est vraie pour les motifs de CHM(Fq)ab de poids pair et dont la réalisation a dimension un.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient X un tel motif et n son poids (pair).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On a une variété abélienne A, telle que X est facteur direct de hn(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On veut montrer que X ∼= X∨(−n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 il suffit de montrer πnum(X) ∼= πnum(X)∨(−n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par la semisimplicité de Jannsen, il suffit alors de montrer que Hom(πnum(X), πnum(X)∨(−n)) ̸= 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cet énoncé peut se démontrer après extension des scalaires.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On étend les scalaires au corps F0 de la Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13(2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilise la décomposition de cette même remarque.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par semisimplicité on peut supposer que X soit un facteur direct d’un des facteurs M de cette décomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Rappelons que l’on dispose d’un accouplement qn non-dégénéré sur M et que M a dimension un ou deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si M a dimension un alors X = M et on a terminé.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si M a dimension deux, remontons aux motifs homologiques, via le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On pourra alors utiliser la réalisation et il suffira de montrer que l’accouplement restreint à la droite qui est la réalisation de X reste non-dégénéré, autrement dit que la droite n’est pas isotrope.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (C’est ici que l’on utilisera que le poids est pair.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons notamment que si le poids est impair l’accouplement sur M est alterné et donc toute droite est isotrope.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La subtilité est que la catégorie des motifs homologiques dépend a priori de la cohomologie choisie mais la bonne nouvelle est qu’il suffit d’étudier une seule cohomologie bien choisie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilise alors la cohomologie λ-adique comme dans la preuve du Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4 avec le même choix de λ : on y avait montré que la réalisation de M est sans vecteurs isotropes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les foncteurs de réalisation ℓ-adique Rℓ : CHM(Fq)ab −→ GrVectQℓ sont conservatifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit f : X → Y une application dans CHM(Fq)ab telle que Rℓ(f) : Rℓ(X) → Rℓ(Y ) soit un isomorphisme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On veut montrer que f est un isomorphisme également.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 il suffira de travailler avec l’équivalence homologique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En utilisant la décomposition de Künneth, il suffira de supposer que Rℓ(X) et Rℓ(Y ) sont concentrés en un même degré cohomologique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons d’abord que Rℓ(X) et Rℓ(Y ) aient dimension un.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On dispose d’ap- plications 1 −→ Y ⊗ X∨ ∼= Y ∨ ⊗ X −→ 1, où la première et la dernière application sont obtenues par adjonction à partir de f et l’isomorphisme central vient de la Proposition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas les réalisations des applications ci-dessus sont des isomorphismes et par la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 on a 1 ∼= Y ⊗ X∨ donc X ∼= Y .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 32 GIUSEPPE ANCONA Travaillons maintenant dans le cas général : soit d la dimension de Rℓ(X) et Rℓ(Y ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que leur degré cohomologique soit pair (sinon il faudra remplacer des produits extérieurs par des produits symétriques dans la suite).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’application Λdf : ΛdX −→ ΛdY retombe dans le cas particulier de la dimension un traité au-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est donc un isomorphisme et on dispose de l’application (Λdf)−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut maintenant construire une application g : Y → X via Y ∼= ΛdY ⊗ (Λd−1Y )∨ −→ ΛdX ⊗ (Λd−1X)∨ ∼= X où le premier et le dernier isomorphisme viennent du lemme ci-dessous et l’applica- tion centrale est (Λdf)−1⊗(Λd−1f)∨.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par construction, Rℓ(g) est un isomorphisme, on conclut alors par la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Lemme 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ([O’S05, Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2]) Soit M un motif homologique dont la réalisa- tion est concentrée en un degré pair et de dimension d, alors M ∼= ΛdM ⊗ (Λd−1M)∨ Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le motif ΛdM est un facteur direct de M ⊗ Λd−1M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci fournit deux applications entre M et ΛdM ⊗ (Λd−1M)∨ dans les deux directions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Leur réalisation est un isomorphisme, on conclut par la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Positivité en caracteristique positive Dans cette section nous étudions la Conjecture de positivité 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le résultat principal dit que la conjecture est vérifiée pour les motifs de dimension 2 et à réduction supersingulière (Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous expliquons ensuite comment ap- pliquer ce résultat pour déduire la Conjecture standard de type Hodge 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15 pour certaines variétés, par exemples les variétés abéliennes de dimension quatre.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Puis nous discutons le rôle de l’hypothèse de dimension 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient M un motif homologique sur un corps k de caractéristique p et q : Sym2 M → 1 un morphisme dans Mot(k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que q soit la réduction modulo p d’une application �q : Sym2 � M → 1 définie en caractéristique zéro.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définissons qZ comme la restriction de q à toutes les classes algébriques Z(M) = Hom(1, M) de M et qB comme la réalisation singulière de �q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que qB soit une polarisation et supposons avoir M ∼= 1⊕2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors qZ est définie positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Sur les hypothèses : relevabilité et rang 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Il n’est pas rare d’avoir des motifs qui se relèvent à la caractéristique zéro.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple il est attendu que tout motif sur un corps fini se relève, car la conjecture de Tate prédit qu’un tel motif serait de type abélien.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En général, même si un motif se relève, ses classes algébriques ne se relèveront pas à la caractéristique zéro, ce qui rend les résultats de positivité difficiles, puisqu’ils ne peuvent pas se déduire des propriétés de positivité des polarisation, voir Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’hypothèse restrictive dans le théorème ci-dessus est la dimension deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La façon d’utiliser ce résultat pour déduire la conjecture standard de type Hodge pour certaines variétés est la suivante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On décompose le motif d’une variété donnée autant que possible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Certains facteurs ne posséderont pas de classes algébriques, d’autres MÉMOIRE HDR 33 en posséderont uniquement certaines pour lesquels la conjecture standard de type Hodge peut se déduire des cas connus : par exemple ce sont des classes qui se relèvent à la caractéristique zéro ou qui sont construites à partir de diviseurs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Enfin, ils resteront parfois des facteurs qui possèdent des classes algébriques qui ne se ramènent pas à des cas connus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le point est alors de trouver des variétés pour lesquels ces derniers facteurs sont de dimension deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un exemple est donné dans le corollaire ci-dessous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Corollaire 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit A une variété abélienne de dimension quatre définie sur un corps de caractéristique p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors (1) La Conjecture standard de type Hodge 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15 est vraie pour A, (2) Le produit d’intersection CH2(A)/num × CH2(A)/num −→ Q est de signature (ρ2 − ρ1 + 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ρ1 − 1), où ρn = dimQ(CH2(A)/num), (3) Il y a une infinité de nombres premiers ℓ ̸= p pour lesquels l’équivalence nu- mérique sur A coïncide avec l’équivalence homologique pour la cohomologie ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par un argument de spécialisation on peut supposer que le corps de définition est fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors utiliser la décomposition de la Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13(3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les facteurs qui sont a priori mystérieux pour la Conjecture standard de type Hodge 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15 sont ceux de type (b), dans la notation de la même remarque.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il se trouve que tous ces facteurs de toutes les variétés abéliennes de dimension quatre sont bien de dimension deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce fait est un petit miracle combinatoire, voir [Anc21, §7] pour les détails ou la Remarque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13(3) pour un aperçu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les points (1) et (2) sont en fait équivalents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette équivalence n’utilise pas le fait que A est une variété abélienne mais uniquement la dimension quatre.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle se déduit de la décomposition en parties primitives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si le corps de définition est fini le point (3) est un cas particulier du Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4 de Clozel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour se ramener aux corps finis on spécialise et on utilise la Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Dimension supérieure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Pour les variétés abéliennes de dimension quelconque on pourra encore utiliser la décomposition de la Re- marque 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13(3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En général les facteurs (b) de la remarque auront dimension plus grande que deux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut tout de même trouver des exemples spo- radiques pour lesquels ces facteurs de type (b) ont rang 2 et déduire la conjecture standard de type Hodge à l’aide du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci a été récemment étudié par Koshikawa [Kos22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Supersingularité vs Frobenius invariant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') On remarquera un petit décalage entre l’hypothèse de supersingularité M ∼= 1⊕2 du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 et la caractérisation des facteurs (b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tout d’abord remarquons que ces deux descriptions sont équivalentes sous la conjecture de Tate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inconditionnellement, a priori, parmi les facteurs de type (b) certains pourraient ne pas posséder de classe algébrique : ces facteurs pourront être négligés à l’étude de la conjecture standard de type Hodge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour les autres on a besoin de montrer que dès qu’un facteur a une classe algébrique il est 34 GIUSEPPE ANCONA engendré par des classes algébriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci se montre en utilisant l’action CM mais l’argument nécessite de travailler avec l’équivalence numérique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si on travaillait avec l’équivalence homologique on se trouverait devant des problèmes similaires à ceux qui empêchent l’argument de Clozel du Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4 de fonctionner pour tout nombre premier ℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Symbole de Hilbert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soit q une Q-forme quadratique de dimen- sion deux et ν = 2, 3, 5, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', ∞ une place de Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On définit le symbole de Hilbert εν(q) de q en ν comme étant +1 si q(x, y) = z2 a une solution non-nulle dans la complétion Qν et −1 sinon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Gardons les notations du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 et soit n l’unique entier tel que la réalisation singulière de � M soit une structure de Hodge de type (−n, +n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors qZ est définie positive si et seulement si εp(qZ) = (−1)nεp(qB), (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) ce qui est encore équivalent au fait que qZ ⊗ Qp est isomorphe à qB ⊗ Qp si et seulement si n est pair.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’idée de la preuve de cette proposition a déjà été introduite dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9 et la remarque qui le suit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette proposition est par ailleurs le point crucial où l’hypothèse de la dimension deux est nécessaire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tout d’abord on remarque que, pour tout nombre premier ℓ ̸= p, on a qB ⊗ Qℓ ∼= qZ ⊗ Qℓ, c’est la combinaison du théorème de comparaison d’Artin et du changement de base propre et lisse en cohomologie ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela implique en particulier que εℓ(qZ) = εℓ(qB), mais aussi que le discriminant de qB et qZ coïncident dans Q∗/(Q∗)2, car un nombre rationnel est un carré s’il l’est dans presque toute complétion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part, en suivant la Définition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6, on a que qB est (−1)n-définie positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela implique en particulier que ε∞(qB) = (−1)n, et que le discriminant de qB est positif.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On en déduit que qZ a discriminant positif et donc, puisqu’on est en rang deux, que qZ est définie positive ou définie négative.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La positivité de qZ est équivalente alors à ε∞(qZ) = +1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A l’aide de la formule du produit sur les symboles de Hilbert on a � ν εν(qZ) = 1 = � ν εν(qB).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) En simplifiant les facteurs ℓ-adiques on obtient ε∞(qZ)·εp(qZ) = ε∞(qB)·εp(qB) et donc ε∞(qZ)·εp(qZ) = (−1)n ·εp(qB).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On conclut que ε∞(qZ) = +1 si et seulement si la formule (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) est satisfaite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La dernière équivalence de la proposition suit du fait que deux Qp-formes qua- dratiques non-dégénérées de même rang sont isomorphes si et seulement si elles ont le même discriminant et le même symbol de Hilbert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Démonstration du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par la proposition précédente on est ramené à un problème purement p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce dernier a en plus une interprétation cohomologique qui permet de le traduire en une question de théorie de Hodge p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour expliquer cette traduction définissons VB,p comme la réalisation étale p-adique de � M et VZ,p comme la partie Frobenius invariante de la réalisation cristalline de M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 35 Chacun de ces deux Qp-espaces vectoriels de dimension deux est muni d’une forme quadratique induite respectivement par �q et q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces deux formes quadratiques ne sont rien d’autre que qB ⊗Qp et qZ ⊗Qp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La première identification suit du théorème de comparaison d’Artin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La deuxième vient du fait que la partie Frobenius invariante contient toujours l’espace engendré par les classes algébriques et dans ce cas cette inclusion est une égalité par dimension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le théorème de comparaison p-adique montré par Faltings fournit un isomor- phisme VB,p ⊗ Bcris = VZ,p ⊗ Bcris (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) fonctoriel, compatible à toutes les structures que l’on pourrait imaginer et en par- ticulier avec les formes quadratiques qB et qZ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On conclut à l’aide des deux phéno- mènes suivants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (a) La matrice de changement de base est calculable dans Mat2×2(Bcris).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elle ne dépend que de l’entier n de la filtration de Hodge et de l’algèbre End(VB,p) des endomorphismes de VB,p comme représentation galoisienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (b) La description de cette matrice est suffisante pour déduire que qB ⊗ Qp et qZ ⊗ Qp sont isomorphes si et seulement si n est pair.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ On passe maintenant à la description des deux phénomènes (a) et (b) de la fin de la preuve ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si le deuxième est élémentaire le premier est une propriété remarquable de la théorie de Hodge p-adique qui la différencie de la théorie de Hodge classique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons avoir deux R-formes quadratiques définies et de dimen- sion deux q1(x, y) = a1x2 + b1y2 et q2(x, y) = a2x2 + b2y2 et une identification q1 ⊗R C = q2 ⊗R C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons savoir que la matrice de changement de base de l’identification est �i 0 0 i � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On pourra alors en déduire que q1 et q2 ne sont pas isomorphes (une est définie négative et l’autre est définie positive).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les arguments au point (b) dans la preuve du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 sont tout aussi élémentaires et ressemblent à cet exemple, avec R et C qui sont remplacés par Qp et Bcris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Théorème de comparaison p-adique vs classique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Le théorème de comparaison p-adique Rp ⊗ Bcris = Rcris ⊗ Bcris, (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) dont (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) en est une instance, est souvent considéré comme l’analogue p-adique du théorème de comparaison entre cohomologie singulière et cohomologie de de Rham algébrique RB ⊗ C = RdR ⊗ C (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) que l’on a discuté dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En fait le théorème de comparaison p- adique a des avantages par rapport à sa version classique, que l’on liste ci-dessous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (C’est grâce à ces propriétés que l’on peut notamment calculer certaines matrices de changement de base et déterminer des relations entre leurs entrées, cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' le point (a) de la preuve du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') 36 GIUSEPPE ANCONA Rappelons que les réalisations classiques d’un motif M possèdent des structures supplémentaires.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier, RdR(M) est munie d’une filtration, RB(M) est munie d’une structure de Hodge, Rp(M) est munie de l’action du groupe de Galois d’un corps p-adique et Rcris(M) est un ϕ-module filtré, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' elle est munie d’une filtration et d’une action du Frobenius absolu ϕ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Rappelons aussi que les coefficients des matrices qui apparaissent dans les com- paraisons (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) ou (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) sont appelés périodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) L’anneau Bcris est muni des actions du Frobenius absolu ϕ et du groupe de Galois absolu de Qp ainsi que d’une filtration 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La comparaison (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) respecte ces trois structures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Quand on dispose d’objets cohomologiques suffisamment concrets, comme ceux de (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3), on peut explicitement décrire ces structures sur les périodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le corps C en revanche n’est pas muni de structures qui imiterait la filtra- tion ou la structure de Hodge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Connaître la structure de Hodge sous-jacente à RB(M) n’aide pas à avoir des informations sur les périodes complexes du motif M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Dans [Fal89], Faltings montre une équivalence de catégories entre certains ϕ-modules filtrés, dits admissibles, et certaines représentations de groupes de Galois de corps p-adiques, dites cristallines.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La condition d’admissibi- lité est toujours vérifiée par les modules d’origine géométrique, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' par les réalisations de motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette équivalence est en plus compatible au théorème de comparaison (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier, des périodes dans Bcris qui ont un certain comporte- ment par rapport à ϕ et à la filtration doivent correspondre à un unique ϕ-module filtré et donc aussi à une unique représentation de Galois.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Autre- ment dit, les périodes p-adique associées à un motif donné sont caractérisées par leur comportement par rapport à deux structures : Frobenius et filtra- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une structure de Hodge est beaucoup plus riche qu’une filtration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple, pour deux variétés d’une même famille, les espaces vectoriels filtrés correspondant seront isomorphes alors que les structures de Hodge ne le seront pas, en général 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On ne peut pas avoir une équivalence de catégories entre ces deux structures cohomologiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (De façon informelle, le passage du cas complexe (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) au cas p-adique (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) correspond à enrichir la structure de de Rham et à réduire celle de Hodge suffisamment pour avoir deux structures équivalentes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet, la réalisation cristalline hérite la filtration de de Rham mais elle possède en plus l’action du Frobenius absolu ϕ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part, d’après la conjecture de Mumford-Tate, une structure de Hodge est grosso-modo équivalente à une représentation de Galois d’un corps de nombres, or sur Rp(M) on ne regarde que l’action d’un certain de ses sous-groupes de décomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Certaines structures sont définies dans un plus gros anneau noté BdR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On ignore ici ce point pour simplifier l’exposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si on considère la courbe elliptique Et : y2 = x(x − 1)(x − t) pour t ∈ Q − {0, 1} et le motif Mt = h1(E) alors RdR(Mt) est la donnée d’un Q-espace vectoriel de dimension 2 muni d’une droite à l’intérieur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En revanche la structure de Hodge RB(Mt) détermine Et à isogénie près, notamment il y aura des structures de Hodge CM et d’autres qui ne le sont pas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 37 (3) La condition d’admissibilité, discutée au point précédent, se trouve être re- lativement élémentaire à vérifier, grâce à [CF00].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci permet de construire facilement des ϕ-modules filtrés admissibles et donc des matrices de pé- riodes avec action de Frobenius et filtration prescrites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On ne dispose pas de méthode élémentaire de construction de périodes complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il s’agit d’intégrales de formes différentielles algébriques qui sont souvent difficiles à calculer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Leurs relations sont prédites par la conjectures des périodes de Grothendieck et restent mystérieuses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En résumant, le point (a) de la preuve du Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 est le calcul de la matrice de périodes associée à (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce calcul procède comme suit : on décrit l’action de ϕ et la filtration sur ces pédiodes (point (1) ci-dessus), puis on montre que cette description caractérise les périodes en question (point (2)), enfin on construit de telles périodes (point (3)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Un calcul de périodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soit M un motif comme dans le Théorème 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 pour lequel on veut montrer la relation (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela passe par le calcul de la ma- trice de périodes de (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce calcul dépend de l’entier n (défini dans la Proposition 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) et de l’algèbre End(VB,p) des endomorphismes de VB,p comme représentation galoisienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons n = 1 : on doit alors montrer que qZ ⊗ Qp et qB ⊗ Qp ne sont pas isomorphes, voir la Proposition 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons également que End(VB,p) soit le corps Qp2, l’unique extension non ramifiée de degré 2 de Qp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Comme Bcris contient toutes les extensions non ramifiées, on peut alors écrire (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) comme (VB,p ⊗Qp Qp2) ⊗Qp2 Bcris = (VZ,p ⊗Qp Qp2) ⊗Qp2 Bcris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) L’action de End(VB,p) = Qp2 sur VB,p décompose VB,p ⊗Qp Qp2 en deux droites échangées par le groupe de Galois Gal(Qp2/Qp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier on peut choisir deux vecteurs vB et wB échangés par le groupe de Galois et appartenant à ces droites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’algèbre Qp2 agit également sur VZ,p, grâce à l’équivalence de catégories de Faltings, voir aussi le point (2) de la Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors construire vZ et wZ de façon analogue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus, comme (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) est compatible à cette action, il existe deux périodes α, β ∈ Bcris telles que αvB = vZ et βvB = vZ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je prétends qu’elles satisfont aux relations ϕ(α) = β et ϕ(β) = α (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7) ainsi que α ∈ Fil1 − Fil2 et β ∈ Fil−1 − Fil0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8) En effet Gal(Qp2/Qp) est engendré par le Frobenius et d’autre part l’action du Frobenius sur VB,p et VZ,p est triviale par définition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Quant à la relation sur la filtration, on la déduit du fait que les droites propres que l’on a construites doivent être isotropes, or la droite définie par la filtration sur la réalisation cristalline de M doit aussi l’être, en particulier elle doit coïncider avec une de ces droites propres (après extension des scalaires).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les relations (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7) et (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8) sont un exemple du principe (1) expliqué dans la Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 38 GIUSEPPE ANCONA Considérons maintenant le Qp2 espace vectoriel P = {γ ∈ Bcris, ϕ2(γ) = γ, γ ∈ Fil1 − Fil2 et ϕ(γ) ∈ Fil−1 − Fil0}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Je prétends qu’il a dimension 1, autrement dit que n’importe quel élément de P est en fait une période de M construite ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci est un exemple du principe (2) expliqué dans la Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le montrer remarquons d’abord qu’une période α de M est inversible dans Bcris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors considérer l’espace P/α ⊂ Bcris, il correspondra à α−1P = {λ ∈ Bcris, ϕ2(λ) = λ, λ ∈ Fil0 − Fil−1 et ϕ(λ) ∈ Fil0 − Fil−1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or cet espace est Qp2 ⊂ Bcris par [Fon94, Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour conclure construisons une période t avec les propriétés t, ϕ(t) ∈ B∗ cris, ϕ2(t) = p · t, t ∈ Fil1 − Fil2 et ϕ(t) ∈ Fil0 − Fil1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De ces propriétés on déduit que α = t/ϕ(t) ∈ P, autrement dit α est une période du motif M, et α · ϕ(α) = 1/p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette dernière relation implique que qZ ⊗ Qp et qB ⊗ Qp ne sont pas isomorphes par un calcul élémentaire qui est analogue à celui de l’Exemple 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La construction de t suit le principe (3) expliqué dans la Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Consi- dérons le ϕ-module filtré N = Q2 p muni du Frobenius ϕ = � 0 1/p 1 0 � et de la filtration Fil−1 = N, Fil0 = Qp · e2, Fil1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On vérifie que c’est un ϕ-module admissible, donc il existe une représentation de Galois V qui lui correspond par l’équivalence de catégorie de Faltings et qui donne une comparaison V ⊗ Bcris = N ⊗ Bcris = B2 cris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Puisque cette identification est compatible à toutes les structures, on voit que les vecteurs de V ⊂ B2 cris sont exactement de la forme (t, ϕ(t)) où t satisfait aux propriétés voulues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Généralisations possibles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La partie p-adique de l’argument présenté se généralise aux motifs de dimension plus grande : les principes généraux expliqués dans la Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9 restent valables, les calculs de l’Exemple 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10 de- viennent plus compliqués mais peuvent être traités.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans un travail avec Adriano Marmora [AM22] nous avons pu en déduire une généralisation de la formule (6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En suivant l’argument de la Proposition 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 cela donne la positivité du symbole de Hilbert à l’infini ε∞(qZ) = +1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Malheureusement cette information ne suffit pas à déduire que qZ est défini positive, c’est le point crucial où l’on utilisait l’hypothèse de dimension 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour passer à la dimension supérieure il faudrait trouver un invariant défini en toute place, tel que la place à l’infini soit contrôlée par toutes les places finies et d’autre part tel que l’invariant à l’infini détermine toute la signature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une tenta- tive pourrait passer par la cohomologie galoisienne : les k-formes quadratiques non dégénérées et de rang donné sont en bijection avec H1(k, O) où O est un k-groupe MÉMOIRE HDR 39 orthogonal de rang convenable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’application H1(Q, O) −→ � ν H1(Qν, O).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' est injective, c’est le théorème d’Hasse–Minkowski.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Son défaut de surjectivité est justement contrôlé par la formule du produit des symboles de Hilbert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le problème déjà soulevé se reformule alors ainsi : l’application H1(Q, O) −→ � ν̸=∞ H1(Qν, O) n’est plus injective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors essayer d’exploiter plus d’informations géométriques de notre si- tuation et faire surgir des groupes plus petits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple les motifs qui appa- raissent dans le problème sont munis non seulement d’une forme quadratique mais aussi de l’action d’un corps CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ajouter cette donnée au problème correspond à étudier la cohomologie galoisienne d’un tore maximal T du groupe orthogonal O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On dispose encore d’un principe local-global : l’application H1(Q, T ) −→ � ν H1(Qν, T ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' est injective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le point crucial serait alors d’avoir l’injectivité aussi de l’application H1(Q, T ) −→ � ν̸=∞ H1(Qν, T ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Malheureusement elle n’est pas injective : on peut calculer son noyau à l’aide de la suite exacte de Poitou–Tate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Périodes p-adiques à la André Dans cette section nous présentons un travail en collaboration avec Dragos Fra- tila.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les motivations sont d’origine géométrique - l’étude des classes algébriques en caractéristique p - mais le résultat final est plutôt arithmétique : on construit une algèbre de périodes p-adiques ainsi qu’un cadre tannakien pour l’étudier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Des telles périodes devraient avoir des analogies avec les périodes complexes que l’on a rencontré dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les périodes p-adiques de Fontaine possèdent des propriétés cohomologiques analogues à celles des périodes complexes et même plus fortes (Remarque 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par contre les propriétés arithmétiques des périodes complexes, comme leur transcendence ou leur lien avec les fonctions spéciales, n’ont pas de bon analogue dans les périodes p-adiques de Fontaine [And90].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pendant que notre travail avançait nous avons découvert qu’André avait tissé des liens similaires [And95, And03].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Son travail nous a été utile pour raffiner notre étude et notamment pour formuler la condition de ramification (Définition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motivation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le point de départ vient d’une remarque de Tate : la conjecture de Tate prédit l’existence de classes algébriques mais elle ne prédit pas quelle classe cohomologique est algébrique [Mil07, Aside 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une façon d’interpréter cette re- marque est que l’on a une description du Qℓ-espace vectoriel engendré par les classes algébrique mais on n’a pas de description du Q-espace vectoriel engendré par ces dernières.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est un point délicat qui est présent dès le travail de Tate sur la conjec- ture de Tate pour les diviseurs sur les variétés abéliennes sur un corps fini [Tat66], 40 GIUSEPPE ANCONA et plus récemment dans le travail de Charles sur la conjecture de Tate pour les diviseurs sur les surfaces K3 [Cha13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un exemple élémentaire qui illustre cette subtilité est le suivant : il existe des Qℓ-droites dans la cohomologie ℓ-adique d’une variété X, disons définie sur un corps fini, qui sont Galois invariantes et pourtant elles ne contiennent pas d’élément du Q-espace vectoriel Im clX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour construire de tels exemples prenons X une surface abélienne, ou une K3, et fixons deux classes de diviseurs α et β linéairement indé- pendantes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Prenons maintenant une constante c ∈ Qℓ−Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors la droite engendrée par α + cβ convient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le montrer on peut utiliser le produit d’intersection et le fait qu’il est défini à coefficients rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un échec : cas ℓ-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Faute de savoir décrire le Q-espace vectoriel des classes algébriques, un premier pas est de le comparer à un autre Q-espace vectoriel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est notamment ce que l’on a fait dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9 et la remarque qui le suit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Prenons une variété Xp définie sur Fp et supposons qu’elle se relève à une va- riété X définie sur Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Au moyen d’un plongement σ : Q ֒→ C on dispose d’une identification H∗ B(X(C), Q) ⊗ Qℓ = H∗ ℓ (Xp) (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) induite par le théorème de comparaison d’Artin et le changement de base propre et lisse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sous cette identification on peut étudier la position du Q-espace vectoriel Zp des classes algébriques sur Xp par rapport à H∗ B(X(C), Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le premier fait que l’on remarque est que l’intersection Zp ∩H∗ B(X(C), Q) contient le Q-espace vectoriel Z0 des classes algébriques sur X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inspiré par les différentes versions de la conjectures des périodes de Grothendieck on peut se demander si cette inclusion est en fait une égalité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Après avoir montré que cette question a réponse affirmative dans certains cas, nous avons compris que la réponse est négative en général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les cas affirmatifs sont les surfaces à rang de Picard maximal - par une méthode similaire celle présentée dans l’Exemple 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9 - et les variétés abéliennes CM [AF22, §10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il est possible de construire des contre-exemples avec le carré d’une courbe elliptique non CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’ar- gument suit en fait la technique qui a permis à André de montrer que l’analogue de la conjecture des périodes de Grothendieck est fausse pour le théorème de com- paraison p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) dépend du choix de σ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut faire varier σ en utilisant le groupe de Galois absolu du corps de nombre sur lequel X est défini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans certains cas on sait que l’action de ce groupe de Galois sur H∗ ℓ (Xp) est hautement non triviale [Ser72], ce qui permet faire varier le Q-espace vectoriel H∗ B(X(C), Q) et notamment de le faire rencontrer Zp de façon inattendue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cas p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fixons une fois pour toutes un plongement Q ⊂ Qp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Prenons comme auparavant une variété X définie sur Q à bonne réduction et notons Xp sa réduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On dispose de la comparaison entre cohomologie de de Rham et co- homologie cristalline H∗ dR(X, Q) ⊗ Qp = H∗ cris(Xp, Qp) due à Berthelot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notons par Zp le Q-espace vectoriel des classes algébriques sur Xp et par Z0 le Q-espace vectoriel des classes algébriques sur X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On a comme dans le cas ℓ-adique l’inclusion Z0 ⊂ Zp ∩ H∗ dR(X, Q) MÉMOIRE HDR 41 et on peut encore une fois se demander si cette inclusion est en fait une égalité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (pGPCw : Analogue p-adique de la version faible de la conjecture des périodes de Grothendieck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Est-ce que l’inclusion Z0 ⊂ Zp ∩ H∗ dR(X, Q) est une égalité ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour rendre cette question raisonnable il est nécessaire d’imposer une condition de ramification que l’on discutera plus tard et que l’on ignore pour l’instant (Défini- tion 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette question apparaît comme l’analogue p-adique de la version faible de la conjecture des périodes de Grothendieck (appelée parfois conjecture de de Rham–Betti [And04, §7]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La remarque ci dessous fait le lien entre cette conjecture et différentes conjectures classiques sur les cycles algébriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tout comme son pendant classique, cette conjecture prédit de la transcendence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet elle prédit qu’une classe algébrique en caractéristique p qui n’est pas re- levable à la caractéristique zéro ne peut pas être dans H∗ dR(X, Q), autrement dit, au moins une de ses coordonnées par rapport à une base de H∗ dR(X, Q) doit être transcendante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La version forte de la conjecture de Grothendieck p-adique prédira de façon précise le degré de transcendance de toutes ces coordonnées (Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (pGPCw vs conjectures classiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Comparons maintenant la question qui a été soulevée au paragraphe précédent, notée (pGPCw), avec trois conjectures classiques que nous rappelons de façon informelle (voir [And04, §7] pour plus de détails).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces trois conjectures sont la conjecture de Hodge (HC), la version faible de la conjecture des périodes de Grothendieck (GPCw) et la conjecture de Hodge variationnelle p-adique de Fontaine et Messing (pHC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (HC) Une classe rationnelle en cohomologie singulière est algébrique si et seule- ment si elle appartient au bon degré de la filtration de de Rham.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (GPCw) Une classe rationnelle en cohomologie de de Rham est algébrique si et seulement si elle est rationnelle pour la cohomologie singulière (pHC) Une classe algébrique en cohomologie cristalline se relève à la caractéris- tique zéro si et seulement si elle appartient au bon degré de la filtration de de Rham.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De façon informelle, on peut voir (pGPCw) comme « le produit fibré de (pHC) et (GPCw) au-dessus de (HC) ».' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (pGPCw) → (GPCw) ↓ ↓ (pHC) → (HC) La partie droite du diagramme concerne la caractéristique zéro, celle de gauche la caractéristique mixte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La conjectures du bas comparent une structure rationnelle et une filtration, celles du haut comparent deux structures rationnelles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Périodes p-adiques à la André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soit M ∈ Mot(Q) un motif homologique 16 à bonne réduction et soit Mp ∈ Mot(Fp) sa réduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notons leurs 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tout comme dans le cas classique on travaillera qu’avec des motifs vérifiant hom = num, par exemple les motifs issues de produits de courbes elliptiques : ceci est nécessaire pour avoir des catégories tannakiennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 42 GIUSEPPE ANCONA classes algébriques par Z0(M) = HomMot(Q)(1, M) et Zp(M) = HomMot(Fp)(1, Mp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons le théorème de comparaison de Berthelot RdR(M) ⊗Q Qp = Rcris(Mp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) Pour tout choix de base B de Zp(M) et B′ de RdR(M) définissons MatB,B′(M) comme la matrice ayant comme vecteurs colonnes les coordonnées de B par rapport à B′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous appelons les coefficients de cette matrice les périodes p-adique d’André de M et définissons Pp(M) ⊂ Qp comme la Q-algèbre engendrée par ces périodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Périodes classiques vs périodes p-adiques à la André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') (1) Si un élément de B est une classe algébrique qui n’est pas relevable à la caractéristique zéro au moins une de ses périodes devrait être transcendante par la Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) La matrice MatB,B′(M) dépend bien du choix des bases B et B′, par contre l’algèbre Pp(M) n’en dépend pas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) Pour que l’espace Zp(M) ne soit pas réduit à zéro il faut que M contienne des facteurs directs de poids zéro.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il faut typiquement imaginer M = h2n(X)(n) pour une variété X à bonne réduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus, pour avoir des périodes intéressantes, il faut que M admette des classes algébriques mo- dulo p qui ne sont pas relevables, sinon toutes les périodes p-adiques seraient algébriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (4) Contrairement au cas classique, la matrice de périodes p-adique n’est pas carrée, en effet l’inégalité 17 #B′ ≤ #B est stricte en général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (5) Il est possible de définir les périodes p-adiques pour les motifs mixtes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il faut dans ce cas considérer uniquement les motifs mixtes vérifiant (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) - cette relation n’est automatique pour les variétés ouverte.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Courbes elliptiques CM et valeurs Gamma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Considérons E une courbe elliptique CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les périodes complexes de son h1(E) sont un produit de certaines valeurs spéciales de la fonction gamma ΓC en certains rationnels explicites dépendant uniquement du corps CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il n’y a pas de pé- riode p-adique associée à h1(E), puisqu’il n’y a pas de classe algébrique dans le h1(E), par contre on peut considérer le motif M = h1(E) ⊗ h1(E)∨.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ses classes algébriques correspondent aux endomorphismes de E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour avoir des périodes p-adiques intéressantes considérons un premier p à réduction su- persingulière, autrement tous les endomorphismes se relèveraient et toutes les périodes seraient algébriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il s’agit de décrire l’action de ces endo- morphismes par rapport à une base de H1 dR(E, Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce genre de calculs a été traités par Coleman et Ogus [Col90, Ogu90].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On y voit apparaître des produits de valeurs spéciales de la fonction gamma p-adique Γp en des rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Point technique : cette inégalité a encore besoin de l’hypothèse hom = num.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilise le fait que l’équivalence numérique commute à l’extension des scalaires, voir la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='18 et la remarque qui la suit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 43 (2) (Motifs de Kummer et logarithme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Considérons le motif de Tate mixte de type Kummer Ka = h1(Gm, {1, a})∨, avec a ∈ Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce motif s’insère dans une suite exacte 0 −→ 1(+1) −→ Ka −→ 1 −→ 0 qui est non scindée pour a ̸= 0, 1, −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier ce motif n’a pas de classe algébrique non nulle, qui est la raison d’avoir considéré Ka et non pas son dual h1(Gm, {1, a}).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sa matrice de périodes complexes est �2πi log(a) 0 1 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fixons un nombre premier p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour a ̸≡ 0, 1 [p], le motif Ka a bonne réduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus les motifs de Tate sur un corps fini forment une catégorie semisimple, en particulier la suite exacte ci-dessus se scinde modulo p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On en déduit que le motif possède une classe algébrique non nulle modulo p qui est donc non relevable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sa matrice de périodes p-adiques est �logp(a) 1 � dont l’analogie avec son pendant complexe est encore une fois frappante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette matrice s’obtient à partir du calcul de la matrice du Frobenius agis- sant sur RdR(Ka) qui est dû à Deligne [Del89, §2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On y voit apparaître le logp(a1−p) et on trouve curieux que le passage de la matrice de Frobenius à la matrice de périodes corrige cet exposant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (La correction de logp(a1−p) à logp(a) aurait pu s’obtenir en changeant de base, or pour ces motifs on dis- pose de bases canoniques et toutes les matrices décrites ci-dessus utilisent uniquement ces bases).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) (Fonctions hypergéométriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soient M et N deux motifs non isomorphes mais dont les réductions modulo p le sont.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors le motif M ⊗ N ∨ a une classe algébrique modulo p non relevable qui est justement associée à cet isomorphisme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ses périodes p-adiques sont les coordonées de RdR(N) par rapport à RdR(M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par exemple on peut considérer M = h1(E) et N = h1(E′) où E et E′ sont deux relèvements non isogènes d’une courbe elliptique ordinaire sur un corps fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut notamment choisir E comme le relèvement ca- nonique de Serre–Tate et E′ comme une courbe elliptique non CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les périodes p-adiques qui apparaissent dans ce cas là sont décrites par Katz [Kat80].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On y voit notamment apparaître des valeurs spéciales de fonctions hypergéométriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (4) (Matrice du Frobenius.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Soit f ∈ End(Mp) un endomorphisme modulo p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut considérer la matrice de son action par rapport à une base de RdR(M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ses coefficients pourront s’interpréter comme périodes p-adiques à la André en regardant f comme une classe algébrique modulo p du motif M ⊗ M ∨.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’intérêt particulier est le cas où f est le Frobenius : certains auteurs [Fur07, Bro17] ont définis les périodes p-adique associés à M comme ses coefficients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 44 GIUSEPPE ANCONA Le point de vue des périodes p-adiques à la André est meilleure pour plusieurs raisons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Entre autres, il donne des bornes plus fine à la trans- cendance ainsi qu’une interprétation motivique de certaines relations natu- relles, comme celles provenant du polynôme caractéristique du Frobenius, voir [AF22, Remark 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Transcendance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Comme expliqué dans la Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4, la Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 prédit la transcendance de certaines périodes p-adiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le prochain but est de donner une borne au degré de transcendance de ces périodes (Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) ainsi qu’une conjecture qui prédira ce degré (Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On montrera que cette dernière conjecture implique en fait la Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1, voir la Proposition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Gardons les notations de la Définition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons les catégories tanna- kiennes ⟨M⟩ et ⟨Mp⟩ engendrées par M et Mp ainsi que les groupes tannakiens GdR(M) et Gcris(Mp) associés aux foncteurs fibres RdR et Rcris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans le cadre des périodes classiques, Grothendieck démontre que leur degré de transcendence est borné par la dimension de GdR(M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notre résultat principal en est l’analogue p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le degré de transcendance des périodes p-adiques vérifie l’inégalité degtrPp(M) ≤ dim GdR(M) − dim Gcris(Mp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Transcendance classique vs transcendance p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') (1) Cette inégalité pourrait sembler plus forte que celle du cas complexe mais la matrice rectangulaire des périodes p-adiques est en général plus petite que celle des périodes complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Elles ont la même taille uniquement dans le cas de réduction supersingulière ce qui revient à Gcris(Mp) = {1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Dans le cadre complexe, le point crucial est d’interpréter les périodes comme les coordonnées d’un C-point du foncteur T (M) = Isom⊗ ⟨M⟩(RB, RdR).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par la théorie tannakienne ce foncteur est représentable par une variété affine sur Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’action naturelle de GdR(M) = Aut⊗ ⟨M⟩(RdR) sur T (M) rend cette variété un torseur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans le cas p-adique le foncteur Zp n’est pas un foncteur fibre, pour des questions de dimension, ce qui le fait sortir du cadre tannakien.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est tout de même un foncteur lax-monoïdal (ce qui revient à dire que le produit de classes algébriques est une classe algébrique).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le foncteur Isom⊗ ⟨M⟩(Zp, RdR) est vide en général, encore pour des rai- sons de dimension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut en revanche considérer les transformations na- turelles tensorielles ou les plongements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’inclusion de foncteurs Emb⊗ ⟨M⟩(Zp, RdR) ⊆ Nat⊗ ⟨M⟩(Zp, RdR) est une égalité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces foncteurs sont représentables par une variété H(M) affine sur Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’action naturelle de GdR(M) sur H(M) est transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus on a un iso- morphisme H(M)Qp = GdR(M)Qp/Gcris(Mp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 45 Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 implique Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') En analogie avec le cas classique on peut interpréter Zp ⊗ Qp ֒→ Rcris = RdR ⊗ Qp (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) comme un Qp-point de H(M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’évaluation en ce point donne un morphisme d’al- gèbres eval : O(H(M)) −→ Qp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) Par construction, l’image de ce morphisme est l’algèbre Pp(⟨M⟩) engendrée par toutes les périodes p-adiques de tous les motifs appartenant à la catégorie ⟨M⟩.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Contrairement au cas classique, ces périodes contiennent strictement celles de M, en général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela vient du fait que l’inclusion Zp(M)⊗n ⊂ Zp(M ⊗n) est stricte en général : c’est le défaut d’une formule de Künneth pour les classes algébriques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8 et les relations Pp(M) ⊂ Pp(⟨M⟩) = Im eval ⊂ Qp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) prouvent le Théorème 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 et même l’inégalité plus forte degtrPp(M) ≤ degtrPp(⟨M⟩) ≤ dim H(M) = GdR(M) − Gcris(Mp).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) La définition suivante est inspirée de travaux d’André [And95, And03].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Condition de ramification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') On dit qu’un motif N est CM si End(N) est un corps de nombres tel que dimQ End(N) = dim RdR(N).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On dit que le nombre premier p ne ramifie pas dans ⟨M⟩ si, pour tout N dans ⟨M⟩ qui est CM, le nombre premier p ne ramifie pas dans le corps de nombre End(N).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exemple 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit A une variété abélienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans le cas où N = h1(A), imposer que End(N) soit un corps de nombres tel que dimQ End(N) = dim RdR(N) revient à demander que A soit simple et CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas demander que p ne ramifie pas dans End(N) correspond à demander que p ne ramifie pas dans son corps CM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour un motif M ′ fixé, les nombres premiers qui ramifient dans ⟨M ′⟩ sont en nombre fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (p-GPCs : Analogue p-adique de la version forte de la conjecture des périodes de Grothendieck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Si p ne ramifie pas dans ⟨M⟩ alors l’application d’évaluation (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) est injective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De façon équivalente, l’espace homogène H(M) est connexe et le Qp-point de H(M) induit par (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) vit au-dessus du point générique de H(M) (ou encore l’inégalité degtrPp(⟨M⟩) ≤ dim H(M) = GdR(M) − Gcris(Mp) est en fait une égalité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Proposition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si M vérifie la Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13 alors pour tout N dans ⟨M⟩ on a une Z0(N) = Zp(N) ∩ RdR(N), c’est-à-dire la Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 (pGPCw) a réponse affirmative pour N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 46 GIUSEPPE ANCONA Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) Les preuves des résultats de cette section utilisent des techniques tannakiennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Comme déjà mentionné, notamment dans la Re- marque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7(2), on ne peut pas utiliser les résultats classiques tel quels mais il faut plutôt adapter leurs preuves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) La condition de ramification (Définition 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10) est inspirée d’une condition qu’André a imposée dans l’étude de ce genre de questions pour les variétés abéliennes à réduction supersingulière.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il avait remarqué que ces variétés pouvaient avoir des périodes p-adiques vérifiant des relations algébriques non motiviques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Comme mentionné dans l’Exemple 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5(1), les périodes qui apparaissent pour de tels motifs sont liées aux valeurs spéciales de la fonc- tion Gamma p-adique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ces dernières se trouvent être plus souvent algé- briques que leurs analogues complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) Assez peu est connu sur la conjecture classique des périodes de Grothen- dieck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La version forte a été démontrée pour les courbes elliptiques CM par Chudnovsky [Chu80].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le cas particulier de la courbe elliptique de Fermat implique notamment la transcendance de ΓC(1/3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La version faible (voir (GPCw) de la Remarque 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) a été montrée pour les diviseurs sur les varié- tés abéliennes et sur les surfaces K3 par Bost et Charles [BC16], en utilisant entre autre le théorème du sous-groupe analytique de Wüstholtz [Wüs89].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce sont des résultats difficiles et on peut s’attendre à ce que leurs analogues p-adiques le soient aussi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le seul cas où la Conjecture 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='13 est vérifiée est pour le motif de Kum- mer Ka de l’Exemple 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5(2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela suit de la transcendence des valeurs spéciales du logarithme p-adique [Ber77].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le premier cas ouvert intéressant serait celui des courbes elliptiques à réduction supersingulière, ce qui don- nerait notamment la transcendence de Γp(1/3), pour p ≡ 2[3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La version faible semble aussi difficile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une petit résultat dans cette direction a été donné dans le cas des courbes elliptique non CM à réduction supersingu- lière [AF22, Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motifs des schémas en groupes commutatifs Cette section résume deux travaux en collaboration avec Stephan Enright-Ward, Annette Huber et Simon Pepin Lehalleur [AEWH15, AHPL16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ils portent sur l’anneau de Chow et le motif d’un schéma en groupes commutatifs et généralisent les théorèmes de Beauville [Bea86] et Deninger–Murre [DM91] qui traitent le cas des schémas abéliens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous expliquons quelles sont les subtilités qui apparaissent quand on quitte le cadre des schémas en groupes projectifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les motifs de Voevodsky deviennent essentiels, non seulement leur existence mais aussi la nature de leur construction : c’est un exemple du principe expliqué au §1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce qui suit S est une variété de type fini et lisse sur un corps k qui jouera le rôle d’une base fixée.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tout S-schéma en groupe que l’on considérera sera lisse et de type fini à fibres connexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Certains énoncés sont valables dans des meilleures généralités.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour un S-schéma lisse f : X → S, CH(X) indique l’anneau de Chow de l’espace total X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On continue à travailler avec les coefficients rationnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 47 Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Beauville [Bea86], Deninger–Murre [DM91]) Soient A un S- schéma abélien de dimension relative g et nA : A → A le morphisme de multi- plication par n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors on a une décomposition CHi(A) = g+i � r=i CHi (r)(A) où CHi (r)(A) = {α ∈ CHi(A), n∗ Aα = nrα, ∀n ∈ Z}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) Ce théorème a été démontré par Beauville dans le cas S = Spec(k) et Deninger–Murre dans le cas général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les deux résultats utilisent de façon cruciale la transformée de Fourier que l’on rappelle plus loin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Quand S = Spec(k), on a l’inclusion ker cli A ⊇ g+i � r̸=2i CHi (r)(A) (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) qui vient du fait que n∗ A agit sur la cohomologie de degré s comme ns · Id .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) La décomposition du théorème induit une bigraduation sur l’anneau CH∗ (•)(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Quand S = Spec(k), la nouvelle graduation scinde la filtration de Bloch– Beilinson qui est conjecturée avoir certaine propriétés par rapport à l’appli- cation classe de cycle, notamment l’inclusion (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) devrait être une égalité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La filtration de Bloch–Beilinson est conjecturée exister pour tous les an- neaux de Chow de toutes les variétés projectives et lisses sur k, mais en général cette filtration ne se scinde pas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient A∨ le schéma abélien dual, P ∈ CH1(A × A∨) le diviseur associé au fibré de Poincaré et π1, π2 les projections de A×A∨ sur les deux facteurs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On définit la transformée de Fourier FA : CH∗(A) −→ CH∗(A∨), α �→ (π2)∗(exp(P) · π∗ 1α).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On vérifie que c’est un isomorphisme, dont l’inverse est essentiellement FA∨.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On en déduit la décomposition CHi(A) = � s {α ∈ CHi(A), FA(α) ∈ CHs(A)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) Ensuite un calcul direct permet de voir comment n∗ A agit sur chaque facteur de la décomposition (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On retrouve ainsi la décomposition de l’énoncé et l’identifica- tion CHi (r)(A) = {α ∈ CHi(A), FA(α) ∈ CHg+i−r(A)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Nous montrons la généralisation suivante.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient G un S-schéma en groupes commutatifs de dimension re- lative d et nG : G → G le morphisme de multiplication par n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors on a une décomposition CH∗(G) = 2d � r=0 CH∗ (r)(G) où CH∗ (r)(G) = {α ∈ CH∗(G), n∗ Gα = nrα, ∀n ∈ Z}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 48 GIUSEPPE ANCONA Remarque 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) On ne sait pas définir une transformée de Fourier pour un tel G : l’application π2 n’est pas propre, donc (π2)∗ n’existe pas, et de plus le dual d’un tel G est un 1-motif en général et non pas une variété.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Le premier cas non trivial pour les groupes non projectifs est donné par S = Spec(k) et G qui admet une suite exacte 0 −→ Gm −→ G −→ A −→ 0 où A est une variété abélienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas le Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3 est facile pour Gm et connu pour A mais on ne peut pas le déduire directement pour G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le problème est que cet énoncé se comporte bien pour les sommes directes mais mal pour les suites exactes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’énoncé qui suivra sera plus adapté à ce genre de dévissage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Définition 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient Sm/S la catégorie des S-schémas lisses de type fini et PSh(S) la catégorie des préfaisceaux sur Sm/S à valeur dans les Q-espaces vec- toriels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient Q(G) ∈ PSh(S) le préfaisceau qui associe à chaque Y ∈ Sm/S le Q-espace vectoriel ayant comme base l’ensemble HomS(Y, G) et G ∈ PSh(S) celui qui associe à chaque Y ∈ Sm/S le Q-espace vectoriel HomS(Y, G) ⊗Z Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La loi de groupe de G induit une transformation naturelle sG : Q(G) −→ G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par construction de DM(S), la catégorie des motifs relatifs 18, les préfaisceaux ci- dessus induisent des motifs et la transformation naturelle un morphisme entre eux que l’on notera αG/S : M(G/S) −→ M1(G/S).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le motif M(G/S) est appelé le motif de G et le motif M1(G/S) est appelé le 1-motif de G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Gardons les notations de la définition ci-dessus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors le motif Symr M1(G/S) est nul pour r assez grand et le morphisme αG/S se prolonge en un unique morphisme de motifs en algèbres de Hopf ϕG/S : M(G/S) −→ � r=0 Symr M1(G/S) (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) qui est de plus un isomorphisme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) (Motifs vs faisceaux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Ce théorème n’est pas une consé- quence formelle d’un énoncé sur les préfaisceaux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons par exemple que le faisceau Symr G n’est pas nul, puisque G est un faisceau en espaces vectoriels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un phénomène plus subtile est le suivant : le théorème montre en parti- culier l’existence d’applications non nulles de M1(G/S) vers M(G/S).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En revanche, il n’y a pas d’application non nulle du faisceau G vers Q(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On considère ici uniquement la version stable de cette catégorie, c’est-à-dire la catégorie obtenue après ⊗-inversion du motif de Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il y a plusieurs descriptions de la catégorie stable DM(S).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il se trouve qu’elles sont équivalentes sous des hypothèses assez générales qui sont notamment satisfaites pour les bases S que l’on considère.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La version qui est adaptée à la Définition 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 est celle des motifs étales étudiés par Ayoub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette catégorie DM(S) est obtenue à partir de D(PSh(S)) en localisant pour imposer la descente étale et l’invariance par A1-homotopie, puis en stabilisant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 49 Esquissons l’argument.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soient α : G → Q(G) une telle transformation na- turelle et idG ∈ G(G) l’application identité.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La naturalité de α implique qu’il suffit de voir que α(idG) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Posons α(idG) = ai · fi où ai sont des nombres rationnels et les fi sont des endomorphismes de G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour montrer ai · fi = 0 considérons la naturalité par rapport aux mor- phismes nG de multiplication par n : idG ❴ � ✤ � � ai · fi ❴ � G nG � α � Q(G)(G) nG � G α � Q(G)(G) nG = n · idG ✤ � � nai · fi � ai · (nG ◦ fi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On prétend que l’égalité � nai · fi = � ai · (nG ◦ fi) force tous les fi à être nuls.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' tout d’abord supposons par l’absurde qu’il y avait un fi, disons f1, qui n’était pas de torsion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors la liste des morphismes nG ◦ f1 serait infinie et on pourrait choisir un n tel que nG ◦ f1 n’apparaisse pas dans la liste des fi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ceci contredirait l’égalité � nai · fi = � ai · (nG ◦ fi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors supposer que les fi soient tous de torsion et on peut donc choisir un n tel que les nG ◦ fi soient tous nuls.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’égalité � nai · fi = � ai · (nG ◦ fi) = 0 implique alors � ai · fi = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) (Décomposition de Chow–Künneth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') On continue à noter par nG : G → G le morphisme de multiplication par n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarquons que son action sur M1(G/S) vaut n · Id.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En particulier l’isomorphisme (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) donne une dé- composition de M(G/S) en espaces propres par rapport à l’action de nG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On en déduit par ailleurs que cette décomposition est une décomposition de Chow–Künneth relative (voir la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 pour le cas absolu).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) (Décomposition de l’anneau de Chow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La décomposition du point (2) donne une décomposition de HomDM(S)(M(G/S), 1(p)[q]) (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) en espaces propres par rapport à l’action de nG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Comme ces Hom cal- culent 19 les groupes de Chow supérieurs [CD19, Corollary 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='14] on dé- duit le Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (4) (Décomposition de motifs vs décomposition d’anneaux de Chow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Pour les schémas abéliens G = A les décompositions des anneaux de Chow au point (3) permettent de retrouver celle du motif au point (2) comme l’ont remar- qué Deninger et Murre.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le point est de considérer A ×S A comme schéma abélien sur A et d’appliquer le Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 à ce schéma abélien : la dia- gonale se décomposera alors en somme de vecteurs propres.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part la 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les constructions de Grothendieck et de Voevodsky ont une convention de covariance diffé- rente, notamment les motifs de Chow se plongent dans les motifs de Voevodsky par un foncteur contravariant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est la raison pour laquelle l’objet 1(p)[q] apparaît à droite dans la formule (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 50 GIUSEPPE ANCONA diagonale s’interprète comme l’identité du motif M(A/S) et on peut veri- fier que cette décomposition de Id ∈ End(M(A/S)) est une décomposition en somme de projecteurs orthogonaux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour un G général, la formule EndDM(S)(M(G/S)) = HomDM(S)(M(G/S) ⊗ M(G/S)∨, 1) ne permet pas de relier ce groupe à l’anneau de Chow de G ×S G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet la dualité de Poincaré identifie, à un twist et shift près, M(G/S)∨ avec Mc(G/S) qui n’est pas, en général, M(G/S).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas la décomposition des anneaux de Chow du Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3 ne permet pas de retrouver celle des motifs au point (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (L’argument ci-dessus montre qu’en général les endomorphismes d’un motif sont reliés aux groupes de Chow uniquement dans le cas propre et lisse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cela a déjà été signalé au §1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 et c’est le point qui limite la construction classique de Grothendieck au cadre propre et lisse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') (5) (Voevodsky vs Chow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Dans le cas G = A d’un schéma abélien, une formule M(A/S) ∼= � r=0 Symr M1(A/S) (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) a été montré par Künnemann encore à l’aide de la transformée de Fourier [Kün94].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une version plus faible, valable pour les motifs homologiques, a été discutée dans la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A l’époque de [Kün94] on ne disposait pas des motifs de Voevodsky et le travail a été fait dans les motifs de Chow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans ce cas l’existence du motif M1(A/S) présent dans la formule (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) n’est pas du tout triviale : il faut construire un projecteur convenable de End(M(A/S)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce motif est en revanche facile à définir dans le cadre de Voevodsky (Définition 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un des avantage des motifs de Voevodsky sur les motifs de Chow est notam- ment cette possibilité de disposer facilement de motifs par des constructions faisceautiques : c’est le principe que nous avons mentionné au §1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La preuve se base sur deux dévissages qui font chacun l’objet d’un article.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un premier dévissage sert à se réduire au cas d’un corps algébriquement clos S = Spec(K), [AHPL16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le deuxième [AEWH15] est une réduction aux cas des variétés abéliennes, ce qui nous ramène essentiellement au résultat de Künnemann [Kün94].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Réduction à S = Spec(K).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le premier dévissage on utilise le théorème suivant d’Ayoub [Ayo14, Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si f : M → N est un morphisme de motifs dans DM(S) alors pour voir si f est un isomorphisme il suffit de voir si son tiré en arrière en tout point géométrique l’est.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit i : Spec(K) → S un point géométrique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour compléter le premier dévissage il suffit alors de montrer que i∗ϕG/S = ϕG×SSpec(K)/ Spec(K).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) La formule (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) est en fait le point technique du travail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour tout morphisme g : T → S, le foncteur g∗ est caractérisé par la propriété que, pour tout S-schéma lisse X, on ait le changement de base g∗M(X/S) = M(X ×S T/T ) (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7) MÉMOIRE HDR 51 comme pour la cohomologie à support compact.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Or parmi les deux motifs qui interviennent dans le morphisme ϕG/S, seulement M(G/S) est de cette forme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le point est alors de trouver une résolution de M1(G) par un complexe dont tous les termes sont des sommes d’objets de la forme M(X/S) en tous les degrés et toutes les flèches de connections sont induites par des morphismes de S-schémas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous construisons une telle résolution inspirée par des construction de [Bre70].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilise uniquement des X qui sont des puissances de G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Avec les notations de la Définition 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 on peut écrire son début sous la forme · · −→ Q(G × G) tG −→ Q(G) sG −→ G, où tG([g1, g2]) = [g1] + [g2] − [g1 + g2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Réduction à G = A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le deuxième dévissage fixons un corps algébriquement clos K et considérerons S = Spec(K).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans la suite on allégera la notation en enlevant les /S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le théorème de Chevalley nous dit qu’un groupe algébrique sur un corps K s’insère dans une suite exacte 0 −→ L −→ G −→ A −→ 0, où A est une variété abélienne et L est un groupe linéaire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors raisonner par récurrence sur la dimension de L et se ramener à 0 −→ Ga −→ G −→ H −→ 0 (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8) ou 0 −→ Gm −→ G −→ H −→ 0, (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9) où dans les deux cas H est un groupe pour lequel l’énoncé est connu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le cas (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8) est facile : par homotopie l’énoncé pour G est équivalent à l’énoncé pour H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour le cas (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9) l’argument est plus délicat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Premièrement, la suite exacte (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9) donne une suite exacte au niveau des fonc- teurs des points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si on applique Symn a cette dernière on obtient un triangle Symr−1 M1(H) ⊗ M1(Gm) −→ Symr M1(G) −→ Symr M1(H), où on a utilisé Sym2 M1(Gm) = 0 [Voe00, Corollary 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette suite exacte permet de déduire par récurrence que Symr M1(G) s’annule pour r assez grand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Deuxièmement, on complète le Gm-fibré G −→ H en un A1-fibré ¯G −→ H avec une section nulle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le triangle de localisation par rapport à la section donne M(G) −→ M( ¯G) −→ M(H)(1)[2], or par homotopie M( ¯G) = M(H).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilise maintenant l’hypothèse de récurrence et on en déduit le diagramme M(G) � ψ �✤ ✤ ✤ ✤ ✤ ✤ M(H) � ϕH � M(H)(1)[2] ϕH(1)[2] � � r Symr M1(G) � � r Symr M1(H) � � r Symr M1(H)(1)[2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='10) 52 GIUSEPPE ANCONA L’existence de ψ suit de la commutativité du carré de droite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cette commutativité n’est pas gratuite, elle utilise la définition de l’application ϕH.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par hypothèse de récurrence on déduit que l’application ψ est un isomorphisme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Malheureusement la récurrence n’est pas terminée puisqu’on ne sait pas lier ψ à ϕG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’existence de ψ a tout de même une conséquence importante : M(G) est un motif de dimension finie, voir la Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='21(2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On montre que la réalisation de ϕG est un isomorphisme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors appliquer la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5 à ϕG et ψ−1 pour déduire que ϕG est un isomorphisme après passage au quotient par l’équivalence homologique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut ensuite utiliser les propriétés des motifs de dimension finie pour conclure que ϕG est un isomorphisme même avant passage au quotient (voir la Remarque 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='21(2), cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' le Théorème 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2 dans le cas pur).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Remarque 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) La notion de dimension finie dans les motifs a toujours été appliquée pour les motifs purs : elle ne se comporte pas bien par suite exacte et on connait des exemples de motifs mixtes qui ne sont pas de dimension finie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' À notre connaissance cette preuve est le premier exemple d’application de ces idées aux motifs mixtes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Le Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 a permis à Huber et Kings de construire le polylogarithme pour tous les schémas en groupes commutatifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) Le Théorème 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6 donne une description du motif d’un groupe algébrique commutatif G en terme d’un motif assez simple, M1(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On pourrait espérer que cela puisse aider à une meilleure compréhension des anneaux de Chow de G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Construction de classes algébriques Cette section concerne la conjecture standard de type Lefschetz, introduite dans la Conjecture 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans un travail en cours en collaboration avec Mattia Cavicchi, Robert Laterveer et Giulia Saccà, nous étudions cette conjecture pour les variétés hyper-kähler qui admettent une fibration lagrangienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le point de départ est une idée récente de Voisin [Voi22] que l’on regarde dans la perspective du théorème de décomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous nous concentrerons dans la suite sur les variétés complexes, la conjecture standard de type Lefschetz est alors une instance particulière de la conjecture de Hodge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De façon assez surprenante elle en est aussi le pilier principal : André dé- montre que sous la conjecture standard de type Lefschetz le transport parallèle de classes algébriques est encore algébrique [And96].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il en déduit que cette conjecture impliquerait la conjecture de Hodge pour les variétés abéliennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La conjecture standard de type Lefschetz est connue pour les variétés abéliennes, voir la Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7, et on sait en déduire le cas des surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Plus récemment Charles et Markmann l’ont montrée pour les variétés hyper-kähler de type KS[n], [CM13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La dimension d’une variété X sera notée dX et la dimension de la fibre générique d’un morphisme f sera notée df.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Même en présence de faisceaux pervers on utilisera la convention classique pour les degrés cohomologiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une approche naïve.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Considérons une variété projective et lisse X et sup- posons qu’elle admette un morphisme f : X → B vers une base B qui est aussi MÉMOIRE HDR 53 projective et lisse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Nous nous demandons jusqu’à quel point connaître la conjecture standard de type Lefschetz pour B et pour les fibres lisses de f peut aider pour montrer la conjecture standard de type Lefschetz pour X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le théorème de décomposition implique en particulier une décomposition 20 de structures de Hodge Hn(X) = � a+b=n Ha(B,pRbf∗Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) Soient η un diviseur sur X qui soit relativement ample, LB un diviseur ample sur B et β son tiré en arrière sur X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le théorème de décomposition fournit aussi les isomorphismes ∪ηb : Ha(B,pRdf−bf∗Q) ∼ −−→ Ha(B,pRdf+bf∗Q), (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) ∪βa : HdB−a(B,pRbf∗Q) ∼ −−→ HdB+a(B,pRbf∗Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) La combination de (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1), (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) et (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) peut suggérer que la conjecture standard de type Lefschetz se ramène à montrer que les inverses de ces isomorphismes sont algébriques et à première vue on pourrait penser que ces derniers se ramènent uniquement à l’étude des fibres de f et de la base B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mais il faut en fait faire attention à un certain nombre de subtilités.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) Les faisceaux pervers pRbf∗Q dépendent aussi des fibres singulières du mor- phisme f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Inverser l’action de LB sur la cohomologie de B ne suffit pas à inverser β.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En effet f ∗ H(B) ⊂ H(X) ne représente que le facteur H∗(B,pR0f∗Q) de la décomposition (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (3) Même si on était capables de construire des correspondances algébriques Ληb et Λβa qui agiraient comme les inverses de (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2) et (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) il n’est pas clair de pouvoir les mettre ensemble pour déduire une correspondance algébrique Λn : HdX+n(X) ∼ −−→ HdX−n(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il faudrait par exemple contrôler comment Ληb agit sur les degrés pervers différents de b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est délicat, notamment parce que les opérateurs η et β ne sont pas bigradués en général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Dans le cas des variétés hyper-kähler qui admettent une fibration lagrangienne on peut espérer contourner ces problèmes : le théorème du support de Ngô donne une description explicite des faisceaux pervers pRbf∗Q, un argument de Voisin permet grosso-modo de construire une deuxième fibration lagrangienne qui inverse le rôle de η et β, enfin Shen et Yin ont montré que les opérateurs η et β sont en fait bigradués pour les fibrations lagrangiennes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proposition 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit f : X → B une application entre variétés projectives, lisses et connexes et soit U l’ouvert de B sur lequel l’application est lisse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (1) Supposons que la fibre générique vérifie la conjecture standard de type Lef- schetz et que pRbf∗Q = IC((Rbf∗Q)|U).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4) 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une telle décomposition n’est pas unique en général, seulement la filtration perverse associée l’est.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Deligne, puis De Cataldo, ont proposé des décompositions qui se comportent mieux que les autres [Del94b, dC13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 54 GIUSEPPE ANCONA Alors il existe des correspondances dans X ×B X dont l’action sur la coho- mologie relative induit des isomorphismes pRdf+bf∗Q ∼ −−→ pRdf−bf∗Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2) Supposons qu’il existe des correspondances dans X×BX dont l’action sur la cohomologie relative induit des isomorphismes pRdf+bf∗Q ∼ −−→ pRdf−bf∗Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors il existe une décomposition du motif h(X) = � b h(B,pRbf∗Q) (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) où R(h(B,pRbf∗Q)) = H∗(B,pRbf∗Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus chaque facteur de la décom- position est autodual à un twist près.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Autour de la preuve.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') La preuve de (1) est facile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit Y la fibre générique de f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’hypothèse sur Y dans (1) fournit des correspondances dans Y ×Y , dont les adhérences donnent des correspondances dans X ×B X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Leur action sur la cohomologie relative IC((Rbf∗Q)|U) est contrôlée par leur action sur le système local (Rbf∗Q)|U et donc par l’action sur la cohomologie de Y .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour (2) on suit l’argument classique qui montre que Lefschetz implique Kün- neth [Kle68].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par rapport au cas absolu on prendra garde au fait que la décom- position (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) n’est pas unique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' L’action des correspondances relatives ne respecte pas cette graduation mais seulement la filtration associée.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Par ailleurs on ne sait pas démontrer que toute décomposition cohomologique (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) est la réalisation d’une décomposition motivique (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5) mais seulement qu’il en existe au moins une qui est d’origine motivique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Corollaire 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit f : X → P1 une fibration de Lefschetz dont la fibre générique vérifie la conjecture standard de type Lefschetz (par exemple f est une fibration en surfaces).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors il existe une décomposition du motif h(X) = � b h(P1,pRbf∗Q) (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6) où R(h(P1,pRbf∗Q)) = H∗(P1,pRbf∗Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus chaque facteur de la décomposition est autodual à un twist près.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Du relatif à l’absolu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Toute variété de dimension trois admet une fibration de Lefschetz après éclatement le long d’une courbe C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part la formule de l’éclatement h(BlC(X)) = h(X) ⊕ h(C)(−1), due à Manin [Man68], montre que la conjecture standard de type Lefschetz pour X se ramène à celle pour BlC(X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si l’on veut étudier la conjecture standard de type Lefschetz pour une variété X de dimension trois on peut donc supposer que X admet une telle fibration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ce qui manque au corollaire ci-dessus pour déduire cette conjecture est l’existence d’une décomposition h(P1,pRbf∗Q) = 2 � a=0 ha(P1,pRbf∗Q), où R(ha(P1,pRbf∗Q)) = Ha(P1,pRbf∗Q), telle que chaque facteur de la décomposi- tion soit autodual à un twist près.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 55 Pour construire cette décomposition on pourrait vouloir utiliser l’isomorphisme (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) et essayer de construire une correspondance algébrique qui induise l’inverse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est une question qui ressemble au problème original de construction de l’inverse de l’opérateur de Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On ne sait pas si c’est un problème plus simple, voir aussi la Remarque 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorème 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit X une variété hyper-kähler de dimension 2n et f : X → Pn une fibration lagrangienne dont toutes les fibres sont irréductibles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que le schéma en groupes Aut0(f) soit polarisable (au sens de Ngô).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors h(X) = 2n � b=0 h(Pn,pRbf∗Q) (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7) où R(h(Pn,pRbf∗Q)) = H∗(Pn,pRbf∗Q).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' De plus chaque facteur de la décomposition est autodual à un twist près.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilise le théorème du support de Ngô, [Ngô10] : quand les fibres sont toutes irréductibles on a bien l’hypothèse (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' D’autre part la fibre générique est une variété abélienne, donc elle vérifie la conjecture standard de type Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors utiliser la Proposition 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 pour conclure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Corollaire 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soit X une variété hyper-kähler de dimension 2n de rang de Pi- card 2 qui admet une fibration lagrangienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Supposons que pour toute fibration lagrangienne g de n’importe quel variétés hyper-kähler birationnelle à X, les fibres de g soient irréductibles et le schéma en groupes Aut0(g) soit polarisable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Alors la conjecture standard de type Lefschetz est vraie pour X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Démonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fixons f : X → Pn une fibration lagrangienne et soit η et β les diviseurs introduits dans §9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Un argument de Voisin [Voi22] montre grosso-modo l’existence d’une fibration lagrangienne g : X → Pn où le rôle de η et β est inversé.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' En fait g existe seulement sur une variété hyper-kähler qui est birationnelle à X, mais elles ont le même motif [Rie16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On peut alors appliquer le théorème ci-dessus à f et g pour déduire deux décom- positions du motif h(X) dont tous les facteurs sont autoduaux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il n’est pas clair que ces deux décompositions soient compatibles, voir aussi le point (3) dans §9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mais un théorème de Shen et Yin montre que les opérateurs η et β sont bigradués pour les fibrations lagrangiennes : on peut voir que ceci force les deux décompositions à être compatibles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La bidécomposition que l’on en déduit est plus fine que (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1) et a tous les facteurs autoduaux, ce qui implique la conjecture standard de type Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' □ Le corollaire ci-dessus s’applique par exemple aux variétés hyper-kähler construites par Laza–Saccà–Voisin [LSV17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Généralisations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') On souhaite se débarrasser de l’hypothèse d’ir- réductibilité des fibres dans le théorème ci-dessus, ce qui permettrait de l’enlever aussi dans son corollaire et de pouvoir l’appliquer à beaucoup plus de variétés.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Si les fibres ne sont pas irréductible les faisceaux pervers pRbf∗Q ne vérifient plus (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cependant le théorème du support de Ngô décrit aussi la nature des faisceaux pervers supportés sur les sous-variétés strictes de la base.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Il montre l’existence 56 GIUSEPPE ANCONA de certaines variétés abéliennes contenues dans certaines fibres singulières dont la cohomologie contrôle les faisceaux pervers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On pense qu’une variante stratifiée de la Proposition 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 devrait s’appliquer à ce contexte général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On utilisera encore que les variétés abéliennes vérifient la conjecture standard de type Lefschetz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Remarque 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (Du relatif à l’absolu, suite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=') Reprenons les notations de la Re- marque 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On souhaiterait montrer l’existence d’une décomposition h(P1,pRbf∗Q) = 2 � a=0 ha(P1,pRbf∗Q), (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8) où R(ha(P1,pRbf∗Q)) = Ha(P1,pRbf∗Q), telle que chaque facteur de la décomposi- tion soit autodual à un twist près : ceci montrerait la conjecture standard de type Lefschetz pour les variétés de dimension trois.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour construire cette décomposition on pourrait vouloir utiliser l’isomorphisme (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='3) induit par la classe β et essayer de construire une correspondance algébrique qui induise l’inverse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inspirés par l’idée de Voisin esquissé dans la preuve du co- rollaire ci-dessus on peut essayer de construire une nouvelle fibration g : X → P2 telle que β soit relativement ample (ces constructions sont toujours possibles après éclatement).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une telle fibration induit une décomposition h(X) = 2 � a=0 h(P2,pRag∗Q) (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9) où ∪β : h(P2,pR0g∗Q) ∼ −−→ h(P2,pR2g∗Q) admet un inverse algébrique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le problème est qu’en général la décomposition (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='9) n’induit pas la décomposition (9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C’est encore lié au fait que les actions de η et β ne sont pas bigraduées en général.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 57 Références [ACLS22] Giuseppe Ancona, Mattia Cavicchi, Robert Laterveer, and Giulia Saccà.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Standard conjectures for some lagrangian fibrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' prépublication, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [AEWH15] Giuseppe Ancona, Stephen Enright-Ward, and Annette Huber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On the motive of a commutative algebraic group.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Documenta Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 20 :807–858, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [AF22] Giuseppe Ancona and Dragos Fratila.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Algebraic classes in mixed characteristic and André’s p-adic periods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' prépublication, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [AHPL16] Giuseppe Ancona, Annette Huber, and Simon Pepin Lehalleur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On the relative motive of a commutative group scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Algebr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Geom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 3(2) :150–178, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [AM22] Giuseppe Ancona and Adriano Marmora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' The Hilbert symbol in the Hodge standard conjecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' prépublication, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Anc21] Giuseppe Ancona.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Standard conjectures for abelian fourfolds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 223(1) :149–212, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Anc22] Giuseppe Ancona.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Conservativity of realization functors on motives of abelian type over finite fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Proceedings of "Motives and Complex Multiplication", Fresán, Jet- chev, Jossen and Pink editors, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [And90] Yves André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' p-adic Betti lattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In p-adic analysis (Trento, 1989), volume 1454 of Lecture Notes in Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 23–63.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Springer, Berlin, 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [And95] Yves André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Théorie des motifs et interprétation géométrique des valeurs p-adiques de G-functions (une introduction).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Number theory (Paris, 1992–1993), volume 215 of London Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Lecture Note Ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 37–60.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cambridge Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Press, Cambridge, 1995.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [And96] Yves André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pour une théorie inconditionnelle des motifs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Hautes Études Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Publ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', (83) :5–49, 1996.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [And03] Yves André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Period mappings and differential equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' From C to Cp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tôhoku- Hokkaidô lectures in arithmetic geometry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' With appendices : A : Rapid course in p-adic analysis by F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Kato, B : An overview of the theory of p-adic unifomization by F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Kato, C : p-adic symmetric domains and Totaro’s theorem by N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tsuzuki, volume 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tokyo : Mathematical Society of Japan, 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [And04] Yves André.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une introduction aux motifs (motifs purs, motifs mixtes, périodes), vo- lume 17 of Panoramas et Synthèses [Panoramas and Syntheses].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Société Mathéma- tique de France, Paris, 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ayo07] Joseph Ayoub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Les six opérations de Grothendieck et le formalisme des cycles évanes- cents dans le monde motivique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Astérisque, (314) :x+466 pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2008), 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ayo10] Joseph Ayoub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Note sur les opérations de Grothendieck et la réalisation de Betti.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Jussieu, 9(2) :225–263, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ayo14] Joseph Ayoub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' La réalisation étale et les opérations de Grothendieck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' École Norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sup.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 47 :1–141, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ayo15] Joseph Ayoub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Une version relative de la conjecture des périodes de Kontsevich-Zagier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2), 181(3) :905–992, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [BC16] Jean-Benoît Bost and François Charles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Some remarks concerning the Grothendieck period conjecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Reine Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 714 :175–208, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Bea86] Arnaud Beauville.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sur l’anneau de Chow d’une variété abélienne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 273(4) :647–651, 1986.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ber77] Daniel Bertrand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sous-groupes à un paramètre p-adique de variétés de groupe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 40(2) :171–193, 1977.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Bon10] Mikhail V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Bondarko.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Weight structures vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' t-structures ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' weight filtrations, spectral sequences, and complexes (for motives and in general).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' K-Theory, 6(3) :387–504, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Bre70] Lawrence Breen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Extensions of abelian sheaves and Eilenberg-MacLane algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 9 :15–44, 1969/1970.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Bro12] Francis Brown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mixed Tate motives over Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2), 175(2) :949–976, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 58 GIUSEPPE ANCONA [Bro17] Francis Brown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Notes on motivic periods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Number Theory Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 11(3) :557–655, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [CD19] Denis-Charles Cisinski and Frédéric Déglise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Triangulated categories of mixed motives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Springer Monographs in Mathematics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Springer, Cham, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [CDN22] Mattia Cavicchi, Frédéric Déglise, and Jan Nagel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motivic decompositions of families with tate fibers : smooth and singular cases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' prépublication, 3(2), 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [CF00] Pierre Colmez and Jean-Marc Fontaine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Construction des représentations p-adiques semi-stables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 140(1) :1–43, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Cha13] François Charles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' The Tate conjecture for K3 surfaces over finite fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 194(1) :119–145, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Chu80] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Chudnovsky.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Algebraic independence of values of exponential and elliptic func- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Proceedings of the International Congress of Mathematicians (Helsinki, 1978), pages 339–350.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fennica, Helsinki, 1980.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Clo99] Laurent Clozel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Equivalence numérique et équivalence cohomologique pour les variétés abéliennes sur les corps finis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2), 150(1) :151–163, 1999.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [CM13] François Charles and Eyal Markman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' The standard conjectures for holomorphic sym- plectic varieties deformation equivalent to Hilbert schemes of K3 surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Compos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 149(3) :481–494, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Col90] Robert F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Coleman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On the Frobenius matrices of Fermat curves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In p-adic analysis (Trento, 1989), volume 1454 of Lecture Notes in Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 173–193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Springer, Berlin, 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [dC13] Mark Andrea A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' de Cataldo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Hodge-theoretic splitting mechanisms for projective maps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Singul.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 7 :134–156, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' With an appendix containing a letter from P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Deligne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Del89] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Deligne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le groupe fondamental de la droite projective moins trois points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Galois groups over Q (Berkeley, CA, 1987), volume 16 of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Publ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 79–297.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Springer, New York, 1989.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Del94a] Pierre Deligne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' à quoi servent les motifs ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Motives (Seattle, WA, 1991), volume 55 of Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sympos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pure Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 143–161.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', Providence, RI, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Del94b] Pierre Deligne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Décompositions dans la catégorie dérivée.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Motives (Seattle, WA, 1991), volume 55 of Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sympos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Pure Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 115–128.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', Providence, RI, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Del13] Pierre Deligne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Multizêtas, d’après Francis Brown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Number 352, pages Exp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 1048, viii, 161–185.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Séminaire Bourbaki.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 2011/2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Exposés 1043–1058.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [DM91] Christopher Deninger and Jacob Murre.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Motivic decomposition of abelian schemes and the Fourier transform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Reine Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 422 :201–219, 1991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Fal89] Gerd Faltings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Crystalline cohomology and p-adic Galois-representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Algebraic analysis, geometry, and number theory (Baltimore, MD, 1988), pages 25–80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Johns Hopkins Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Press, Baltimore, MD, 1989.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Fon94] Jean-Marc Fontaine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le corps des périodes p-adiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Astérisque, (223) :59–111, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' With an appendix by Pierre Colmez, Périodes p-adiques (Bures-sur-Yvette, 1988).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Fur07] Hidekazu Furusho.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' p-adic multiple zeta values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tannakian interpretations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 129(4) :1105–1144, 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Jan07] Uwe Jannsen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On finite-dimensional motives and Murre’s conjecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Algebraic cycles and motives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 2, volume 344 of London Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Lecture Note Ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 112–142.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Cambridge Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Press, Cambridge, 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Kah03] Bruno Kahn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Équivalences rationnelle et numérique sur certaines variétés de type abélien sur un corps fini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' École Norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sup.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (4), 36(6) :977–1002 (2004), 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Kat80] Nicholas M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Katz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' p-adic L-functions, Serre-Tate local moduli, and ratios of solutions of differential equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Proceedings of the International Congress of Mathemati- cians (Helsinki, 1978), pages 365–371.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Fennica, Helsinki, 1980.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' MÉMOIRE HDR 59 [Kim05] Shun-Ichi Kimura.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Chow groups are finite dimensional, in some sense.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 331(1) :173–201, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Kle68] Steven Kleiman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Algebraic cycles and the Weil conjectures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Dix esposés sur la cohomologie des schémas, pages 359–386.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' North-Holland, Amsterdam, 1968.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [KM74] Nicholas M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Katz and William Messing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Some consequences of the Riemann hypothesis for varieties over finite fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 23 :73–77, 1974.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Kos22] Teruhisa Koshikawa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' The numerical Hodge standard conjecture for the square of a simple abelian variety of prime dimension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' prépublication, 3(2), 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Kün94] Klaus Künnemann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On the Chow motive of an abelian scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Motives, volume 55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='1 of Proceedings of Symposia in pure mathematics, pages 189–205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' American ma- thematical society, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [LSV17] Radu Laza, Giulia Saccà, and Claire Voisin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A hyper-Kähler compactification of the intermediate Jacobian fibration associated with a cubic 4-fold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Acta Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 218(1) :55– 135, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Man68] Ju.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Manin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Correspondences, motifs and monoidal transformations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' ), 77 (119) :475–507, 1968.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Mil07] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Milne.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' The tate conjecture over finite fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' pages 1–27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Available on the author’s webpage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Mum68] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Mumford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Rational equivalence of 0-cycles on surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Kyoto Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 9 :195–204, 1968.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ngô10] Bao Châu Ngô.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Le lemme fondamental pour les algèbres de Lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Publ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Inst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Hautes Études Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', (111) :1–169, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ogu90] Arthur Ogus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' A p-adic analogue of the Chowla-Selberg formula.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In p-adic analysis (Trento, 1989), volume 1454 of Lecture Notes in Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 319–341.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Springer, Berlin, 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [O’S05] Peter O’Sullivan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' The structure of certain rigid tensor categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Paris, 340(8) :557–562, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Rie16] Ulrike Rieß.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On Beauville’s conjectural weak splitting property.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' IMRN, (20) :6133–6150, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Sch15] Stefan Schreieder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On the construction problem for Hodge numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Geom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Topol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 19(1) :295–342, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Ser72] Jean-Pierre Serre.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Propriétés galoisiennes des points d’ordre fini des courbes ellip- tiques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 15(4) :259–331, 1972.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [SK79] Tetsuji Shioda and Toshiyuki Katsura.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On Fermat varieties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Tôhoku Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2), 31(1) :97–115, 1979.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Tat66] John Tate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Endomorphisms of abelian varieties over finite fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Invent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 2 :134– 144, 1966.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Voe00] Vladimir Voevodsky.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Triangulated categories of motives over a field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' In Cycles, trans- fers, and motivic homology theories, volume 143 of Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Stud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', pages 188– 238.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Princeton Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Press, Princeton, NJ, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Voi22] Claire Voisin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' On the Lefschetz standard conjecture for Lagrangian covered hyper- Kähler varieties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 396 :Paper No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' 108108, 29, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Wil17] Jörg Wildeshaus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Intermediate extension of Chow motives of Abelian type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=', 305 :515–600, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' [Wüs89] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Wüstholz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Algebraische Punkte auf analytischen Untergruppen algebraischer Grup- pen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' (2), 129(3) :501–517, 1989.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content=' IRMA, Université de Strasbourg 7, rue René Descartes, 67000 Strasbourg Email address: ancona@math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='unistra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='fr URL: http://irma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='unistra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} +page_content='fr/~ancona/' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/99E0T4oBgHgl3EQfgAA0/content/2301.02411v1.pdf'} diff --git a/9NE5T4oBgHgl3EQfRA5T/content/2301.05517v1.pdf b/9NE5T4oBgHgl3EQfRA5T/content/2301.05517v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ed056fc68f63017e4dd5039d9553aca87db2a6c8 --- /dev/null +++ b/9NE5T4oBgHgl3EQfRA5T/content/2301.05517v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e88a9bfbdcb28d48ff8aabe0c0b28589ad0b1fd2f70353125b6d2490637bafa8 +size 385476 diff --git a/9NE5T4oBgHgl3EQfRA5T/vector_store/index.faiss b/9NE5T4oBgHgl3EQfRA5T/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..3b9e8b91d69fb94b72398de5d8a6eccdadf5e509 --- /dev/null +++ b/9NE5T4oBgHgl3EQfRA5T/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f2c5a9daa79a86c939a140328709879332a4ed97f3f1cb5942e2806be5be970 +size 1769517 diff --git a/9NE5T4oBgHgl3EQfRA5T/vector_store/index.pkl b/9NE5T4oBgHgl3EQfRA5T/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..3f18fdbda87b70f34d7bf30490ac5b628f58f7f7 --- /dev/null +++ b/9NE5T4oBgHgl3EQfRA5T/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2aee50ab517f83cdca4629d30b2b95c566d1cb366780280ea5ed34abd4514a3 +size 73536 diff --git a/9NFLT4oBgHgl3EQfBy4O/vector_store/index.faiss b/9NFLT4oBgHgl3EQfBy4O/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..3ddc71bf600178109b42756f77912e0be9b46dc0 --- /dev/null +++ b/9NFLT4oBgHgl3EQfBy4O/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc77c1faf429d2804d18325fea1574460523ab981ba1c86021d036115caf3616 +size 6357037 diff --git a/ANE1T4oBgHgl3EQfowXh/vector_store/index.faiss b/ANE1T4oBgHgl3EQfowXh/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..1bfdd1e1beeb2fa0cea28e4a129c401b459c1ff0 --- /dev/null +++ b/ANE1T4oBgHgl3EQfowXh/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c481c41c05720e00937c3ad3953e492c418f366514dbbb2e4c320ae3bc4de34b +size 4063277 diff --git a/ANE1T4oBgHgl3EQfowXh/vector_store/index.pkl b/ANE1T4oBgHgl3EQfowXh/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..ebe0a94a0137e66fcb0ceceea4a282c3ad80f319 --- /dev/null +++ b/ANE1T4oBgHgl3EQfowXh/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ad39d418ee7920c9644c411e1247c9f4acf24f900bff2ca9c7f698b5800098b +size 140442 diff --git a/BtAzT4oBgHgl3EQfTfyw/content/tmp_files/2301.01251v1.pdf.txt b/BtAzT4oBgHgl3EQfTfyw/content/tmp_files/2301.01251v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..04473259635d1bc3b07e52595925d417b3798b20 --- /dev/null +++ b/BtAzT4oBgHgl3EQfTfyw/content/tmp_files/2301.01251v1.pdf.txt @@ -0,0 +1,1243 @@ +Introducing Variational Inference in +Undergraduate Statistics and Data Science +Curriculum +Vojtech Kejzlar +Department of Mathematics and Statistics, Skidmore College +and +Jingchen Hu +Department of Mathematics and Statistics, Vassar College +January 4, 2023 +Abstract +Probabilistic models such as logistic regression, Bayesian classification, neural net- +works, and models for natural language processing, are increasingly more present in +the undergraduate statistics and data science curriculum due to their wide range of +applications. In this paper, we present a one-week undergraduate course module on +variational inference, a popular optimization-based approach for approximate infer- +ence with probabilistic models. Our proposed module is guided by active learning +principles: In addition to lecture materials on variational inference, we provide an +accompanying class activity, an R shiny app, and a guided lab based on a real data +application of clustering documents using Latent Dirichlet Allocation with R code. +The main goal of our module is to expose undergraduate students to a method that +facilitates statistical modeling and inference with large datasets. Using our proposed +module as a foundation, instructors can adopt it and adapt to introduce more realistic +use cases and applications in data science, Bayesian statistics, multivariate analysis, +and statistical machine learning courses. +Keywords: Active learning, Bayesian Statistics, Statistical Computing, Probabilistic Mod- +els, Undergraduate Curriculum +1 +arXiv:2301.01251v1 [stat.OT] 3 Jan 2023 + +1 +Introduction +With the recent and rapid expansion of undergraduate curricula with offerings in data +science, Bayesian statistics, multivariate data analysis, and statistical machine learning, +probabilistic models and Bayesian methods have grown to become more popular (Schwab- +McCoy et al. 2021, Dogucu & Hu (2022)). In many settings, a central task in applications +of probabilistic models is the evaluation of posterior distribution p(θ | y) of m model +parameters θ ∈ Rm (m ≥ 1) conditioned on the observed data y = (y1, . . . , yn) provided +by the Bayes’ theorem +p(θ | y) = p(y | θ)p(θ) +p(y) +∝ p(y | θ)p(θ). +(1) +Here, p(y | θ) is the sampling density given by the underlying probabilistic model for data, +p(θ) is the prior density that represents our prior beliefs about θ before seeing the data, +and p(y) is the marginal data distribution. The posterior distribution p(θ | y), however, +has closed form only in a limited number of scenarios (e.g., conjugate priors) and there- +fore typically requires approximation. By far the most popular approximation methods +are Markov chain Monte Carlo (MCMC) algorithms including Gibbs sampler, Metropo- +lis, Metropolis-Hastings, and Hamiltonian Monte Carlo (Gelman et al. 2013), to name a +few. See Albert & Hu (2020) for a review of these algorithms in undergraduate Bayesian +courses. While useful for certain scenarios, these MCMC algorithms do not scale well with +large datasets and can have a hard time approximating multimodal posteriors (Rudoy & +Wolfe 2006, Bardenet et al. (2017)). Such challenges therefore limit the applications of +probabilistic models that can be discussed in the classroom and restrict students’ exposure +to more realistic use cases that include applying neural networks, pattern recognition, and +natural language processing to massive datasets. +Variational inference is an alternative to the sampling-based approximation via MCMC +that approximates a target density through optimization. Statisticians and computer sci- +entists (starting with Peterson & Anderson (1987), Jordan et al. (1999), Blei et al. (2017)) +have been using variational techniques in a variety of settings because these techniques tend +to be faster and easier to scale to massive datasets. Despite its popularity among statistics +and data science practitioners, variational inference is rarely discussed in undergraduate +2 + +courses as it is believed to be a rather advanced topic (Dogucu & Hu 2022). With this in +mind, we have developed a one-week course module to disrupt this notion of being an ad- +vanced topic and help instructors to introduce variational techniques in their undergraduate +courses for more realistic use cases of probabilistic models. Our proposed one-week module +is based on the best practices of active learning, which have been shown to improve student +learning and engagement (Michael 2006, Freeman et al. (2014), Deslauriers et al. (2019)). +Our main guiding principle in designing the module is to involve students in the learning +process by introducing student-centered class activities and labs. The guiding principle +also includes assigning open-ended questions, focusing on problem-solving, providing ap- +propriate scaffolding for activities, and creating opportunities to work collaboratively with +peers. +Our module is designed for students to gain a fundamental understanding and practical +experience with variational inference over the course of two class meetings. During the first +meeting, students are exposed to the fundamentals of variational inferences including the +Kullback-Leibler divergence, evidence lower bound, gradient ascent, and coordinate ascent. +Additionally, they gain their first hands-on experience by applying variational inference to +a simple probabilistic model for count data. To encourage and empower undergraduate +instructors to adopt and adapt this variational inference module, we provide an accompa- +nying in-class handout and an R Shiny app with details explained in the supplementary +materials. During the second class meeting, students work on a guided R lab to apply +variational inference to a realistic scenario of clustering documents with Latent Dirichlet +Allocation (Blei et al. 2003). See Table 1 for the breakdown of the module. +Content +1st class +Lecture: Fundamentals of variational inference +Class activity: Probabilistic model for count data with variational inference +2nd class +Lab: Document clustering +Table 1: +Outline of the one-week variational inference module. +As for the audience, we believe that the module can be seamlessly integrated into any +intermediate- or upper-level undergraduate course in data science, Bayesian statistics, mul- +tivariate data analysis, and statistical machine learning that covers topics on clustering, +3 + +classification, or text analysis. The prerequisites needed for the module are a basic under- +standing of statistical modeling, probability distributions, and elementary calculus. +The remainder of the paper is organized as the following. In Section 2, we provide an +overview of variational inference essentials that can be readily used as a basis for a lecture +instruction. Section 3 presents a motivating example and the Gamma-Poisson model for +count data that serves as the first hands-on class activity with variational inference. In +Section 4, we offer a realistic use case for variational inference based on a Latent Dirichlet +Allocation model for document clustering with implementation details in R, which can be +used as a computing lab. We end the paper in Section 5 with a few concluding remarks. +2 +Lecture: Foundations of Variantional Inference +In this section, we introduce concepts and definitions of variational inference in Section +2.1, discuss the choices of variational families in Section 2.2, and present details of ELBO +optimization in Section 2.3. We also include recommendations of variational families and +ELBO optimization strategies with pedagogical considerations for an undergraduate audi- +ence. Instructors can design their lecture based on these materials tailored to their needs. +2.1 +Concepts and Definitions +The main idea behind variational inference is to approximate the target probability density +p(θ | y) by a member of some relatively simple family of densities q(θ | λ), indexed by +the variational parameter λ ∈ Rm′ (m′ ≥ 1), over the space of model parameters θ. Note +that λ = (λ1, . . . , λm) has m components of (potentially) varying dimensions. Variational +approximation is done by finding the member of variational family that minimizes the +Kullback-Leibler (KL) divergence of q(θ | λ) from p(θ | y): +q∗ = arg min +q(θ|λ) +KL(q(θ | λ)||p(θ | y)), +(2) +with KL divergence being the expectation of the log ratio between the q(θ | λ) and p(θ | y) +with respect to q(θ | λ): +KL(q(θ | λ)||p(θ | y)) = Eq +� +log q(θ | λ) +p(θ | y) +� +. +(3) +4 + +Figure 1: Illustration of variational inference as the optimization-based approximation. The +goal of variational inference is to find a member of the variational family that minimizes +KL divergence with the target distribution. +The KL divergence measures how different is the probability distribution q(θ | λ) from +p(θ | y) (Kullback & Leibler 1951). Note that while we use the KL divergence to measure +the similarity between two densities, it is not a metric because the KL divergence is not +symmetric and does not satisfy the triangle inequality. In fact, the order of q(θ | λ) and +p(θ | y) in Equation (2) is deliberate as it leads to taking the expectation with respect +to the variational distribution q(θ | λ). One can naturally think of reversing the roles of +q(θ | λ) and p(θ | y). However, this leads to a “different kind” of variational inference called +expectation propagation (Minka (2001)), which loses computational efficiency of variational +inference defined in Equation (2). +In a nutshell, rather than sampling, variational inference approximates densities us- +ing optimization. See Figure 1 for a graphical illustration, i.e., by finding the values of +variational parameters from λinit to λ∗ through optimization which lead to a variational +distribution q(θ | λ) that is close to the target posterior distribution p(θ | y) defined by +the smallest KL divergence. Finding the optimal q∗ is done in practice by maximizing an +equivalent objective function, L(λ), the evidence lower bound (ELBO), because the KL +5 + +q(0|入) +Optimization +init +Smallest KL +p(0ly)divergence is intractable as it requires the evaluation of the marginal distribution p(y): +L(λ) = +Eq[log p(y, θ) − log q(θ|λ)] += +Eq[log p(y|θ)] +� +�� +� +Expected log-likelihood of data +− +KL(q(θ|λ)||p(θ)) +� +�� +� +KL div. between the variational and prior densities +. +(4) +The ELBO is the sum between the negative KL divergence of the variational density from +the target density and the log of the marginal density p(y). Since the term log p(y) is +constant with respect to q(θ | λ), the objective functions in Equation (3) and Equation (4) +are equivalent. Examining the ELBO also reveals the intuition behind variational inference. +On the one hand, the first term in Equation (4) encourages the variational approximation +to place mass on parameter values that maximize the sampling density p(y | θ). On the +other hand, the second term in Equation (4) prefers closeness of the variational density to +the prior. Therefore, the ELBO shows a similar tension between the sampling density and +the prior known in Bayesian inference. +2.2 +Variational Families with Pedagogical Recommendations +We now move on to the implementation details of variational inference starting with the +selection of the variational family q(θ | λ). This choice is crucial as it affects the complexity +of optimization outlined in Section 2.1 as well as the quality of variational approximation. +Mean-field Variational Family +By far the most popular is the mean-field variational family which assumes that all the +unknown parameters are mutually independent, each approximated by its own univariate +variational density: +q(θ | λ) = +m +� +i=1 +q(θi | λi). +(5) +For example, a typical choice for real-valued parameters is the normal variational family +q(θ | µ, σ2) and the log-normal or Gamma for non-negative parameters. The main advan- +tage of the mean-field family is in its simplicity as it requires only a minimum number +of parameters to be estimated (no correlation parameters) and often leads to uncompli- +cated optimization. However, the mutually independent parameter assumption comes at +6 + +a price because the mean-field family cannot capture relationships between model param- +eters. To illustrate the pitfalls of mean-field approximation, consider a simple case of a +two-dimensional normal target density with highly correlated components. Figure 2 shows +the optimal mean-field variational approximation given by the product of two normal den- +sities. One can clearly see that the optimal variational densities match well with the means +of the target density, but the marginal variances are underestimated. To further understand +this common flaw of mean-field approximation, consider the definition of KL divergence +in Equation (3). The objective function penalizes more larger density in q(θ | λ) in areas +where p(θ | y) has low density than the opposite direction (recall that the expectation is +taken with respect to the variational density). +Recommendation for Undergradudate Instruction +It is worth noting that the development of new variational families which improves on the +trade-off between complexity and expressiveness of variational approximations has been +a fruitful and active area of research. To keep the scope of the undergraduate one-week +variational inference module manageable to both the students and the instructors, we +recommend solely focusing on the mean-field approximation. For interested students who +want to explore further, we encourage the instructors to refer them to the recent work +of Ambrogioni et al. (2021) that provides a detailed discussion on many state-of-the-art +variational families and their associated implementation challenges. +2.3 +ELBO Optimization with Pedagogical Recommendations +Besides the choice of variational family, another key implementation detail to address is +the way in which we find the member of the variational family that maximizes the ELBO. +Since this is a fairly general optimization problem, one can in principle use any optimization +procedure. In the variational inference literature, the coordinate ascent and the gradient +ascent procedures are the most prominent and widely used (Blei et al. (2017)). +7 + +3 +2 +1 +0 +1 +2 +3 +1 +3 +2 +1 +0 +1 +2 +3 +2 +2 +0 +2 +1 +0.0 +0.2 +0.4 +0.6 +Density +2 +0 +2 +2 +Target density +M-F approximation +Figure 2: Mean-field variational approximation of a two-dimensional normal target density. +The figure illustrates the common pitfall of the mean-field approximation in situations with +correlated model parameters. +8 + +Corrdinate Ascent +The coordinate ascent approach is based on the simple idea that one can maximize ELBO, +which is a multivariate function, by cyclically maximizing it along one direction at a time. +Starting with initial values (denoted by superscript 0) of the m variational parameters λ0 +λ0 = (λ0 +1, . . . , λ0 +m), +one obtains the (k + 1)th updated value of variational parameters by iteratively solving +λk+1 +i += arg max +x +L(λk+1 +1 +, . . . , λk+1 +i−1 , x, λk +i+1, . . . , λk +m), +which can be accomplished without using gradients (Blei et al. 2017). +Gradient Ascent +Variational inference via gradient ascent uses the standard iterative optimization algorithm +based on the idea that the ELBO grows fastest in the direction of its gradient (Hoffman +et al. 2013). In particular, the update of variational parameters λ at the (k +1)th iteration +is given by +λk+1 ← λk + η × ∇λL(λk), +where ∇λL(λ) is the ELBO gradient, and η is the step size which is also called the learning +rate. The step size controls the rate at which one updates the variational parameters. +For both coordinate and gradient ascent, we typically declare convergence of variational +parameters once the change in ELBO falls below some small threshold (Blei et al. 2017). +Recommendation for Undergradudate Instruction +Our recommendation for an undergraduate variational inference module is to take the +route of gradient ascent. This pedagogical choice is guided by our combined experience of +teaching statistical modeling, Bayesian statistics, and data science at various undergradu- +ate levels to students with diverse statistical backgrounds. Our recommendation has also +taken into account the pedagogical advantages and disadvantages of gradient ascent and +coordinate ascent for undergraduates: Variational inference via coordinate ascent, while +conceptually straightforward, requires non-trivial and model-specific derivations which can +9 + +easily obscure the overall goal of this one-week module to expand students’ exposure to +the state-of-the-art approximate inference for probabilistic models; gradient-based varia- +tional inference, in contrast, leads to a black-box optimization that does not require any +model-specific derivations due to an extensive autodifferentiation capabilities of modern +statistical software such as RStan (Stan Development Team 2022) and Python packages +PyTorch (Paszke et al. 2019) and TensorFlow (Abadi et al. 2015), to name a few. +We believe that from an undergraduate-level pedagogical perspective, gradient descent +reflects better the current data science pipeline and allows the instruction to be focused +on conceptual understanding of variational inference rather than technical details. +Of +course, using gradient-based optimization requires the students to be familiar with partial +derivatives. Such a pre-requisite potentially restricts the audience for our module to a course +with a multivariable calculus prerequisite. Nevertheless, we believe that an instructor with +sufficient preparation can explain the basics behind gradient ascent to an audience with a +minimal calculus background. +3 +Class Activity: +A Probabilistic Model for Count +Data with Variational Inference +In this section, we provide a fully developed hands-on class activity with variational in- +ference for count data. Starting with a motivating example in Section 3.1, we give an +overview of the Gamma-Poisson model in Section 3.2, and discuss details of the variational +inference of this model in Section 3.3, illustrated with an R Shiny app we have developed +for instruction purpose. Instructors can adopt and adapt this class activity based on these +materials tailored to their needs. +3.1 +A Motivating Example +To illustrate how ELBO optimization leads to a good approximation of target posterior +distribution, we consider Poisson sampling with a Gamma prior, which is a popular one- +parameter model for count data (Gelman et al. 2013, Albert & Hu (2019), Johnson et al. +(2022)). To get started, we provide the following motivating example: +10 + +Our task is to estimate the average number of active users of a popular mas- +sively multiplier online role-playing game (mmorpg) playing between the peak +evening hours 7 pm and 10 pm. This information can help game developers +in allocating server resources and optimizing user experience. To estimate the +average number of active users, we will consider the counts (in thousands) of +active players collected during the peak evening hours over a two-week period in +the past month. +We have chosen the Gamma-Poisson model as the probabilistic model in this class ac- +tivity for two reasons. First, the Gamma-Poisson model is relatively easy to understand for +students with an elementary knowledge of probability distributions. Second, the Gamma +is a conjugate prior for Poisson sampling which means that one can derive the exact poste- +rior distribution (another Gamma) and check the fidelity of variational approximation by +comparing to the analytical Gamma solution. The learning objective of this class activity +is to get students familiarized with various aspects of variational inference presented in Sec- +tion 2, such as ELBO and variational family, with a simple example. Afterwards, students +are better prepared to move on to more realistic scenarios, such as document clustering, +described in Section 4. +3.2 +Overview of the Gamma-Poisson Model +We now provide an overview of the Gamma-Poisson model which can be readily turned +into a class lecture. Suppose that y = (y1, . . . , yn) represent the observed counts in n time +intervals where the counts are independent, and each yi follows a Poisson distribution with +the same rate parameter θ > 0. The joint probability mass function of y = (y1, . . . , yn) is +p(y | θ) = +n +� +i=1 +p(yi | θ) ∝ θ +�n +i=1 yie−nθ. +(6) +The posterior distribution for the rate parameter θ is our inference target as θ represents +the expected number of counts that occur during the given time intervals. Note that the +Poisson sampling relies on several assumptions about the sampling process. +First, one +assumes that the time interval is fixed. Second, the counts occurring during different time +11 + +intervals are independent. Lastly, the rate θ at which the counts occur is constant over +time. +The Gamma-Poisson conjugacy states that if θ follows a Gamma prior distribution with +shape and rate parameters α and β, it can be shown that the posterior distribution p(θ | y) +will also have a Gamma density. Namely, if +θ ∼ Gamma(α, β), +(7) +then +θ | y ∼ Gamma(α + +n +� +i=1 +yi, β + n). +(8) +In other words, given α, β, and y, one can derive the analytical solution to the posterior +of p(θ | y) and can subsequently sample from Gamma(α + �n +i=1 yi, β + n) to get posterior +samples of θ. While no approximation is needed, it serves as a good example of illustrating +how variational inference works in such a setting and allows evaluations of the performance +of variational inference. +3.3 +Variational Inference of the Gamma-Poisson Model +Recall from Section 2 that variational inference approximates the (unknown) posterior +distribution of a parameter by a simple family of distributions. In this case, we will ap- +proximate the posterior distribution p(θ | y) by a log-normal distribution with mean µ and +standard deviation σ: +q(θ | µ, σ) = +1 +θσ +√ +2πe− (ln θ−µ)2 +2σ2 +. +(9) +The log-normal distribution is a continuous probability distribution of a random variable +whose logarithm is normally distributed. It is a popular variational family for non-negative +parameters because it can be expressed as a (continuously) transformed normal distribu- +tion, and therefore it is amenable to automatic differentiation. Automatic differentiation is +a computation method for derivatives in computer programs that relies on the application +of chain rule in differential calculus. It provides accurate and fast numerical derivative +evaluations that leads to machine learning algorithms (such as variational inference) that +do not require users to manually work out and code derivatives (Kucukelbir et al. 2017, +Baydin et al. (2018)). +12 + +0.0 +0.1 +0.2 +0.3 +40 +50 +60 +θ +Density +Prior +True posterior +log−normal(3.7, 0.05) +ELBO maximization: log−normal(3.9, 0.04) +Figure 3: Variational approximation based on the motivating scenario of mmorpg’s player +activity. The true Gamma(792, 100) posterior and the prior Gamma(100,2) distributions +are included. +In the supplementary materials, we provide an accompanying in-class handout and an R +Shiny app based on the motivating scenario of mmorpg described in Section 3.1. The first +two parts of the handout present the motivating example and the overview of the Gamma- +Poisson model. In the third part of the handout, students carry out exact posterior inference +for the unknown rate parameter θ using a small dataset of observed counts of mmorpg’s +active players. In the fourth and final part, students find variational approximation of +p(θ | y) and check how well their approximation matches the true posterior distribution. +Figure 3 shows the final variational approximation compared to the true Gamma(792, 100) +posterior distribution from the handout example. We can see, on the one hand, the resulting +log-normal(3.9, 0.04) distribution (the black dash line) that maximizes the ELBO visually +overlaps with the true posterior (ELBO = −42.52, KL divergence < 0.001). On the other +hand, another member of the variational family, the log-normal(3.7, 0.05) distribution (the +blue dot-dash line; with ELBO = −57.55 and KL divergence = 15.085), clearly differs from +the target. This example illustrates the good performance of variational inference through +optimization for the Gamma-Poisson count model. +13 + +The design of this class activity is guided by the active-learning principles listed in +Section 1 and the goal is to give students their first hands-on experience with variational +inference without the need of coding. Specifically, we include open-ended questions that +focus on problem-solving and create opportunities for students to collaborate with peers. +Moreover, the accompanying R Shiny app provides appropriate and sufficient scaffolding so +that students can concentrate on conceptual understanding instead of the technical details, +which follows our pedagogical recommendations in Section 2. +We now turn to a guided R lab to illustrate the use of variational inference for a more +realistic use case of document clustering, applied to a sample of Associated Press newspaper +articles. +4 +Lab: Document Clustering +Among the many models approximated by variational inference techniques, Latent Dirichlet +Allocation (LDA) might be one of the most popular (Blei et al. 2003). LDA is a mixed- +membership clustering model, commonly used for document clustering. Specifically, LDA +models each document to have a mixture of topics, where each word in the document is +drawn from a topic based on the mixing proportions (Stan Development Team 2022). While +the LDA model is relatively easy and straightforward to follow, using conventional MCMC +estimation techniques has proven to be too computationally demanding due to the large +number of parameters involved. Therefore, researchers and practitioners turn to variational +inference techniques when using LDA for document clustering (Blei et al. 2003). +In Section 4.1, we briefly introduce the LDA model following the presentation in Stan +Development Team (2022). In Section 4.2, we present an LDA application to a collection +of Associate Press newspaper articles where variational inference is implemented by the +cmdstanr R package. For brevity, we focus on the interpretation of results and discuss +pedagogical considerations and leave a Stan script for the LDA model and the details of +the guided lab assignment with R code in the supplementary materials. +14 + +4.1 +Overview of the LDA model +The LDA model considers K topics for M documents made up of words drawn from a +vocabulary of V distinct words. For a document m, a topic distribution θm over K topics +is drawn from a Dirichlet distribution, +θm ∼ Dirichlet(α), +(10) +where �K +k=1 θm,k = 1 (0 ≤ θm,k ≤ 1) and α is a vector of length K with positive values. +Each of the Nm words {wm,1, . . . , wm,Nm} in document m is then generated indepen- +dently conditional on θm. To do so, first, the topic zm,n for word wm,n in document m is +drawn from +zm,n ∼ categorical(θm), +(11) +where θm is the document-specific topic-distribution defined in Equation (14). +Next, the word wm,n in document m is drawn from +wm,n ∼ categorical(φz[m,n]), +(12) +which is the word distribution for topic zm,n. Note that z[m, n] in Equation (16) refers to +zm,n. +Lastly, a Dirichlet prior is given to distributions φk over words for topic k as +φk ∼ Dirichlet(β), +(13) +where β is the prior a vector of length V (i.e., the total number of words) with positive +values. Figure 10 shows a graphical model representation of LDA. +4.2 +Clustering of Associated Press Newspaper Articles +As a realistic application of variational inference, we consider a collection of 2246 Asso- +ciated Press newspaper articles to be clustered using the LDA model. +The dataset is +(conveniently) part of the topicmodels R package. We believe this dataset is well suited +to demonstrate the capabilities of variational inference in the classroom as it is too large +for the MCMC approximation to be feasible but small enough for the variational inference +to take just a few minutes to converge. For brevity, we highlight the results based on a +15 + +Figure 4: Graphical model representation of LDA. The largest box represents the docu- +ments. On the left, the inner box represents the topics and words within each document. +On the right, the box represents the topics. +two-topic LDA model (i.e., K = 2) and leave the details to the guided lab in the sup- +plementary materials. The number of topics is set to 2 for demonstration purposes and +simplicity of interpretations. Comparing LDA with a different number of topics is often +done with metrics such as semantic coherence or held-out data likelihood (Mimno et al. +2011). While such a comparison is beyond the scope of this lab, interested students are +encouraged to explore mentored by the instructors. +Figure 5 shows the evolution of ELBO for the two-topic LDA model which converged +after a little bit over 100 iterations of the gradient ascent algorithm described in Section +2.3. On a standard laptop computer, this typically takes between 5-10 minutes depending +on the CPU speed. We recommend running the algorithm repeatedly (i.e., 2-3 times) with +a different random seed in the classroom and discussing the dependency of variational +inference on initial values of variational parameters which can occur in practice. +Figures 6 and 7 are examples of graphical displays of the topics that were extracted +from the collection of articles based with the LDA. In particular, Figure 6 shows the 10 +most common words for each topic; that is, the parts of distribution φk, for k ∈ {1, 2}, +with the largest mass. Figure 7 displays similar information for the 20 most common words +for each topic in the form of a word cloud. The most common words in topic 1 include +people, government, president, police, and state, suggesting that this topic may represent +political news. In contrast, the most common words in topic 2 include percent, billion, +16 + +N +α +m +β +K +M−1120000 +−1100000 +−1080000 +−1060000 +−1040000 +0 +30 +60 +90 +Iteration +ELBO +Figure 5: The evolution of ELBO for the two-topic LDA model based on 2246 Associated +Press newspaper articles. +million, market, American, and states, hinting that this topic may represent news about +the US economy. +5 +Concluding Remarks +In this paper, we present a newly-developed one-week course module that exposes un- +dergraduate students to approximation via variational inference. The proposed module +is self-contained in the sense that it encourages and empowers potential instructors to +adopt and adapt the module as we provide an overview of variational inference, an active- +learning-based class activity with an R Shiny app, and a guided lab based on a realis- +tic application with R code (see the supplementary materials or https://github.com/ +kejzlarv/variational_inference_module). Its design is rooted in the best practices of +active learning that have been demonstrated to improve student learning and engagement. +The module can be integrated into any intermediate- or upper-level undergraduate +course where students learn probabilistic models (including logistic regression, Bayesian +classifiers, neural networks, or models for natural language processing), such as Bayesian +statistics, multivariate data analysis, and data science courses. The applications discussed +17 + +Topic 1 +Topic 2 +0.000 +0.005 +0.010 +0.015 +0.020 +0.025 +0.000 +0.005 +0.010 +0.015 +0.020 +0.025 +american +billion +first +states +united +last +million +year +new +percent +told +officials +soviet +state +government +police +two +president +people +i +Word distributions ( ϕ ) +Word +Figure 6: Word distributions based on the two-topic LDA model. The 10 most common +words are displayed. +Figure 7: World clouds consisting of the 20 most common words for each of the two topics +extracted by the LDA. +18 + +market company +south house +first +t united +partygovernment +billion +daylastyear +I federal +told +people +rs +york +eal +percentreport +twO I police city +@national +bush +new +news +president +states +court +soviet +military +million +S +week +officials say +american +time +thursdayin these courses are typically limited to scenarios with relatively small datasets, since +the required use of MCMC does not scale well to large datasets. Given the popularity +and scalability of variational inference, we hope that undergraduate instructors adopting +and adapting this module will be able to integrate more realistic and fun use cases in +their classrooms. Moreover, the references and further readings provided in this paper are +readily available resources for a deeper dive of variational inference by interested students +with appropriate mentoring by their undergraduate instructors. +References +Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G. S., Davis, +A., Dean, J., Devin, M., Ghemawat, S., Goodfellow, I., Harp, A., Irving, G., Isard, M., +Jia, Y., Jozefowicz, R., Kaiser, L., Kudlur, M., Levenberg, J., Man´e, D., Monga, R., +Moore, S., Murray, D., Olah, C., Schuster, M., Shlens, J., Steiner, B., Sutskever, I., +Talwar, K., Tucker, P., Vanhoucke, V., Vasudevan, V., Vi´egas, F., Vinyals, O., Warden, +P., Wattenberg, M., Wicke, M., Yu, Y. & Zheng, X. (2015), ‘TensorFlow: Large-scale +machine learning on heterogeneous systems’. Software available from tensorflow.org. +URL: https://www.tensorflow.org/ +Albert, J. & Hu, J. (2019), Probability and Bayesian Modeling, 1 edn, Chapman and +Hall/CRC. +Albert, J. & Hu, J. (2020), ‘Bayesian computing in the undergraduate statistics curriculum’, +Journal of Statistics Education 28, 236–247. +Ambrogioni, L., Lin, K., Fertig, E., Vikram, S., Hinne, M., Moore, D. & van Gerven, +M. (2021), Automatic structured variational inference, in A. Banerjee & K. Fukumizu, +eds, ‘Proceedings of The 24th International Conference on Artificial Intelligence and +Statistics’, Vol. 130 of Proceedings of Machine Learning Research, PMLR, pp. 676–684. +URL: https://proceedings.mlr.press/v130/ambrogioni21a.html +Bardenet, R., Doucet, A. & Holmes, C. (2017), ‘On markov chain monte carlo methods for +tall data’, J. Mach. Learn. Res. 18(1), 1515–1557. +19 + +Baydin, A. G., Pearlmutter, B. A., Radul, A. A. & Siskind, J. M. (2018), ‘Automatic +differentiation in machine learning: a survey’, Journal of Machine Learning Research +18(153), 1–43. +URL: http://jmlr.org/papers/v18/17-468.html +Blei, D. M., Kucukelbir, A. & McAuliffe, J. D. (2017), ‘Variational inference: A review for +statisticians’, Journal of the American Statistical Association 112(518), 859–877. +URL: https://doi.org/10.1080/01621459.2017.1285773 +Blei, D. M., Ng, A. Y. & Jordan, M. I. (2003), ‘Latent dirichlet allocation’, Journal of +Machine Learning Research 3, 993–1022. +Deslauriers, L., McCarty, L. S., Miller, K., Callaghan, K. & Kestin, G. (2019), ‘Measuring +actual learning versus feeling of learning in response to being actively engaged in the +classroom’, Proceedings of the National Academy of Sciences 116(39), 19251–19257. +URL: https://www.pnas.org/doi/abs/10.1073/pnas.1821936116 +Dogucu, M. & Hu, J. (2022), ‘The current state of undergraduate bayesian education and +recommendations for the future’, The American Statistician 76(4), 405–413. +URL: https://doi.org/10.1080/00031305.2022.2089232 +Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H. & Wen- +deroth, M. P. (2014), ‘Active learning increases student performance in science, engineer- +ing, and mathematics’, Proceedings of the National Academy of Sciences 111(23), 8410– +8415. +URL: https://www.pnas.org/doi/abs/10.1073/pnas.1319030111 +Gelman, A., Carlin, J., Stern, H., Dunson, D., Vehtari, A. & Rubin, D. (2013), Bayesian +Data Analysis, third edn, CRC Pres. +URL: https://books.google.com/books?id=ZXL6AQAAQBAJ +Hoffman, M. D., Blei, D. M., Wang, C. & Paisley, J. (2013), ‘Stochastic variational infer- +ence’, Journal of Machine Learning Research 14, 1303–1347. +URL: http://jmlr.org/papers/v14/hoffman13a.html +20 + +Johnson, A. A., Ott, M. & Dogucu, M. (2022), Bayes Rules! An Introduction to Applied +Bayesian Modeling, 1 edn, Chapman and Hall/CRC. +Jordan, M. I., Ghahramani, Z., Jaakkola, T. S. & Saul, L. K. (1999), ‘An introduction to +variational methods for graphical models’, Machine Learning 37, 183–233. +Kucukelbir, A., Tran, D., Ranganath, R., Gelman, A. & Blei, D. M. (2017), ‘Automatic +differentiation variational inference’, Journal of Machine Learning Research 18(14), 1– +45. +URL: http://jmlr.org/papers/v18/16-107.html +Kullback, S. & Leibler, R. A. (1951), ‘On Information and Sufficiency’, The Annals of +Mathematical Statistics 22(1), 79 – 86. +URL: https://doi.org/10.1214/aoms/1177729694 +Michael, J. (2006), ‘Where’s the evidence that active learning works?’, Advances in Physi- +ology Education 30(4), 159–167. PMID: 17108243. +URL: https://doi.org/10.1152/advan.00053.2006 +Mimno, D., Wallach, H. M., Talley, E., Leenders, M. & McCallum, A. (2011), Optimizing +semantic coherence in topic models, in ‘Proceedings of the Conference on Empirical +Methods in Natural Language Processing’, EMNLP ’11, Association for Computational +Linguistics, USA, p. 262–272. +Minka, T. P. (2001), Expectation propagation for approximate bayesian inference, in ‘Pro- +ceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence’, UAI’01, +Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, p. 362–369. +Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, +Z., Gimelshein, N., Antiga, L., Desmaison, A., Kopf, A., Yang, E., DeVito, Z., Raison, +M., Tejani, A., Chilamkurthy, S., Steiner, B., Fang, L., Bai, J. & Chintala, S. (2019), +Pytorch: An imperative style, high-performance deep learning library, in ‘Advances in +Neural Information Processing Systems 32’, Curran Associates, Inc., pp. 8024–8035. +URL: +http://papers.neurips.cc/paper/9015-pytorch-an-imperative-style-high- +performance-deep-learning-library.pdf +21 + +Peterson, C. & Anderson, J. R. (1987), ‘A mean field theory learning algorithm for neural +networks’, Complex Systems 1, 995–1019. +Rudoy, D. & Wolfe, P. J. (2006), Monte carlo methods for multi-modal distributions, in +‘2006 Fortieth Asilomar Conference on Signals, Systems and Computers’, pp. 2019–2023. +Schwab-McCoy, A., Baker, C. M. & Gasper, R. E. (2021), ‘Data science in 2020: Com- +puting, curricula, and challenges for the next 10 years’, Journal of Statistics and Data +Science Education 29(sup1), S40–S50. +URL: https://doi.org/10.1080/10691898.2020.1851159 +Stan Development Team (2022), Stan Modeling Language User’s Guide and Reference Man- +ual, Version 2.31. +URL: http://mc-stan.org/ +22 + +Supplementary Materials for Introducing Variational Inference In +Undergraduate Statistics and Data Science Curriculum +The supplementary materials include: 1) Details of the class activity on probabilistic +model for count data with variational inference, introduced in Section 3 in the main text; +2) The manual of the R shiny app we have developed for the module, mentioned in Section +3 in the main text; and 3) Details of the guided R lab of the LDA application to a sample +of the Associated Press newspaper articles with variational inference, presented in Section +4 in the main text. +6 +Class Activity: Probabilistic Model for Count Data +with Variational Inference +The goal of this activity is to illustrate variational inference on a simple example of Gamma- +Poisson conjugate model, which is a popular model for count data. +6.1 +A Motivating Example +Our task is to estimate the average number of active users of a popular massively multiplier +online role-playing game (mmorpg) playing between the peak evening hours 7 pm and 10 +pm. +This information can help the game developers in allocating server resources and +optimizing user experience. To make this estimate, we will consider the following counts +(in thousands) of active players collected during the peak evening hours over a two-week +period past month. +Sun +Mon +Tue +Wed +Thu +Fri +Sat +Week 1 +50 +47 +46 +52 +49 +55 +53 +Week 2 +48 +45 +51 +50 +53 +46 +47 +6.2 +Overview of the Gamma-Poisson Model +Sampling density: +23 + +Suppose that y = (y1, . . . , yn) represent the observed counts in n time intervals where +the counts are independent, then each yi follows a Poisson distribution with rate θ > 0. +Namely, +yi | θ ∼ Poisson(θ) +• E(yi | θ) = θ +• Var(yi | θ) = θ +Prior distribution: +θ ∼ Gamma(α, β) +• α > 0 is the shape parameter +• β > 0 is the rate parameter +• E(θ) = α +β +• Var(θ) = +α +β2 +Posterior distribution: +θ | y1, . . . , yn ∼ Gamma(α + +n +� +i=1 +yi, β + n) +6.3 +Exact Inference with the Gamma-Poisson Model +We will start by selecting a prior distribution for the unknown average number of active +users. Suppose that we elicit an expert’s advice on the matter, and they tell us that a +similar mmorpg has typically about 50,000 users during peak hours. However, they are +not too sure about that, so the interval between 45,000 and 55,000 users should have a +reasonably high probability. Suppose that we elicit an expert’s advice on the matter, and +they tell us that a similar mmorpg has typically about 50,000 users during peak hours. +However, they are not too sure about that, so the interval between 45,000 and 55,000 users +should have a reasonably high probability. This reasoning leads to a Gamma(100, 2) as a +reasonable prior for the average number of active users. +24 + +Task 1: Explain the reasoning behind using Gamma(100, 2) as the prior distri- +bution. +Task 2: Use the information above to find the exact posterior distribution for +the average number of active users. +Task 3: What are the mean and standard deviation of the posterior distribu- +tion that you just obtained? What is your recommendation about the typical +number of active users playing the mmorpg between the peak evening hours +7pm and 10pm? +6.4 +Variational Inference with the Gamma-Poisson Model +Variational inference approximates the (unknown) posterior distribution of a parameter +by a simple family of distributions. In this case, we will try to approximate the posterior +distribution of the mmorpg’s average number of active users between the peak hours θ by +a log-normal distribution with mean µ and standard deviation σ. Log-normal distribution +is a continuous probability distribution of a random variable whose logarithm is normally +distributed. It also happens to be a popular variational family for non-negative parameters +as it is amenable to autodifferentiation. Since we know exactly how the posterior distri- +bution for Gamma-Poisson model looks like, we will be able to check the fidelity of the +variational approximation. Use the accompanying applet titled Variational Inference with +Gamma-Poisson Model for count data to complete the following task. +25 + +Task 4: Use the sliders in the applet to manually find the member of a log- +normal variational family that well approximates the posterior distribution of +θ. What is your strategy? +Task 5: Compare your approximation with a neighbor. Whose approximation +is closer to the exact posterior distribution of θ? How are you deciding? +Task 6: Check the Fit a variational approximation box in the applet to find +the variational approximation using the gradient ascent algorithm. How close +was the variational approximation that you found manually to the one found +here? +7 +Manual of the R shiny app +This document describes the elements of R Shiny applet that accompanies the “Proba- +bilistic Model for Count Data with Variational Inference” class activity. Note that the +numbering in Section 7.1 and Section 7.2 corresponds to the numbered boxes in Figure 8 +and Figure 9. +7.1 +Manual search for variational approximation +1. Sliders to control the mean µ and the standard deviation σ of log-normal variational +family. +2. The ELBO and KL divergence values for variational approximation based on the +mean and standard deviations selected in box 1. +3. A plot that displays the true Gamma(792, 100) posterior distribution, the Gamma(100, 2) +prior distribution, and the variational approximation based on the selection in box 1. +4. A checkbox to display the results of ELBO maximization via gradient ascent algo- +rithm. The resulting variational approximation is plotted in box 3. +26 + +Figure 8: The applet is based on the class activity presented in Section 1 of the supplemen- +tary materials. The applet visual before checking the “Fit a variational approximation“ +checkbox is displayed. +7.2 +Variational approximation based on ELBO maximization +5. The resulting mean µ, standard deviation σ, and ELBO values of variational approx- +imation based on ELBO maximization. +6. A plot depicting ELBO values for each iteration of the gradient ascent algorithm. +8 +Lab: Document Clustering +The goal of this lab is to gain a practical experience with variational inference on a real- +istic use case based on Latent Dirichlet Allocation (LDA) and implement the model in R +applied to a dataset of documents. To do so, we consider a collection of 2246 Associated +Press newspaper articles to be clustered using the LDA model. The dataset is part of the +topicmodels R package. You can load the dataset AssociatedPress with the following R +command. +data("AssociatedPress", package = "topicmodels") +27 + +Variational Inference with Gamma-Poisson Model for count data +Variational approximation using log-normal variational family: +Prior +True posterior-. VI- Manual +3 +μ +1 +0.3- +3.5 +3.7 +4.2 +3.82 +3.9 +4.14 4.2 +0.2 +6 +PDF +0.01 +0.05 +0.1 +0.1 +ELBO value: +2 +-57.496 +0.0 +40 +50 +KL Divergence value: +60 +0 +15.031 +OFit a variational approximation +4Figure 9: The applet visual after checking the “Fit a variational approximation“ checkbox +is displayed. +28 + +Variational Inference with Gamma-Poisson Model for count data +Variational approximation using log-normal variational family: +:Prior +True posterior -: VI-Manual - VI -ELBO maximization +μ +0.3- +3.5 +3.7 +4.2 +3.5 + 3.58 3.66 3.74 3.82 3.9 3.984.064.14 4.2 +0.2 +PDF +0.01 +0.05 +0.1 +0.1 +0.010.020.030.040.050.060.070.080.090.1 +ELBO value: +-57.496 +0.0- +40 +50 +60 +KL Divergence value: +15.031 + Fit a variational approximation +0 +D +Results of ELBo maximization via gradient ascent: +5 +μ +-50 +3.901 +0 +.BO +-100 +0.035 +E +ELBO value: +-150 +-42.52 +-200 +0 +50 +100 +150 +200 +Iteration8.1 +Overview of the LDA model and Stan script +The LDA is a mixed-membership clustering model, commonly used for document clustering. +LDA models each document to have a mixture of topics, where each word in the document is +drawn from a topic based on the mixing proportions. Specifically, the LDA model assumes +K topics for M documents made up of words drawn from V distinct words. For document +m, a topic distribution θm is drawn over K topics from a Dirichlet distribution, +θm ∼ Dirichlet(α), +(14) +where �K +k=1 θm,k = 1 (0 ≤ θm,k ≤ 1) and α is the prior a vector of length K with positive +values. +Each of the Nm words {wm,1, . . . , wm,Nm} in document m is then generated indepen- +dently conditional on θm. To do so, first, the topic zm,n for word wm,n in document m is +drawn from +zm,n ∼ categorical(θm), +(15) +where θm is the document-specific topic-distribution defined in Equation (14). +Next, the word wm,n in document m is drawn from +wm,n ∼ categorical(φz[m,n]), +(16) +which is the word distribution for topic zm,n. Note that z[m, n] in Equation (16) refers to +zm,n. +Lastly, a Dirichlet prior is given to distributions φk over words for topic k as +φk ∼ Dirichlet(β), +(17) +where β is the prior a vector of length V (i.e., the total number of words) with positive +values. Figure 10 shows a graphical model representation of LDA. +Below, we include the Stan script for the LDA model provided by Stan Development +Team (2022). Note that Stan supports the calculation of marginal distributions over the +continuous parameters by summing out the discrete parameters in mixture models (Stan +Development Team 2022). This process corresponds to the gamma parameter in the Stan +script below. We refer interested readers to Stan Development Team (2022) for further +details. +29 + +Figure 10: Graphical model representation of LDA. The outer box represents the docu- +ments. On the left, the inner box represents the topics and words within each document. +On the right, the box represents the topics. +data { +int K; +// number of topics +int V; +// number of words +int M; +// number of docs +int N; +// total word instances +int w[N]; +// word n +int doc[N]; +// doc ID for word n +vector[K] alpha; +// topic prior vector of length K +vector[V] beta; +// word prior vector of length V +} +parameters { +simplex[K] theta[M]; +// topic distribution for doc m +simplex[V] phi[K]; +// word distribution for topic k +} +model { +for (m in 1:M) +theta[m] ~ dirichlet(alpha); +30 + +N +α +m +β +K +Mfor (k in 1:K) +phi[k] ~ dirichlet(beta); +for (n in 1:N) { +real gamma[K]; +for (k in 1:K) +gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]); +target += log sum exp(gamma); +// likelihood; +} +} +8.2 +Variational inference with the LDA model +For demonstration purposes, we shall start with a two-topic LDA model (i.e., K = 2). +Before that, we recommend removing the words from AssociatedPress datasets that are +rare using the function removeSparseTerms() from the tm package. These words have a +minimal effect on the LDA parameter estimation. Nevertheless, they increase the compu- +tational cost of variational inference and therefore should be removed using the following +R command. +dtm <- removeSparseTerms(AssociatedPress, 0.95) +We are now ready to fit the LDA model using variational inference capabilities of the +cmdstanr package. The following code achieves the goal: +LDA model cmd <- cmdstan model(stan file = "LDA.stan") +N TOPICS <- 2 +data <- list(K = N TOPICS, +V = dim(dtm)[2], +M = dim(dtm)[1], +N = sum(dtm$v), +w = rep(dtm$j,dtm$v), +31 + +doc = rep(dtm$i,dtm$v), +#according to Griffiths and Steyvers(2004) +alpha = rep(50/N TOPICS,N TOPICS), +beta = rep(1,dim(dtm)[2]) +) +vi fit <- LDA model cmd$variational(data = data, +seed = 1, +output samples = 1000, +eval elbo = 1, +grad samples = 10, +elbo samples = 10, +algorithm = "meanfield", +output dir = NULL, +iter = 1000, +adapt iter = 20, +save latent dynamics=TRUE, +tol rel obj = 10^-4) +The “LDA.stan” file contains the Stan script for the LDA model provided in Section +8.1. We recommend the usage of the R help to get familiar with the variational() method +of the cmdstan model() function. The variable vi fit contains the results of variational +approximation of the LDA parameters. For example, one can obtain the word distributions +for the each of the topics with vi fit$summary("phi"). +Finally, to access the ELBO values, use the following: +vi diag <- utils::read.csv(vi fit$latent dynamics files()[1], +comment.char = "#") +ELBO <- data.frame(Iteration = vi diag[,1], ELBO = vi diag[,3]) +32 + +Task 1: Use a graphical display to show the 10 most common words for each +of the two-topics and their probabilities. +Task 2: Use the function wordcloud() from the wordcloud package and display +the most common words for each of the topics as a world clowd. What kinds +of articles do these topics represent? +Task 3: Fit a three-topic LDA model, display the most common words for each +of the topics. How do the results differ from the two-topic LDA? +Task 4 (Advanced): Use the three-topic LDA model and diplay the topic preva- +lence among the 2246 Associated Press articles. That is, show what proportions +of articles fall under each topic. +All necessary R code for fitting the LDA model to the Associated Press sample, including +the graphical displays shown in the main text, is included in a separate R script file called +LDA LAB.R available as a part of the supplementary materials. We also include a printout +of the R script below for interested readers. +library(cmdstanr) +# Checking integrity of installation of cmdstanr +check cmdstan toolchain() +install cmdstan(cores = 2) +cmdstan path() +cmdstan version() +# Auxiliary packages +library(tm) +library(tidyverse) +library(tidytext) +library(topicmodels) +33 + +## Get data +data("AssociatedPress", package = "topicmodels") +## Removing rare words from the vocabulary +dtm <- removeSparseTerms(AssociatedPress, 0.95) +dim(dtm) +## Input for stan model +N TOPICS <- 2 +data <- list(K = N TOPICS, +V = dim(dtm)[2], +M = dim(dtm)[1], +N = sum(dtm$v), +w = rep(dtm$j,dtm$v), +doc = rep(dtm$i,dtm$v), +#according to Griffiths and Steyvers(2004) +alpha = rep(50/N TOPICS,N TOPICS), +beta = rep(1,dim(dtm)[2]) +) +### VB fit +LDA model cmd <- cmdstan model(stan file = "LDA.stan") +LDA model cmd$print() +vb fit <- LDA model cmd$variational(data = data, +seed = 1, +output samples = 1000, +eval elbo = 1, +grad samples = 10, +34 + +elbo samples = 10, +algorithm = "meanfield", +output dir = NULL, +iter = 1000, +adapt iter = 20, +save latent dynamics=TRUE, +tol rel obj = 10^-4) +# Plotting ELBO +vb diag <- utils::read.csv(vb fit$latent dynamics files()[1], +comment.char = "#") +ELBO <- data.frame(Iteration = vb diag[,1], +ELBO = vb diag[,3]) +ggplot(data = ELBO, aes(x = Iteration, y = ELBO)) + geom line(lwd=1.5) + +theme(text = element text(size = 20), +panel.background = element rect(fill = "transparent", +color = "lightgrey"), +panel.grid.major = element line(colour = "lightgrey")) + +xlim(0,110) +## Accessing parameters +vb fit$summary("theta") # dim: M-by-K +vb fit$summary("phi") # dim: K-by-V +## Word distribution per topic +V <- dim(dtm)[2] +odd rows <- rep(c(1,0), times = V) +Topic1 <- vb fit$summary("phi")[odd rows == 1,] +Topic2 <- vb fit$summary("phi")[odd rows == 0,] +35 + +word probs <- data.frame(Topic = c(rep("Topic 1", V), +rep("Topic 2", V)), +Word = rep(dtm$dimnames$Terms,N TOPICS), +Probability = c(Topic1$mean, Topic2$mean)) +# Selecting top 10 words per topic +top words <- word probs %>% group by(Topic) %>% top n(10) %>% +ungroup() %>% arrange(Topic, -Probability) +top words %>% +mutate(Word = reorder within(Word, Probability, Topic)) %>% +ggplot(aes(Probability, Word, fill = factor(Topic))) + +geom col(show.legend = FALSE) + +facet wrap(~ Topic, scales = "free") + +scale y reordered() + theme(text = element text(size = 15)) + xlim(0,0.025) + +xlab("Word distributions ( \u03d5 )") +# Word Cloud display +#install.packages("wordcloud") +library(wordcloud) +top words <- word probs %>% group by(Topic) %>% top n(20) %>% +ungroup() %>% arrange(Topic, -Probability) +mycolors <- brewer.pal(8, "Dark2") +wordcloud(top words %>% filter(Topic == "Topic 1") %>% .$Word , +top words %>% filter(Topic == "Topic 1") %>% .$Probability, +random.order = FALSE, +color = mycolors) +36 + +mycolors <- brewer.pal(8, "Dark2") +wordcloud(top words %>% filter(Topic == "Topic 2") %>% .$Word , +top words %>% filter(Topic == "Topic 2") %>% .$Probability, +random.order = FALSE, +color = mycolors) +37 + diff --git a/BtAzT4oBgHgl3EQfTfyw/content/tmp_files/load_file.txt b/BtAzT4oBgHgl3EQfTfyw/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..75501cbe10478da31e799945e4783d81720094f4 --- /dev/null +++ b/BtAzT4oBgHgl3EQfTfyw/content/tmp_files/load_file.txt @@ -0,0 +1,787 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf,len=786 +page_content='Introducing Variational Inference in Undergraduate Statistics and Data Science Curriculum Vojtech Kejzlar Department of Mathematics and Statistics,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Skidmore College and Jingchen Hu Department of Mathematics and Statistics,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Vassar College January 4,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2023 Abstract Probabilistic models such as logistic regression,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Bayesian classification,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' neural net- works,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' and models for natural language processing,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' are increasingly more present in the undergraduate statistics and data science curriculum due to their wide range of applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In this paper, we present a one-week undergraduate course module on variational inference, a popular optimization-based approach for approximate infer- ence with probabilistic models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Our proposed module is guided by active learning principles: In addition to lecture materials on variational inference, we provide an accompanying class activity, an R shiny app, and a guided lab based on a real data application of clustering documents using Latent Dirichlet Allocation with R code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The main goal of our module is to expose undergraduate students to a method that facilitates statistical modeling and inference with large datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Using our proposed module as a foundation, instructors can adopt it and adapt to introduce more realistic use cases and applications in data science, Bayesian statistics, multivariate analysis, and statistical machine learning courses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Keywords: Active learning, Bayesian Statistics, Statistical Computing, Probabilistic Mod- els, Undergraduate Curriculum 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='01251v1 [stat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='OT] 3 Jan 2023 1 Introduction With the recent and rapid expansion of undergraduate curricula with offerings in data science, Bayesian statistics, multivariate data analysis, and statistical machine learning, probabilistic models and Bayesian methods have grown to become more popular (Schwab- McCoy et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2021, Dogucu & Hu (2022)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In many settings, a central task in applications of probabilistic models is the evaluation of posterior distribution p(θ | y) of m model parameters θ ∈ Rm (m ≥ 1) conditioned on the observed data y = (y1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , yn) provided by the Bayes’ theorem p(θ | y) = p(y | θ)p(θ) p(y) ∝ p(y | θ)p(θ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (1) Here, p(y | θ) is the sampling density given by the underlying probabilistic model for data, p(θ) is the prior density that represents our prior beliefs about θ before seeing the data, and p(y) is the marginal data distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The posterior distribution p(θ | y), however, has closed form only in a limited number of scenarios (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', conjugate priors) and there- fore typically requires approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' By far the most popular approximation methods are Markov chain Monte Carlo (MCMC) algorithms including Gibbs sampler, Metropo- lis, Metropolis-Hastings, and Hamiltonian Monte Carlo (Gelman et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2013), to name a few.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' See Albert & Hu (2020) for a review of these algorithms in undergraduate Bayesian courses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' While useful for certain scenarios, these MCMC algorithms do not scale well with large datasets and can have a hard time approximating multimodal posteriors (Rudoy & Wolfe 2006, Bardenet et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2017)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Such challenges therefore limit the applications of probabilistic models that can be discussed in the classroom and restrict students’ exposure to more realistic use cases that include applying neural networks, pattern recognition, and natural language processing to massive datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Variational inference is an alternative to the sampling-based approximation via MCMC that approximates a target density through optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Statisticians and computer sci- entists (starting with Peterson & Anderson (1987), Jordan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (1999), Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2017)) have been using variational techniques in a variety of settings because these techniques tend to be faster and easier to scale to massive datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Despite its popularity among statistics and data science practitioners, variational inference is rarely discussed in undergraduate 2 courses as it is believed to be a rather advanced topic (Dogucu & Hu 2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' With this in mind, we have developed a one-week course module to disrupt this notion of being an ad- vanced topic and help instructors to introduce variational techniques in their undergraduate courses for more realistic use cases of probabilistic models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Our proposed one-week module is based on the best practices of active learning, which have been shown to improve student learning and engagement (Michael 2006, Freeman et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2014), Deslauriers et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2019)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Our main guiding principle in designing the module is to involve students in the learning process by introducing student-centered class activities and labs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The guiding principle also includes assigning open-ended questions, focusing on problem-solving, providing ap- propriate scaffolding for activities, and creating opportunities to work collaboratively with peers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Our module is designed for students to gain a fundamental understanding and practical experience with variational inference over the course of two class meetings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' During the first meeting, students are exposed to the fundamentals of variational inferences including the Kullback-Leibler divergence, evidence lower bound, gradient ascent, and coordinate ascent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Additionally, they gain their first hands-on experience by applying variational inference to a simple probabilistic model for count data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To encourage and empower undergraduate instructors to adopt and adapt this variational inference module, we provide an accompa- nying in-class handout and an R Shiny app with details explained in the supplementary materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' During the second class meeting, students work on a guided R lab to apply variational inference to a realistic scenario of clustering documents with Latent Dirichlet Allocation (Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' See Table 1 for the breakdown of the module.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Content 1st class Lecture: Fundamentals of variational inference Class activity: Probabilistic model for count data with variational inference 2nd class Lab: Document clustering Table 1: Outline of the one-week variational inference module.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' As for the audience, we believe that the module can be seamlessly integrated into any intermediate- or upper-level undergraduate course in data science, Bayesian statistics, mul- tivariate data analysis, and statistical machine learning that covers topics on clustering, 3 classification, or text analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The prerequisites needed for the module are a basic under- standing of statistical modeling, probability distributions, and elementary calculus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The remainder of the paper is organized as the following.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In Section 2, we provide an overview of variational inference essentials that can be readily used as a basis for a lecture instruction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Section 3 presents a motivating example and the Gamma-Poisson model for count data that serves as the first hands-on class activity with variational inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In Section 4, we offer a realistic use case for variational inference based on a Latent Dirichlet Allocation model for document clustering with implementation details in R, which can be used as a computing lab.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We end the paper in Section 5 with a few concluding remarks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2 Lecture: Foundations of Variantional Inference In this section, we introduce concepts and definitions of variational inference in Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1, discuss the choices of variational families in Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2, and present details of ELBO optimization in Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We also include recommendations of variational families and ELBO optimization strategies with pedagogical considerations for an undergraduate audi- ence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Instructors can design their lecture based on these materials tailored to their needs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 Concepts and Definitions The main idea behind variational inference is to approximate the target probability density p(θ | y) by a member of some relatively simple family of densities q(θ | λ), indexed by the variational parameter λ ∈ Rm′ (m′ ≥ 1), over the space of model parameters θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that λ = (λ1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , λm) has m components of (potentially) varying dimensions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Variational approximation is done by finding the member of variational family that minimizes the Kullback-Leibler (KL) divergence of q(θ | λ) from p(θ | y): q∗ = arg min q(θ|λ) KL(q(θ | λ)||p(θ | y)), (2) with KL divergence being the expectation of the log ratio between the q(θ | λ) and p(θ | y) with respect to q(θ | λ): KL(q(θ | λ)||p(θ | y)) = Eq � log q(θ | λ) p(θ | y) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (3) 4 Figure 1: Illustration of variational inference as the optimization-based approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The goal of variational inference is to find a member of the variational family that minimizes KL divergence with the target distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The KL divergence measures how different is the probability distribution q(θ | λ) from p(θ | y) (Kullback & Leibler 1951).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that while we use the KL divergence to measure the similarity between two densities, it is not a metric because the KL divergence is not symmetric and does not satisfy the triangle inequality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In fact, the order of q(θ | λ) and p(θ | y) in Equation (2) is deliberate as it leads to taking the expectation with respect to the variational distribution q(θ | λ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' One can naturally think of reversing the roles of q(θ | λ) and p(θ | y).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' However, this leads to a “different kind” of variational inference called expectation propagation (Minka (2001)), which loses computational efficiency of variational inference defined in Equation (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In a nutshell, rather than sampling, variational inference approximates densities us- ing optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' See Figure 1 for a graphical illustration, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', by finding the values of variational parameters from λinit to λ∗ through optimization which lead to a variational distribution q(θ | λ) that is close to the target posterior distribution p(θ | y) defined by the smallest KL divergence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Finding the optimal q∗ is done in practice by maximizing an equivalent objective function, L(λ), the evidence lower bound (ELBO), because the KL 5 q(0|入) Optimization init Smallest KL p(0ly)divergence is intractable as it requires the evaluation of the marginal distribution p(y): L(λ) = Eq[log p(y, θ) − log q(θ|λ)] = Eq[log p(y|θ)] � �� � Expected log-likelihood of data − KL(q(θ|λ)||p(θ)) � �� � KL div.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' between the variational and prior densities .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (4) The ELBO is the sum between the negative KL divergence of the variational density from the target density and the log of the marginal density p(y).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Since the term log p(y) is constant with respect to q(θ | λ), the objective functions in Equation (3) and Equation (4) are equivalent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Examining the ELBO also reveals the intuition behind variational inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the one hand, the first term in Equation (4) encourages the variational approximation to place mass on parameter values that maximize the sampling density p(y | θ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the other hand, the second term in Equation (4) prefers closeness of the variational density to the prior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Therefore, the ELBO shows a similar tension between the sampling density and the prior known in Bayesian inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 Variational Families with Pedagogical Recommendations We now move on to the implementation details of variational inference starting with the selection of the variational family q(θ | λ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This choice is crucial as it affects the complexity of optimization outlined in Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 as well as the quality of variational approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Mean-field Variational Family By far the most popular is the mean-field variational family which assumes that all the unknown parameters are mutually independent, each approximated by its own univariate variational density: q(θ | λ) = m � i=1 q(θi | λi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (5) For example, a typical choice for real-valued parameters is the normal variational family q(θ | µ, σ2) and the log-normal or Gamma for non-negative parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The main advan- tage of the mean-field family is in its simplicity as it requires only a minimum number of parameters to be estimated (no correlation parameters) and often leads to uncompli- cated optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' However, the mutually independent parameter assumption comes at 6 a price because the mean-field family cannot capture relationships between model param- eters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To illustrate the pitfalls of mean-field approximation, consider a simple case of a two-dimensional normal target density with highly correlated components.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 2 shows the optimal mean-field variational approximation given by the product of two normal den- sities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' One can clearly see that the optimal variational densities match well with the means of the target density, but the marginal variances are underestimated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To further understand this common flaw of mean-field approximation, consider the definition of KL divergence in Equation (3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The objective function penalizes more larger density in q(θ | λ) in areas where p(θ | y) has low density than the opposite direction (recall that the expectation is taken with respect to the variational density).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Recommendation for Undergradudate Instruction It is worth noting that the development of new variational families which improves on the trade-off between complexity and expressiveness of variational approximations has been a fruitful and active area of research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To keep the scope of the undergraduate one-week variational inference module manageable to both the students and the instructors, we recommend solely focusing on the mean-field approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For interested students who want to explore further, we encourage the instructors to refer them to the recent work of Ambrogioni et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2021) that provides a detailed discussion on many state-of-the-art variational families and their associated implementation challenges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3 ELBO Optimization with Pedagogical Recommendations Besides the choice of variational family, another key implementation detail to address is the way in which we find the member of the variational family that maximizes the ELBO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Since this is a fairly general optimization problem, one can in principle use any optimization procedure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In the variational inference literature, the coordinate ascent and the gradient ascent procedures are the most prominent and widely used (Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2017)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 7 3 2 1 0 1 2 3 1 3 2 1 0 1 2 3 2 2 0 2 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='6 Density 2 0 2 2 Target density M-F approximation Figure 2: Mean-field variational approximation of a two-dimensional normal target density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The figure illustrates the common pitfall of the mean-field approximation in situations with correlated model parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 8 Corrdinate Ascent The coordinate ascent approach is based on the simple idea that one can maximize ELBO, which is a multivariate function, by cyclically maximizing it along one direction at a time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Starting with initial values (denoted by superscript 0) of the m variational parameters λ0 λ0 = (λ0 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , λ0 m), one obtains the (k + 1)th updated value of variational parameters by iteratively solving λk+1 i = arg max x L(λk+1 1 , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , λk+1 i−1 , x, λk i+1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , λk m), which can be accomplished without using gradients (Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Gradient Ascent Variational inference via gradient ascent uses the standard iterative optimization algorithm based on the idea that the ELBO grows fastest in the direction of its gradient (Hoffman et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In particular, the update of variational parameters λ at the (k +1)th iteration is given by λk+1 ← λk + η × ∇λL(λk), where ∇λL(λ) is the ELBO gradient, and η is the step size which is also called the learning rate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The step size controls the rate at which one updates the variational parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For both coordinate and gradient ascent, we typically declare convergence of variational parameters once the change in ELBO falls below some small threshold (Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Recommendation for Undergradudate Instruction Our recommendation for an undergraduate variational inference module is to take the route of gradient ascent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This pedagogical choice is guided by our combined experience of teaching statistical modeling, Bayesian statistics, and data science at various undergradu- ate levels to students with diverse statistical backgrounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Our recommendation has also taken into account the pedagogical advantages and disadvantages of gradient ascent and coordinate ascent for undergraduates: Variational inference via coordinate ascent, while conceptually straightforward, requires non-trivial and model-specific derivations which can 9 easily obscure the overall goal of this one-week module to expand students’ exposure to the state-of-the-art approximate inference for probabilistic models;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' gradient-based varia- tional inference, in contrast, leads to a black-box optimization that does not require any model-specific derivations due to an extensive autodifferentiation capabilities of modern statistical software such as RStan (Stan Development Team 2022) and Python packages PyTorch (Paszke et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2019) and TensorFlow (Abadi et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2015), to name a few.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We believe that from an undergraduate-level pedagogical perspective, gradient descent reflects better the current data science pipeline and allows the instruction to be focused on conceptual understanding of variational inference rather than technical details.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Of course, using gradient-based optimization requires the students to be familiar with partial derivatives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Such a pre-requisite potentially restricts the audience for our module to a course with a multivariable calculus prerequisite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Nevertheless, we believe that an instructor with sufficient preparation can explain the basics behind gradient ascent to an audience with a minimal calculus background.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 3 Class Activity: A Probabilistic Model for Count Data with Variational Inference In this section, we provide a fully developed hands-on class activity with variational in- ference for count data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Starting with a motivating example in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1, we give an overview of the Gamma-Poisson model in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2, and discuss details of the variational inference of this model in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3, illustrated with an R Shiny app we have developed for instruction purpose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Instructors can adopt and adapt this class activity based on these materials tailored to their needs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 A Motivating Example To illustrate how ELBO optimization leads to a good approximation of target posterior distribution, we consider Poisson sampling with a Gamma prior, which is a popular one- parameter model for count data (Gelman et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2013, Albert & Hu (2019), Johnson et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2022)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To get started, we provide the following motivating example: 10 Our task is to estimate the average number of active users of a popular mas- sively multiplier online role-playing game (mmorpg) playing between the peak evening hours 7 pm and 10 pm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This information can help game developers in allocating server resources and optimizing user experience.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To estimate the average number of active users, we will consider the counts (in thousands) of active players collected during the peak evening hours over a two-week period in the past month.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We have chosen the Gamma-Poisson model as the probabilistic model in this class ac- tivity for two reasons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' First, the Gamma-Poisson model is relatively easy to understand for students with an elementary knowledge of probability distributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Second, the Gamma is a conjugate prior for Poisson sampling which means that one can derive the exact poste- rior distribution (another Gamma) and check the fidelity of variational approximation by comparing to the analytical Gamma solution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The learning objective of this class activity is to get students familiarized with various aspects of variational inference presented in Sec- tion 2, such as ELBO and variational family, with a simple example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Afterwards, students are better prepared to move on to more realistic scenarios, such as document clustering, described in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 Overview of the Gamma-Poisson Model We now provide an overview of the Gamma-Poisson model which can be readily turned into a class lecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Suppose that y = (y1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , yn) represent the observed counts in n time intervals where the counts are independent, and each yi follows a Poisson distribution with the same rate parameter θ > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The joint probability mass function of y = (y1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , yn) is p(y | θ) = n � i=1 p(yi | θ) ∝ θ �n i=1 yie−nθ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (6) The posterior distribution for the rate parameter θ is our inference target as θ represents the expected number of counts that occur during the given time intervals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that the Poisson sampling relies on several assumptions about the sampling process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' First, one assumes that the time interval is fixed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Second, the counts occurring during different time 11 intervals are independent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Lastly, the rate θ at which the counts occur is constant over time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The Gamma-Poisson conjugacy states that if θ follows a Gamma prior distribution with shape and rate parameters α and β, it can be shown that the posterior distribution p(θ | y) will also have a Gamma density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Namely, if θ ∼ Gamma(α, β), (7) then θ | y ∼ Gamma(α + n � i=1 yi, β + n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (8) In other words, given α, β, and y, one can derive the analytical solution to the posterior of p(θ | y) and can subsequently sample from Gamma(α + �n i=1 yi, β + n) to get posterior samples of θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' While no approximation is needed, it serves as a good example of illustrating how variational inference works in such a setting and allows evaluations of the performance of variational inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3 Variational Inference of the Gamma-Poisson Model Recall from Section 2 that variational inference approximates the (unknown) posterior distribution of a parameter by a simple family of distributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In this case, we will ap- proximate the posterior distribution p(θ | y) by a log-normal distribution with mean µ and standard deviation σ: q(θ | µ, σ) = 1 θσ √ 2πe− (ln θ−µ)2 2σ2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (9) The log-normal distribution is a continuous probability distribution of a random variable whose logarithm is normally distributed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' It is a popular variational family for non-negative parameters because it can be expressed as a (continuously) transformed normal distribu- tion, and therefore it is amenable to automatic differentiation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Automatic differentiation is a computation method for derivatives in computer programs that relies on the application of chain rule in differential calculus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' It provides accurate and fast numerical derivative evaluations that leads to machine learning algorithms (such as variational inference) that do not require users to manually work out and code derivatives (Kucukelbir et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2017, Baydin et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2018)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 12 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3 40 50 60 θ Density Prior True posterior log−normal(3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='7, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='05) ELBO maximization: log−normal(3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='9, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='04) Figure 3: Variational approximation based on the motivating scenario of mmorpg’s player activity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The true Gamma(792, 100) posterior and the prior Gamma(100,2) distributions are included.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In the supplementary materials, we provide an accompanying in-class handout and an R Shiny app based on the motivating scenario of mmorpg described in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The first two parts of the handout present the motivating example and the overview of the Gamma- Poisson model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In the third part of the handout, students carry out exact posterior inference for the unknown rate parameter θ using a small dataset of observed counts of mmorpg’s active players.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In the fourth and final part, students find variational approximation of p(θ | y) and check how well their approximation matches the true posterior distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 3 shows the final variational approximation compared to the true Gamma(792, 100) posterior distribution from the handout example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We can see, on the one hand, the resulting log-normal(3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='9, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='04) distribution (the black dash line) that maximizes the ELBO visually overlaps with the true posterior (ELBO = −42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='52, KL divergence < 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='001).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the other hand, another member of the variational family, the log-normal(3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='7, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='05) distribution (the blue dot-dash line;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' with ELBO = −57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='55 and KL divergence = 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='085), clearly differs from the target.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This example illustrates the good performance of variational inference through optimization for the Gamma-Poisson count model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 13 The design of this class activity is guided by the active-learning principles listed in Section 1 and the goal is to give students their first hands-on experience with variational inference without the need of coding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Specifically, we include open-ended questions that focus on problem-solving and create opportunities for students to collaborate with peers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Moreover, the accompanying R Shiny app provides appropriate and sufficient scaffolding so that students can concentrate on conceptual understanding instead of the technical details, which follows our pedagogical recommendations in Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We now turn to a guided R lab to illustrate the use of variational inference for a more realistic use case of document clustering, applied to a sample of Associated Press newspaper articles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 4 Lab: Document Clustering Among the many models approximated by variational inference techniques, Latent Dirichlet Allocation (LDA) might be one of the most popular (Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' LDA is a mixed- membership clustering model, commonly used for document clustering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Specifically, LDA models each document to have a mixture of topics, where each word in the document is drawn from a topic based on the mixing proportions (Stan Development Team 2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' While the LDA model is relatively easy and straightforward to follow, using conventional MCMC estimation techniques has proven to be too computationally demanding due to the large number of parameters involved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Therefore, researchers and practitioners turn to variational inference techniques when using LDA for document clustering (Blei et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1, we briefly introduce the LDA model following the presentation in Stan Development Team (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2, we present an LDA application to a collection of Associate Press newspaper articles where variational inference is implemented by the cmdstanr R package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For brevity, we focus on the interpretation of results and discuss pedagogical considerations and leave a Stan script for the LDA model and the details of the guided lab assignment with R code in the supplementary materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 14 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 Overview of the LDA model The LDA model considers K topics for M documents made up of words drawn from a vocabulary of V distinct words.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For a document m, a topic distribution θm over K topics is drawn from a Dirichlet distribution, θm ∼ Dirichlet(α), (10) where �K k=1 θm,k = 1 (0 ≤ θm,k ≤ 1) and α is a vector of length K with positive values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Each of the Nm words {wm,1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , wm,Nm} in document m is then generated indepen- dently conditional on θm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To do so, first, the topic zm,n for word wm,n in document m is drawn from zm,n ∼ categorical(θm), (11) where θm is the document-specific topic-distribution defined in Equation (14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Next, the word wm,n in document m is drawn from wm,n ∼ categorical(φz[m,n]), (12) which is the word distribution for topic zm,n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that z[m, n] in Equation (16) refers to zm,n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Lastly, a Dirichlet prior is given to distributions φk over words for topic k as φk ∼ Dirichlet(β), (13) where β is the prior a vector of length V (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', the total number of words) with positive values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 10 shows a graphical model representation of LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 Clustering of Associated Press Newspaper Articles As a realistic application of variational inference, we consider a collection of 2246 Asso- ciated Press newspaper articles to be clustered using the LDA model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The dataset is (conveniently) part of the topicmodels R package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We believe this dataset is well suited to demonstrate the capabilities of variational inference in the classroom as it is too large for the MCMC approximation to be feasible but small enough for the variational inference to take just a few minutes to converge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For brevity, we highlight the results based on a 15 Figure 4: Graphical model representation of LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The largest box represents the docu- ments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the left, the inner box represents the topics and words within each document.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the right, the box represents the topics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' two-topic LDA model (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', K = 2) and leave the details to the guided lab in the sup- plementary materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The number of topics is set to 2 for demonstration purposes and simplicity of interpretations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Comparing LDA with a different number of topics is often done with metrics such as semantic coherence or held-out data likelihood (Mimno et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' While such a comparison is beyond the scope of this lab, interested students are encouraged to explore mentored by the instructors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 5 shows the evolution of ELBO for the two-topic LDA model which converged after a little bit over 100 iterations of the gradient ascent algorithm described in Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On a standard laptop computer, this typically takes between 5-10 minutes depending on the CPU speed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We recommend running the algorithm repeatedly (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', 2-3 times) with a different random seed in the classroom and discussing the dependency of variational inference on initial values of variational parameters which can occur in practice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figures 6 and 7 are examples of graphical displays of the topics that were extracted from the collection of articles based with the LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In particular, Figure 6 shows the 10 most common words for each topic;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' that is, the parts of distribution φk, for k ∈ {1, 2}, with the largest mass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 7 displays similar information for the 20 most common words for each topic in the form of a word cloud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The most common words in topic 1 include people, government, president, police, and state, suggesting that this topic may represent political news.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In contrast, the most common words in topic 2 include percent, billion, 16 N α m β K M−1120000 −1100000 −1080000 −1060000 −1040000 0 30 60 90 Iteration ELBO Figure 5: The evolution of ELBO for the two-topic LDA model based on 2246 Associated Press newspaper articles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' million, market, American, and states, hinting that this topic may represent news about the US economy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 5 Concluding Remarks In this paper, we present a newly-developed one-week course module that exposes un- dergraduate students to approximation via variational inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The proposed module is self-contained in the sense that it encourages and empowers potential instructors to adopt and adapt the module as we provide an overview of variational inference, an active- learning-based class activity with an R Shiny app, and a guided lab based on a realis- tic application with R code (see the supplementary materials or https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='com/ kejzlarv/variational_inference_module).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Its design is rooted in the best practices of active learning that have been demonstrated to improve student learning and engagement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The module can be integrated into any intermediate- or upper-level undergraduate course where students learn probabilistic models (including logistic regression, Bayesian classifiers, neural networks, or models for natural language processing), such as Bayesian statistics, multivariate data analysis, and data science courses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The applications discussed 17 Topic 1 Topic 2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='000 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='005 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='010 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='015 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='020 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='025 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='000 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='005 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='010 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='015 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='020 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='025 american billion first states united last million year new percent told officials soviet state government police two president people i Word distributions ( ϕ ) Word Figure 6: Word distributions based on the two-topic LDA model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The 10 most common words are displayed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 7: World clouds consisting of the 20 most common words for each of the two topics extracted by the LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 18 market company south house first t united partygovernment billion daylastyear I federal told people rs york eal percentreport twO I police city @national bush new news president states court soviet military million S week officials say american time thursdayin these courses are typically limited to scenarios with relatively small datasets, since the required use of MCMC does not scale well to large datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Given the popularity and scalability of variational inference, we hope that undergraduate instructors adopting and adapting this module will be able to integrate more realistic and fun use cases in their classrooms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Moreover, the references and further readings provided in this paper are readily available resources for a deeper dive of variational inference by interested students with appropriate mentoring by their undergraduate instructors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' References Abadi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Agarwal, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Barham, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Brevdo, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Chen, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Citro, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Corrado, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Davis, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Dean, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Devin, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Ghemawat, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Goodfellow, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Harp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Irving, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Isard, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Jia, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Jozefowicz, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Kaiser, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Kudlur, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Levenberg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Man´e, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Monga, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Moore, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Murray, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Olah, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Schuster, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Shlens, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Steiner, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Sutskever, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Talwar, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Tucker, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Vanhoucke, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Vasudevan, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Vi´egas, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Vinyals, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Warden, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Wattenberg, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Wicke, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Yu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Zheng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2015), ‘TensorFlow: Large-scale machine learning on heterogeneous systems’.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Software available from tensorflow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='tensorflow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/ Albert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Hu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2019), Probability and Bayesian Modeling, 1 edn, Chapman and Hall/CRC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Albert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Hu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2020), ‘Bayesian computing in the undergraduate statistics curriculum’, Journal of Statistics Education 28, 236–247.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Ambrogioni, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Lin, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Fertig, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Vikram, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Hinne, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Moore, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & van Gerven, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2021), Automatic structured variational inference, in A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Banerjee & K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Fukumizu, eds, ‘Proceedings of The 24th International Conference on Artificial Intelligence and Statistics’, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 130 of Proceedings of Machine Learning Research, PMLR, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 676–684.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://proceedings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='mlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='press/v130/ambrogioni21a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='html Bardenet, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Doucet, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Holmes, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2017), ‘On markov chain monte carlo methods for tall data’, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Mach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Learn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 18(1), 1515–1557.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 19 Baydin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Pearlmutter, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Radul, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Siskind, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2018), ‘Automatic differentiation in machine learning: a survey’, Journal of Machine Learning Research 18(153), 1–43.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: http://jmlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/papers/v18/17-468.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='html Blei, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Kucukelbir, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & McAuliffe, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2017), ‘Variational inference: A review for statisticians’, Journal of the American Statistical Association 112(518), 859–877.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1080/01621459.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1285773 Blei, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Ng, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Jordan, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2003), ‘Latent dirichlet allocation’, Journal of Machine Learning Research 3, 993–1022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Deslauriers, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', McCarty, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Miller, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Callaghan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Kestin, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2019), ‘Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom’, Proceedings of the National Academy of Sciences 116(39), 19251–19257.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/doi/abs/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1821936116 Dogucu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Hu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2022), ‘The current state of undergraduate bayesian education and recommendations for the future’, The American Statistician 76(4), 405–413.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1080/00031305.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2089232 Freeman, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Eddy, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', McDonough, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Smith, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Okoroafor, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Jordt, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Wen- deroth, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2014), ‘Active learning increases student performance in science, engineer- ing, and mathematics’, Proceedings of the National Academy of Sciences 111(23), 8410– 8415.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/doi/abs/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1319030111 Gelman, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Carlin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Stern, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Dunson, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Vehtari, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Rubin, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2013), Bayesian Data Analysis, third edn, CRC Pres.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://books.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='google.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='com/books?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='id=ZXL6AQAAQBAJ Hoffman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Blei, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Wang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Paisley, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2013), ‘Stochastic variational infer- ence’, Journal of Machine Learning Research 14, 1303–1347.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: http://jmlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/papers/v14/hoffman13a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='html 20 Johnson, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Ott, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Dogucu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2022), Bayes Rules!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' An Introduction to Applied Bayesian Modeling, 1 edn, Chapman and Hall/CRC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Jordan, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Ghahramani, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Jaakkola, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Saul, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (1999), ‘An introduction to variational methods for graphical models’, Machine Learning 37, 183–233.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Kucukelbir, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Tran, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Ranganath, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Gelman, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Blei, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2017), ‘Automatic differentiation variational inference’, Journal of Machine Learning Research 18(14), 1– 45.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: http://jmlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/papers/v18/16-107.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='html Kullback, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Leibler, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (1951), ‘On Information and Sufficiency’, The Annals of Mathematical Statistics 22(1), 79 – 86.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1214/aoms/1177729694 Michael, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2006), ‘Where’s the evidence that active learning works?’' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Advances in Physi- ology Education 30(4), 159–167.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' PMID: 17108243.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1152/advan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='00053.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2006 Mimno, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Wallach, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Talley, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Leenders, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & McCallum, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2011), Optimizing semantic coherence in topic models, in ‘Proceedings of the Conference on Empirical Methods in Natural Language Processing’, EMNLP ’11, Association for Computational Linguistics, USA, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 262–272.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Minka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2001), Expectation propagation for approximate bayesian inference, in ‘Pro- ceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence’, UAI’01, Morgan Kaufmann Publishers Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', San Francisco, CA, USA, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 362–369.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Paszke, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Gross, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Massa, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Lerer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Bradbury, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Chanan, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Killeen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Lin, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Gimelshein, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Antiga, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Desmaison, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Kopf, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Yang, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', DeVito, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Raison, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Tejani, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Chilamkurthy, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Steiner, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Fang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Bai, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Chintala, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2019), Pytorch: An imperative style, high-performance deep learning library, in ‘Advances in Neural Information Processing Systems 32’, Curran Associates, Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 8024–8035.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: http://papers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='neurips.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='cc/paper/9015-pytorch-an-imperative-style-high- performance-deep-learning-library.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='pdf 21 Peterson, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Anderson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (1987), ‘A mean field theory learning algorithm for neural networks’, Complex Systems 1, 995–1019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Rudoy, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Wolfe, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2006), Monte carlo methods for multi-modal distributions, in ‘2006 Fortieth Asilomar Conference on Signals, Systems and Computers’, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2019–2023.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Schwab-McCoy, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', Baker, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' & Gasper, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' (2021), ‘Data science in 2020: Com- puting, curricula, and challenges for the next 10 years’, Journal of Statistics and Data Science Education 29(sup1), S40–S50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1080/10691898.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1851159 Stan Development Team (2022), Stan Modeling Language User’s Guide and Reference Man- ual, Version 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' URL: http://mc-stan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='org/ 22 Supplementary Materials for Introducing Variational Inference In Undergraduate Statistics and Data Science Curriculum The supplementary materials include: 1) Details of the class activity on probabilistic model for count data with variational inference, introduced in Section 3 in the main text;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2) The manual of the R shiny app we have developed for the module, mentioned in Section 3 in the main text;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' and 3) Details of the guided R lab of the LDA application to a sample of the Associated Press newspaper articles with variational inference, presented in Section 4 in the main text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 6 Class Activity: Probabilistic Model for Count Data with Variational Inference The goal of this activity is to illustrate variational inference on a simple example of Gamma- Poisson conjugate model, which is a popular model for count data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 A Motivating Example Our task is to estimate the average number of active users of a popular massively multiplier online role-playing game (mmorpg) playing between the peak evening hours 7 pm and 10 pm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This information can help the game developers in allocating server resources and optimizing user experience.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To make this estimate, we will consider the following counts (in thousands) of active players collected during the peak evening hours over a two-week period past month.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Sun Mon Tue Wed Thu Fri Sat Week 1 50 47 46 52 49 55 53 Week 2 48 45 51 50 53 46 47 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 Overview of the Gamma-Poisson Model Sampling density: 23 Suppose that y = (y1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , yn) represent the observed counts in n time intervals where the counts are independent, then each yi follows a Poisson distribution with rate θ > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Namely, yi | θ ∼ Poisson(θ) E(yi | θ) = θ Var(yi | θ) = θ Prior distribution: θ ∼ Gamma(α, β) α > 0 is the shape parameter β > 0 is the rate parameter E(θ) = α β Var(θ) = α β2 Posterior distribution: θ | y1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , yn ∼ Gamma(α + n � i=1 yi, β + n) 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3 Exact Inference with the Gamma-Poisson Model We will start by selecting a prior distribution for the unknown average number of active users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Suppose that we elicit an expert’s advice on the matter, and they tell us that a similar mmorpg has typically about 50,000 users during peak hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' However, they are not too sure about that, so the interval between 45,000 and 55,000 users should have a reasonably high probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Suppose that we elicit an expert’s advice on the matter, and they tell us that a similar mmorpg has typically about 50,000 users during peak hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' However, they are not too sure about that, so the interval between 45,000 and 55,000 users should have a reasonably high probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This reasoning leads to a Gamma(100, 2) as a reasonable prior for the average number of active users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 24 Task 1: Explain the reasoning behind using Gamma(100, 2) as the prior distri- bution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 2: Use the information above to find the exact posterior distribution for the average number of active users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 3: What are the mean and standard deviation of the posterior distribu- tion that you just obtained?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' What is your recommendation about the typical number of active users playing the mmorpg between the peak evening hours 7pm and 10pm?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='4 Variational Inference with the Gamma-Poisson Model Variational inference approximates the (unknown) posterior distribution of a parameter by a simple family of distributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' In this case, we will try to approximate the posterior distribution of the mmorpg’s average number of active users between the peak hours θ by a log-normal distribution with mean µ and standard deviation σ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Log-normal distribution is a continuous probability distribution of a random variable whose logarithm is normally distributed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' It also happens to be a popular variational family for non-negative parameters as it is amenable to autodifferentiation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Since we know exactly how the posterior distri- bution for Gamma-Poisson model looks like, we will be able to check the fidelity of the variational approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Use the accompanying applet titled Variational Inference with Gamma-Poisson Model for count data to complete the following task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 25 Task 4: Use the sliders in the applet to manually find the member of a log- normal variational family that well approximates the posterior distribution of θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' What is your strategy?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 5: Compare your approximation with a neighbor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Whose approximation is closer to the exact posterior distribution of θ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' How are you deciding?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 6: Check the Fit a variational approximation box in the applet to find the variational approximation using the gradient ascent algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' How close was the variational approximation that you found manually to the one found here?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 7 Manual of the R shiny app This document describes the elements of R Shiny applet that accompanies the “Proba- bilistic Model for Count Data with Variational Inference” class activity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that the numbering in Section 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 and Section 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 corresponds to the numbered boxes in Figure 8 and Figure 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 Manual search for variational approximation 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Sliders to control the mean µ and the standard deviation σ of log-normal variational family.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The ELBO and KL divergence values for variational approximation based on the mean and standard deviations selected in box 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A plot that displays the true Gamma(792, 100) posterior distribution, the Gamma(100, 2) prior distribution, and the variational approximation based on the selection in box 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A checkbox to display the results of ELBO maximization via gradient ascent algo- rithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The resulting variational approximation is plotted in box 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 26 Figure 8: The applet is based on the class activity presented in Section 1 of the supplemen- tary materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The applet visual before checking the “Fit a variational approximation“ checkbox is displayed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 Variational approximation based on ELBO maximization 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The resulting mean µ, standard deviation σ, and ELBO values of variational approx- imation based on ELBO maximization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' A plot depicting ELBO values for each iteration of the gradient ascent algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 8 Lab: Document Clustering The goal of this lab is to gain a practical experience with variational inference on a real- istic use case based on Latent Dirichlet Allocation (LDA) and implement the model in R applied to a dataset of documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To do so, we consider a collection of 2246 Associated Press newspaper articles to be clustered using the LDA model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The dataset is part of the topicmodels R package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' You can load the dataset AssociatedPress with the following R command.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' data("AssociatedPress", package = "topicmodels") 27 Variational Inference with Gamma-Poisson Model for count data Variational approximation using log-normal variational family: Prior True posterior-.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' VI- Manual 3 μ 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3- 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='7 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='82 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='9 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='14 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 6 PDF 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='01 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 ELBO value: 2 57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='496 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='0 40 50 KL Divergence value: 60 0 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='031 OFit a variational approximation 4Figure 9: The applet visual after checking the “Fit a variational approximation“ checkbox is displayed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 28 Variational Inference with Gamma-Poisson Model for count data Variational approximation using log-normal variational family: :Prior True posterior -: VI-Manual - VI -ELBO maximization μ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='3- 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='7 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='58 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='66 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='74 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='82 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='9 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='984.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='064.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='14 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 PDF 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='01 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='030.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='040.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='050.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='060.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='070.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='080.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='090.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 ELBO value: 57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='496 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='0- 40 50 60 KL Divergence value: 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='031 Fit a variational approximation 0 D Results of ELBo maximization via gradient ascent: 5 μ 50 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='901 0 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='BO 100 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='035 E ELBO value: 150 42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='52 200 0 50 100 150 200 Iteration8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1 Overview of the LDA model and Stan script The LDA is a mixed-membership clustering model, commonly used for document clustering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' LDA models each document to have a mixture of topics, where each word in the document is drawn from a topic based on the mixing proportions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Specifically, the LDA model assumes K topics for M documents made up of words drawn from V distinct words.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For document m, a topic distribution θm is drawn over K topics from a Dirichlet distribution, θm ∼ Dirichlet(α), (14) where �K k=1 θm,k = 1 (0 ≤ θm,k ≤ 1) and α is the prior a vector of length K with positive values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Each of the Nm words {wm,1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' , wm,Nm} in document m is then generated indepen- dently conditional on θm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' To do so, first, the topic zm,n for word wm,n in document m is drawn from zm,n ∼ categorical(θm), (15) where θm is the document-specific topic-distribution defined in Equation (14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Next, the word wm,n in document m is drawn from wm,n ∼ categorical(φz[m,n]), (16) which is the word distribution for topic zm,n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that z[m, n] in Equation (16) refers to zm,n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Lastly, a Dirichlet prior is given to distributions φk over words for topic k as φk ∼ Dirichlet(β), (17) where β is the prior a vector of length V (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', the total number of words) with positive values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Figure 10 shows a graphical model representation of LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Below, we include the Stan script for the LDA model provided by Stan Development Team (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Note that Stan supports the calculation of marginal distributions over the continuous parameters by summing out the discrete parameters in mixture models (Stan Development Team 2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' This process corresponds to the gamma parameter in the Stan script below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We refer interested readers to Stan Development Team (2022) for further details.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 29 Figure 10: Graphical model representation of LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The outer box represents the docu- ments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the left, the inner box represents the topics and words within each document.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' On the right, the box represents the topics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' data { int K;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // number of topics int V;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // number of words int M;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // number of docs int N;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // total word instances int w[N];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // word n int doc[N];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // doc ID for word n vector[K] alpha;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // topic prior vector of length K vector[V] beta;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // word prior vector of length V } parameters { simplex[K] theta[M];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // topic distribution for doc m simplex[V] phi[K];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // word distribution for topic k } model { for (m in 1:M) theta[m] ~ dirichlet(alpha);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 30 N α m β K Mfor (k in 1:K) phi[k] ~ dirichlet(beta);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' for (n in 1:N) { real gamma[K];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' for (k in 1:K) gamma[k] = log(theta[doc[n], k]) + log(phi[k, w[n]]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' target += log sum exp(gamma);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' // likelihood;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' } } 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='2 Variational inference with the LDA model For demonstration purposes, we shall start with a two-topic LDA model (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=', K = 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Before that, we recommend removing the words from AssociatedPress datasets that are rare using the function removeSparseTerms() from the tm package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' These words have a minimal effect on the LDA parameter estimation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Nevertheless, they increase the compu- tational cost of variational inference and therefore should be removed using the following R command.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' dtm <- removeSparseTerms(AssociatedPress, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='95) We are now ready to fit the LDA model using variational inference capabilities of the cmdstanr package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The following code achieves the goal: LDA model cmd <- cmdstan model(stan file = "LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='stan") N TOPICS <- 2 data <- list(K = N TOPICS,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' V = dim(dtm)[2],' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' M = dim(dtm)[1],' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' N = sum(dtm$v),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' w = rep(dtm$j,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='dtm$v),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' 31 doc = rep(dtm$i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='dtm$v),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' #according to Griffiths and Steyvers(2004) alpha = rep(50/N TOPICS,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='N TOPICS),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' beta = rep(1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='dim(dtm)[2]) ) vi fit <- LDA model cmd$variational(data = data,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' seed = 1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' output samples = 1000,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' eval elbo = 1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' grad samples = 10,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' elbo samples = 10,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' algorithm = "meanfield",' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' output dir = NULL,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' iter = 1000,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' adapt iter = 20,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' save latent dynamics=TRUE,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' tol rel obj = 10^-4) The “LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='stan” file contains the Stan script for the LDA model provided in Section 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We recommend the usage of the R help to get familiar with the variational() method of the cmdstan model() function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' The variable vi fit contains the results of variational approximation of the LDA parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' For example, one can obtain the word distributions for the each of the topics with vi fit$summary("phi").' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Finally, to access the ELBO values, use the following: vi diag <- utils::read.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='csv(vi fit$latent dynamics files()[1], comment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='char = "#") ELBO <- data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='frame(Iteration = vi diag[,1], ELBO = vi diag[,3]) 32 Task 1: Use a graphical display to show the 10 most common words for each of the two-topics and their probabilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 2: Use the function wordcloud() from the wordcloud package and display the most common words for each of the topics as a world clowd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' What kinds of articles do these topics represent?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 3: Fit a three-topic LDA model, display the most common words for each of the topics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' How do the results differ from the two-topic LDA?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' Task 4 (Advanced): Use the three-topic LDA model and diplay the topic preva- lence among the 2246 Associated Press articles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' That is, show what proportions of articles fall under each topic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' All necessary R code for fitting the LDA model to the Associated Press sample, including the graphical displays shown in the main text, is included in a separate R script file called LDA LAB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='R available as a part of the supplementary materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' We also include a printout of the R script below for interested readers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content=' library(cmdstanr) # Checking integrity of installation of cmdstanr check cmdstan toolchain() install cmdstan(cores = 2) cmdstan path() cmdstan version() # Auxiliary packages library(tm) library(tidyverse) library(tidytext) library(topicmodels) 33 ## Get data data("AssociatedPress", package = "topicmodels") ## Removing rare words from the vocabulary dtm <- removeSparseTerms(AssociatedPress, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='95) dim(dtm) ## Input for stan model N TOPICS <- 2 data <- list(K = N TOPICS, V = dim(dtm)[2], M = dim(dtm)[1], N = sum(dtm$v), w = rep(dtm$j,dtm$v), doc = rep(dtm$i,dtm$v), #according to Griffiths and Steyvers(2004) alpha = rep(50/N TOPICS,N TOPICS), beta = rep(1,dim(dtm)[2]) ) ### VB fit LDA model cmd <- cmdstan model(stan file = "LDA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='stan") LDA model cmd$print() vb fit <- LDA model cmd$variational(data = data, seed = 1, output samples = 1000, eval elbo = 1, grad samples = 10, 34 elbo samples = 10, algorithm = "meanfield", output dir = NULL, iter = 1000, adapt iter = 20, save latent dynamics=TRUE, tol rel obj = 10^-4) # Plotting ELBO vb diag <- utils::read.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='csv(vb fit$latent dynamics files()[1], comment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='char = "#") ELBO <- data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='frame(Iteration = vb diag[,1], ELBO = vb diag[,3]) ggplot(data = ELBO, aes(x = Iteration, y = ELBO)) + geom line(lwd=1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='5) + theme(text = element text(size = 20), panel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='background = element rect(fill = "transparent", color = "lightgrey"), panel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='grid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='major = element line(colour = "lightgrey")) + xlim(0,110) ## Accessing parameters vb fit$summary("theta") # dim: M-by-K vb fit$summary("phi") # dim: K-by-V ## Word distribution per topic V <- dim(dtm)[2] odd rows <- rep(c(1,0), times = V) Topic1 <- vb fit$summary("phi")[odd rows == 1,] Topic2 <- vb fit$summary("phi")[odd rows == 0,] 35 word probs <- data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='frame(Topic = c(rep("Topic 1", V), rep("Topic 2", V)), Word = rep(dtm$dimnames$Terms,N TOPICS), Probability = c(Topic1$mean, Topic2$mean)) # Selecting top 10 words per topic top words <- word probs %>% group by(Topic) %>% top n(10) %>% ungroup() %>% arrange(Topic, -Probability) top words %>% mutate(Word = reorder within(Word, Probability, Topic)) %>% ggplot(aes(Probability, Word, fill = factor(Topic))) + geom col(show.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='legend = FALSE) + facet wrap(~ Topic, scales = "free") + scale y reordered() + theme(text = element text(size = 15)) + xlim(0,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='025) + xlab("Word distributions ( \\u03d5 )") # Word Cloud display #install.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='packages("wordcloud") library(wordcloud) top words <- word probs %>% group by(Topic) %>% top n(20) %>% ungroup() %>% arrange(Topic, -Probability) mycolors <- brewer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='pal(8, "Dark2") wordcloud(top words %>% filter(Topic == "Topic 1") %>% .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='$Word , top words %>% filter(Topic == "Topic 1") %>% .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='$Probability, random.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='order = FALSE, color = mycolors) 36 mycolors <- brewer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='pal(8, "Dark2") wordcloud(top words %>% filter(Topic == "Topic 2") %>% .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='$Word , top words %>% filter(Topic == "Topic 2") %>% .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='$Probability, random.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} +page_content='order = FALSE, color = mycolors) 37' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/BtAzT4oBgHgl3EQfTfyw/content/2301.01251v1.pdf'} diff --git a/CdE3T4oBgHgl3EQfUQrt/content/tmp_files/2301.04450v1.pdf.txt b/CdE3T4oBgHgl3EQfUQrt/content/tmp_files/2301.04450v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..c8d4f416d4f74326e855819f76d1663f7968c4a3 --- /dev/null +++ b/CdE3T4oBgHgl3EQfUQrt/content/tmp_files/2301.04450v1.pdf.txt @@ -0,0 +1,720 @@ +Letter +Optica +1 +Subnanometer confinement and bundling of atoms in a +Rydberg empowered optical lattice +MOHAMMADSADEGH KHAZALI +Institute for Research in Fundamental Sciences (IPM), Tehran 19538-33511, Iran +Department of Physics, University of Tehran, North Kargar Ave., Tehran P.O. Box 14395-547, Iran +Compiled January 12, 2023 +Optical lattices are the basic blocks of atomic quan- +tum technology. +The scale and resolution of these +lattices are diffraction-limited to the light wavelength. +Tight confinement of single sites in conventional lat- +tices requires excessive laser intensity which in turn +suppresses the coherence due to enhanced scattering. +This article proposes a new scheme for atomic opti- +cal lattice with sub-wavelength spatial structure. The +potential is formed by the nonlinear optical response +of the three-level Rydberg dressed atoms, which is +not constrained by the diffraction limit of the driv- +ing fields. The lattice consists of a 3D array of ultra- +narrow Lorentzian wells with sub-nanometer widths. +The scheme allows moving adjacent sites to close dis- +tances with sub-nanometer resolution. These extreme +scales are now optically accessible by a hybrid scheme +deploying the dipolar interaction and optical twist of +atomic eigenstates. The interaction-induced two-body +resonance that forms the trapping potential, only oc- +curs at a peculiar laser intensity, localizing the trap +sites to ultra-narrow regions over the standing-wave +driving field. The Lorentzian trapping potentials with +2Å width and 30MHz depth are realizable with scatter- +ing rates as low as 1Hz. The mentioned improvements +allow quantum logic operations with Rydberg-Fermi +interaction. These techniques are particularly demand- +ing for the realization of atomtronics, quantum walks, +Hubbard models, and neutral-atom quantum simula- +tion. +© 2023 Optical Society of America +http://dx.doi.org/10.1364/ao.XX.XXXXXX +1. INTRODUCTION +The primary enabling technology in atomic quantum proces- +sors is the coherent control of the position and motion of atoms +by lasers. The underlying mechanism in conventional optical +lattices is the ac-Stark shift of atomic levels formed by far-off- +resonant laser fields. The diffraction limit, which is about the +wavelength of the light, is what determines the scale and spatial +resolution of such optical potential landscapes. This fundamen- +tally limits the optical manipulation of atoms, affecting some of +⟩ +|𝟓𝒔𝟐, 𝟏𝑺𝟎 +⟩ +|𝟓𝒔𝟓𝒑, 𝟑𝑷𝟏 +⟩ +|𝟓𝒔𝒏𝒔, 𝟑𝑺𝟏 +⟩ +|𝒈 +⟩ +|𝒑 +⟩ +|𝒆 +V +−∆ +𝛀𝟏 +𝛀𝟐(x) +(a) +𝜃 +(b) +0 +0.05 +0.1 +0.15 +0.2 +0.25 +0. +-1.2 +-1 +-0.8 +-0.6 +-0.4 +-0.2 +0 +0.2 +Fig. 1. Ultra-tight confinement of atoms in an interaction- +induced atomic lattice. (a) Rydberg dressing of ground state +Sr atoms with standing-wave laser field (blue line) results +in an interaction-induced periodic trapping potential that +features sharp resonance at a narrow range of laser intensity. +This would form ultra-narrow trapping wells (green line) that +results in sub-nanometer atomic confinement. (b) The level +scheme presents in-resonance two-photon Rydberg excitation. +the quantum technology applications. For instance, in the re- +cently proposed Rydberg-Fermi quantum simulator [2, 3], ultra- +tight confinement of atoms within the single lobe of the Rydberg +wave-function is required for high-fidelity scalable quantum +processing. Tight confinement is also demanding for applica- +tions that are based on distance selective interaction [4, 5] and +controlled Rydberg anti-blockade operations [6]. Finally, tight +confinement is demanded for improving the fidelity of neutral +atom processors [7–10]. +This article presents the first scheme for ultra-tight sub- +nanometer confinement of atoms in an optical lattice with dy- +namic features to move pair of lattice sites close to each other +at extreme scales of about 4Å which is the realm of solid-state +crystals. The dynamic control of the lattice separation allows a +new type of quantum gate operations powered by the remote +Rydberg-Fermi spin-flip. The Rydberg-Fermi spin-flip has been +observed in Bose Einstein Condensate (BEC) at inter-atomic +distances of about 30nm [11]. However, the real application +of this phenomenon in an atomic lattice quantum processor +was elusive. The real application requires dynamic maneuver +of the interatomic distance from micrometer scale during the +laser-addressing of individual sites [12, 13] to 30nm over the +interaction stage [11]. Furthermore, since the Rydberg-Fermi +interaction is proportional to the Rydberg wave-function proba- +bility amplitude, the interatomic distance must be fixed within +nanometer-scale precision. The tight confinement and ultra-high +precision of interatomic distances in the current lattice proposal +arXiv:2301.04450v1 [quant-ph] 11 Jan 2023 + +Letter +Optica +2 +opens new opportunities to develop nano-scale quantum tech- +nologies of this type. +Tight confinement of atoms in conventional optical lattices +requires an extensive power i.e. the spatial width is inversely +proportional to the quadruple root of the laser intensity. The +drawback is the loss of coherence due to the enhanced scattering. +In an alternative approach, this article deploys the nonlinear +response of Rydberg-dressed atoms to the intensity of standing- +wave driving field, as a means to form a lattice of ultra-narrow +trapping potentials. The sub-wavelength resolution arises when +the composition of eigenstates on two-atom basis twists rapidly +at a specific light intensity to form interaction-induced reso- +nance over a short length scale of the standing-wave. Unlike the +conventional ac-Stark shift potentials, this interaction-induced +potential is a quantum effect, with magnitude proportional to +¯h. This effect forms 3D lattices with potential widths as small as +the neutral atom radius. The proposed lattice features dynamic +terms that bundle pairs of atoms and draw them near to sub- +nanometer distances; the realm that used to be exclusive to solid +state crystals. +The recent advances in optical control of Rydberg atoms have +opened a wide range of applications in quantum technology [14– +21]. The required dipolar interaction in this proposal is formed +by the in-resonance dressing of ground-state atoms with the +highly excited Rydberg state [22, 23]. Rydberg dressing of a +BEC with homogeneous laser lights could form triangular and +quasi-ordered droplet crystals [23, 24]. However, this periodic +structure would not be fixed in space. The spatial pattern of the +driving field and intensity dependence of the potential would +spatially pin the lattice sites to the nodes of standing wave. +Therefore, the lattice structure would be fixed in the space. This +feature is required for addressing individual sites in atomic +processors. +The atomic lattice scheme is based on dressing 88Sr atoms +with the highly excited Rydberg level, see Fig. 1. In the two- +photon in-resonance dressing scheme [22, 23], the single atom +Hamiltonian is given by +Hi/¯h = Ω1 +2 (σi +gp + σi +pg) + Ω2(x) +2 +(σi +ep + σi +pe) − ∆σpp, +(1) +where σα,β = |α⟩⟨β| is the transition operator. The two Rabi +frequencies Ω1,2 are applied by 689nm and 318nm lasers that are +detuned from the intermediate state |p⟩ by ∆. With negligible +Rydberg decay rates, the system would follow the dark eigen- +state |d⟩ ∝ Ω2|g⟩ − Ω1|e⟩ with zero light-shift. In the limit of +Ω1 ≪ Ω2, ground state atoms will be partially dressed by Ryd- +berg states with the population of Pe = (Ω1/Ω2)2. The van-der +Waals interaction between Rydberg atoms Vij = ¯hC6/r6 +ijσieeσj +ee +is a function of interatomic distance rij. +The interaction of +|5sns 3S1⟩ Rydberg atoms is repulsive. This strong interaction +could exceed atom-light coupling over several micrometers of +interatomic separations. +The dynamic of the system under Rydberg interaction is gov- +erned by the master equation of two-body density matrices. The +two-body density matrices ρij = Tr¯i,¯jρ are obtained by tracing +over all but i and j particles. The corresponding master equation +would be given by +∂tρij = − i +¯h [Hi + Hj + Vij, ρij] + Li(ρij) + Lj(ρij) +(2) +The internal state dynamics are governed by single-particle dissi- +pation described by Li operator acting on ith atom. The Liouvil- +lian term Li(ρ) = ∑β D(cβ)ρi with D(c)ρi = cρic† − 1/2(c†cρi + +0 +0.5 +1 +1.5 +x/Rc +-1 +-0.5 +0 +U/|U0| +2/2 +(a) +-5 +-1 0 1 +5 +-1 +-0.5 +0 +Num +Ana +(b) +-5 +-1 0 1 +5 +-1 +-0.5 +0 +Num +Ana +(c) +Fig. 2. Interaction-induced atomic lattice. (a) The red line +shows the spatial profile of the Rabi frequency Ω2(x) = +Ω2c + Ω2sw| sin(kx sin(θ/2))|, where k = 2π/λ . The blue +line shows the interaction of two atoms as a function of in- +teratomic distance. When two atoms are within the soft-core +radius and are both located at the nodes of the standing wave +they experience a strong trapping potential. With a single +atom per lattice site, the effective trapping interaction would +be the sum of two-body interactions of all the sites within the +±Rc distance. (b) The interaction induced resonance occurs at +Ω2 = 2|∆|, with maximum depth of U0 = 3¯hΩ4 +1 +8∆γ2 and a HWHM +of Ω2 − 2|∆| = ±γp. The analytical form of Eq. 5 and numerical +calculation of the interaction potential (Eq. 3) presents a per- +fect match. (c) Spatial form of the interaction-induced trap at +the position of the ith standing wave node. The Lorentzian po- +tential of Eq. 6 with the width w (Eq. 7) shows a perfect match +with the numeric calculation of Eq. 3. Chosen parameters in (a) +are Ω2c = 2∆ = 2π × 10MHz, Ω2sw = ∆/2 loss limited to 1Hz, +n = 100, θ = π. +ρic†c) in the Lindblad form governs the dissipative time evolu- +tion. Lindblad terms encounter spontaneous emission from Ry- +dberg cpe = √γe|p⟩⟨e| and intermediate level cgp = √γp|g⟩⟨p|. +The spontaneous emission rates are γp/2π = 7.6kHz and γe can +be found in [25]. +Considering the steady state ¯ρij of Eq. 2, the effective interac- +tion would be given by +U(rij) = Tr[ ¯ρij(Hi + Hj + Vij)]. +(3) +For homogeneous lasers, a plateau-type interaction profile +would be formed with constant interaction within the soft-core +as depicted by the dotted line in Fig. 2a. In Rydberg-dressing +the interaction region is defined by Rc; the interatomic distance +within which interaction-induced laser detuning equals the ef- +fective laser bandwidth PrV(Rc) = Ω1Ω2/2∆ [23]. The soft- +core interaction features a sharp peak at Ω2 = 2|∆|, due to an +interaction-induced resonance, see Fig. 2b. +To form the optical lattice with the mentioned interaction- +induced resonance, a space-dependent variation of the upper +laser is deployed. Using different intensities for the counter- +propagating 318nm lasers, results in the desired spatial pattern + +Letter +Optica +3 +|𝛽"〉 +𝛽$ +|𝛽$〉 +|𝑔𝑔〉 +|𝜆"〉 +(b) +-∆ +−2∆ +2Ω1 +⟩ +|𝑒𝑒 +2Ω1 +2Ω1 +𝑉 +2Ω2 +2Ω2 +-∆ +⟩ +|𝑔𝑒$ +⟩ +|𝑔𝑝$ +⟩ +|𝑔𝑔 +⟩ +|𝑝𝑝 +⟩ +|𝑝𝑒$ +(a) +|𝜆$〉 +|𝜆0〉 +𝛽" +Ω2 +Fig. 3. The origin of the trapping potential is the interaction- +induced resonance at Ω2 = 2|∆|. (a) With Ω1 ≪ Ω2 the two- +atom Hilbert space would be organized in three subspaces +of ground state, single-excitation (green box) and double- +excitations (yellow box) that are coupled by weak Ω1 laser. +(b) The effects of strong coupling Ω2 and interaction V could +be observed by diagonalizing the green and yellow subspaces +with eigen-states of |β±⟩ and |λ0,±⟩ respectively. At the nodes +of the standing wave Ω2(x) = −2∆, the interaction-induced +level shift, makes the |λ−⟩ in-resonance with the ground state, +significantly enhancing the interaction and forming the trap- +ping potential. +of the Rabi frequency +Ω2(x) = Ω2c + Ω2sw| sin(kx sin(θ/2))|, +(4) +where k = 2π/λ is the laser wave-vector and θ is the angle +between counter propagating lights, see Fig. 1a. +Adjusting +Ω2c = −2∆ in Eq. 4 forms periodic trapping potentials at the +nodes of standing wave upon the presence of at least two atoms +within the core radius Rc, see Fig. 2a,c. In a 1D lattice with single +atom ocupation per site, the effective potential experienced by +each site is the sum of two-body interactions of neighboring +lattice sites within the interacting range of ±Rc. Considering the +isotropic Rydberg interaction of the S orbital, extension to the +3D lattice is trivial. +Here we analytically formulate the interaction-induced reso- +nance peak around Ω2(x) = 2|∆|. Considering the level scheme +of Fig. 3a in two-atom basis, the doubly excited Rydberg state +asymptotically decouples within the interaction region Rc as +V → ∞. Taking into account the remaining states, in the limit of +Ω1 ≪ Ω2c the steady state density could be obtained by adding +three orders of perturbative corrections to the initial ground +state. In the limit of γp ≪ ∆ the dressing interaction of the +steady state simplifies to +U(x) = +¯hΩ4 +1 +Ω2(x)2 +4∆[2∆2 + Ω2(x)2] +[4∆2 − Ω2(x)2]2 + 16γ2p∆2 . +(5) +The maximum interaction occurs at Ω2 = 2|∆| with the value +of U0 = 3¯hΩ4 +1 +8∆γ2 . Note that the attractive or repulsive nature of +the potential peak is determined by the sign of detuning ∆. +The half-width at half-maximum of interaction peak occurs at +Ω2(x) − |2∆| = ±γp. The presented analytic model of Eq. 5 +perfectly resemble the numerical results, see Fig. 2. Considering +the spatial variation of the Ω2 in Eq. 4 over the narrow area of +the potential peak k.(x − xi) ≪ 1 with Ω2c = −2∆, the spatial +profile of the ith trapping site has a Lorentzian form +Ui(x) = +U0 +1 + (x − xi)2/w2 +(6) +where the half-width at half-maximum and the depth of the +spatial trap well are given by +w = +γp +k sin(θ/2)Ω2sw +; +U0 = 3¯hΩ4 +1 +8∆γ2 . +(7) +Figure 2c compares this analytical form of Eq. 6 with the numer- +ical results. The scale of the trap width as a function of Ω2sw is +plotted in Fig. 4c. Remarkably, with Ω2sw/2π = 1.7MHz the +trap width would be as tight as the radius of 88Sr atoms. +0 +5 +10 +2c/2 (MHz) +0 +0.5 +1 +1.5 +2 +U0 (kHz) +(a) +0 +2 +4 +6 +8 +10 +2c/2 (MHz) +0 +0.01 +0.02 +0.03 +1/ +2c +(b) +102 +104 +2sw/2 (kHz) +100 +101 +trap width w (nm) +(c) +Fig. 4. (a) The scale of trap depth U0 for two atoms located +within the core distance of Rc is plotted as a function of Ω2c +for the constant scattering rate of 1Hz. Having N lattice sites +within the interaction distance Rc, the trapping potential ex- +perienced by an atom would add up to NU0. (b) The decoher- +ence rate is adjusted to 1Hz by controlling the ratio of Ω1/Ω2c. +(c) The width of Lorentzian traps w (Eq. 7) is plotted as a func- +tion of Ω2sw for θ = π. +The origin of the enhanced interaction at Ω2 = 2|∆| can be +traced to two-atom resonance that occurs in the presence of +strong interaction [22]. Considering the laser coupling on the +two-atom basis, for Ω1 ≪ Ω2 the two-atom Hilbert space would +be organized in three subspaces that are coupled by weak Ω1 +laser, see Fig. 3. These subspaces are the ground state |gg⟩, one +atom excitation {|gp⟩+, |ge⟩+}, and two atom excitation states +{|pp⟩, |pe⟩+, |ee⟩}, with |αβ⟩+ = (|α⟩ + |β⟩)/ +√ +2 represents sym- +metric two-particle states. The strong coupling Ω2, mixes the +states in each subspace. Pre-diagonalizing the subsystems quan- +tifies the light-shifts experienced by the eigen-states, see Fig. 3b. +For the second subspace with single excitation, the coupling +Hamiltonian in the {|gp⟩+, |ge⟩+} basis is given by +S2/¯h = +� +� −∆ +Ω2/2 +Ω2/2 +0 +� +� . +(8) +The eigen-energies in this subspace β±/¯h += +−∆/2 ± +1/2 +� +∆2 + Ω2 +2 does not get resonant with the ground state. In +the third subspace with double excitations, the coupling Hamil- + +Letter +Optica +4 +tonian in the {|pp⟩, |pe⟩+, |ee⟩} basis is given by +S3/¯h = +� +� +� +� +� +−2∆ +Ω2/ +√ +2 +0 +Ω2/ +√ +2 +−∆ +Ω2/ +√ +2 +0 +Ω2/ +√ +2 +V +� +� +� +� +� . +(9) +For large interaction inside the softcore V → ∞, the eigen- +energies are λ0 = V and λ± = − 3¯h +2 ∆ ± ¯h/2 +� +∆2 + 2Ω2 +2. The +doubly excited Rydberg state |λ0⟩ ≈ |ee⟩ decouples asymp- +totically. At Ω2 = 2|∆| one of |λ±⟩ eigen-states couples res- +onantly with the ground state, generating an enhanced light- +shift. As discussed above, small deviation of laser intensity +Ω2 − 2|∆| = ±γp makes the λ− eigen-state out of resonance. +Hence, the interaction-induced resonant peaks would be local- +ized at very narrow areas of the Ω2(x) standing-wave. +The main source of decoherence in this system is the spon- +taneous emission from the intermediate state. Rydberg inter- +action disturbs individual atom’s dark state, populating the +intermediate state |p⟩, and hence increases the loss rate per atom +Γ = Tr[ρi(γpσpp + γeσee)] at the trapping potential teeth. The +loss rate spatial profile is approximately given by γpU(x)/∆. +The maximum loss for a given Ω2(x) profile could be controlled +by adjusting the intensity of Ω1 laser. The scale of trap depth +for two atoms located within the soft-core is plotted in Fig. 4a +as a function of Ω2c for the constant scattering rate of 1Hz. The +interaction-to-loss ratio enhances by applying stronger laser +driving of Ω2c. Having N single-atom-occupied trapping sites +within the ±Rc interaction distance, the trapping potential expe- +rienced by an atom would add up to NU0. In a case study, con- +sidering the lattice constant of λ/2, dressing ground state atoms +to |5s100s 3S1⟩ with Ω2c/2π = 10MHz and limiting the loss rate +to 1Hz, the collective trapping potential experienced by a single +atom in 1D (3D) lattice interacting by neighboring cites within +the soft-core would be N1DU0 = 77kHz, (N3DU0=37MHz). +An important feature of the proposed Rydberg-empowered +optical lattice is the possibility to move the pair lattice sites close +to each other by manipulating the laser intensity. The inter- +atomic distance could approach the extreme scales that used +to be limited to solid-state crystals. As mentioned above, the +resonance trapping potential occurs at the points of standing +waves that fulfill the Ω2c + Ω2sw| sin(kx sin(θ/2))| = 2|∆|. Ac- +cordingly, by adjusting the Ω2c and Ω2sw the resonance could +occur at positions other than the nodes of the standing-wave +driving-field, making a lattice of dimers as shown in Fig. 5b. +To find the precision in adjusting the minimum interatomic +distance, consider the case that the pair sites are very close +to the position of the nodes |x − xn| ≪ (k sin(θ/2))−1 where +xn is the position of a random node. At this regime, the sep- +aration of two lattice sites from a node would be given by +x − xn = ±(2|∆| − Ω2c)/(Ω2swk sin(θ/2)). Therefore, having +larger Ω2sw would enhance the adjustment precision of intra- +dimer lattice spacing. A sample laser parameters for trapping +two lattice sites at 4Å distance are Ω2sw/2π = 10MHz and +2|∆| − Ω2c = 2π × 25kHz and θ = π which are experimentally +realizable. +Outlook- The sub-nanoscale resolution in trapping and +bundling of pair sites demonstrated here extends the toolbox of +neutral atom quantum technology. Ultra-narrow wells in this +proposal allow significant suppression of the lattice constant +with a time-sharing approach [26]. In this approach, the applied +standing wave is stroboscopically shifted in space by λ/2N and +0.02 +0.04 +0.06 +0.08 +0.1 +0.12 +0.14 +0.16 +-1 +-0.8 +-0.6 +-0.4 +-0.2 +0 +0.2 +(a) +0.02 +0.04 +0.06 +0.08 +0.1 +0.12 +0.14 +0.16 +-1 +-0.5 +0 +0.5 +(b) +Fig. 5. Moving lattice sites by adjusting the relative intensity +of counter-propagating lasers. The lattice constant is originally +Λ/4 in (a). By changing the laser intensity, the resonance con- +dition Ω2(x) = 2|∆| fulfills at positions other than the nodes of +standing-wave, forming a lattice of dimers. The distance of the +atomic pairs could be made as small as 4Å. +hence the effective lattice constant would be smaller by a factor +of N−1. This compaction of the atomic lattice is quite demanding +for scaling the lattice sites with the current limited laser powers. +Furthermore, in quantum simulation with optical lattices, the en- +ergy scale of Hubbard models for both hopping and interaction +of atoms is set by the minimum lattice constant which used to be +limited to λ/2, leading to challenging temperature requirements +to observe quantum phases of interest [1]. +A distinct research avenue looks at the applications of the +presented scheme with ultra-narrow repulsive peaks. Equation +5 shows that changing the detuning sign would preserve the +interaction profile but only flip the potential sign from attrac- +tive to repulsive. These ultra-narrow barriers are ideal for the +realization of the Kronig-Penney (KP) lattice model [27]. Fur- +thermore, the three dimensional repulsive δ-function peaks form +nearly perfect box-traps [28]. These repulsive narrow peaks also +realize thin tunnel junctions for atomtronic devices [29, 30]. The +potential is easily generalizable to other geometries in three +dimensions using the holographically designed laser intensity +[31]. +REFERENCES +1. +C. Gross and I. Bloch, Science 357, 995 (2017). +2. +M Khazali, W Lechner, Electron cloud design for Rydberg multi-qubit +gates, arXiv:2111.01581 (2021). +3. +Khazali, Mohammadsadegh, Universal terminal for mobile edge- +quantum computing, arXiv:2204.08522 (2022). +4. +M. Khazali, ?Discrete-Time Quantum-Walk & Floquet Topological In- +sulators via Distance-Selective Rydberg-Interaction?, Quantum 6, 664 +(2022). +5. +Hollerith, S., Srakaew, K., Wei, D., Rubio-Abadal, A., Adler, D., +Weckesser, P., Kruckenhauser, A., Walther, V., van Bijnen, R., Rui, J. +and Gross, C., Realizing distance-selective interactions in a Rydberg- +dressed atom array. Phys. Rev. Lett., 128, 113602 (2022). +6. +C. Ates, T. Pohl, T. Pattard, and J. M. Rost, "Antiblockade in Rydberg +excitation of an ultracold lattice gas." Phys. Rev. Lett. 98, 023002 +(2007). +7. +Graham, T.M., Kwon, M., Grinkemeyer, B., Marra, Z., Jiang, X., Licht- +man, M.T., Sun, Y., Ebert, M. and Saffman, M., Rydberg-mediated +entanglement in a two-dimensional neutral atom qubit array. Phys. Rev. +lett., 123, 230501 (2019). +8. +A. Pagano, S. Weber, D. Jaschke, T. Pfau, F. Meinert, S. Montangero, +and H. P. Büchler, Error budgeting for a controlled-phase gate with +strontium-88 Rydberg atoms, Phys. Rev. Research 4, 033019 (2022). +9. +Cetina, M., Egan, L.N., Noel, C., Goldman, M.L., Biswas, D., Risinger, +A.R., Zhu, D. and Monroe, C., Control of transverse motion for quan- + +Letter +Optica +5 +tum gates on individually addressed atomic qubits. PRX Quantum, 3, +010334 (2022). +10. +Brennen, G.K., Caves, C.M., Jessen, P.S. and Deutsch, I.H., Quantum +logic gates in optical lattices. Phys. Rev. Lett. 82, 1060 (1999). +11. +Niederprüm, T., Thomas, O., Eichert, T., and Ott, H. Rydberg molecule- +induced remote spin flips. Phys. Rev. Lett. 117, 123002 (2016). +12. +C. Weitenberg, M. Endres, J. F. Sherson, M. Cheneau, P. Schauß, T. +Fukuhara, I. Bloch & S. Kuhr, "Single-spin addressing in an atomic +Mott insulator." Nature 471 319 (2011). +13. +M. Saffman, "Quantum computing with atomic qubits and Rydberg +interactions: progress and challenges." J. Phys. B 49, 202001 (2016). +14. +M. Saffman, T. G. Walker, and K. Mølmer. "Quantum information with +Rydberg atoms." Rev. Pod. Phys. 82, 2313 (2010). +15. +M. Khazali and K. Mølmer. Fast multiqubit gates by adiabatic evolution +in interacting excited-state manifolds of rydberg atoms and supercon- +ducting circuits. Phys. Rev. X, 10, 021054, (2020). +16. +M. Khazali, C. R Murray, and T. Pohl. Polariton exchange interactions +in multichannel optical networks. Physical Review Letters, 123 113605, +2019. +17. +M. Khazali, K. Heshami, and C. Simon. Photon-photon gate via the in- +teraction between two collective Rydberg excitations. Physical Review +A, 91 030301, (2015). +18. +M. Khazali, All optical quantum information processing via a single-step +Rydberg blockade gate, arXiv:2211.06998 (2022) +19. +M. Khazali, K. Heshami, and C. Simon. Single-photon source based on +Rydberg exciton blockade. J. Phys. B: At. Mol. Opt. Phys., 50, 215301, +(2017). +20. +M. Khazali, H. W. Lau, A. Humeniuk, and C. Simon. Large energy +superpositions via Rydberg dressing. Phys. Rev. A, 94, 023408, (2016); +M. Khazali. Progress towards macroscopic spin and mechanical su- +perposition via Rydberg interaction. Phys. Rev. A, 98, 043836, (2018). +21. +M. Khazali, "Quantum information and computation with Rydberg +atoms." Iranian Journal of Applied Physics 10 19 (2021); M. Khaz- +ali, Applications of Atomic Ensembles for Photonic Quantum Informa- +tion Processing and Fundamental Tests of Quantum Physics. Diss. +University of Calgary (Canada), (2016). +22. +C. Gaul, B. J. DeSalvo, J. A. Aman, F. B. Dunning, T. C. Killian, and T. +Pohl, Resonant Rydberg Dressing of Alkaline-Earth Atoms via Elec- +tromagnetically Induced Transparency, Phys. Rev. Lett. 116, 243001 +(2016). +23. +M. Khazali, Rydberg noisy dressing and applications in making soliton +molecules and droplet quasicrystals, Phys. Rev. Research 3, L032033 +(2021) +24. +Henkel, N., Cinti, F., Jain, P., Pupillo, G. and Pohl, T., Supersolid vor- +tex crystals in Rydberg-dressed Bose-Einstein condensates. Physical +Review Letters 108,265301 (2012). +25. +S. Kunze, R. Hohmann, H. J. Kluge, J. Lantzsch, L. Monz, J. Stenner, +K. Stratmann, K. Wendt, and K. Zimmer, Lifetime measurements of +highly excited Rydberg states of strontium I, Z. Phys. D 27, 111 (1993). +26. +S. Nascimbene, N. Goldman, N. R. Cooper, and J. Dalibard, Dynamic +optical lattices of sub-wavelength spacing for ultracold atoms, Phys. +Rev. Lett. 115, 140401 (2015). +27. +R. de L. Kronig and W. G. Penney, Proc. Roy. Soc. A 130, 499 (1931). +28. +A. L. Gaunt, T. F. Schmidutz, I. Gotlibovych, R. P. Smith, and Z. Hadz- +ibabic, Bose-Einstein condensation of atoms in a uniform potential, +Phys. Rev. Lett. 110, 200406 (2013). +29. +BT. Seaman, M. Krämer, DZ. Anderson, MJ. Holland, "Atomtronics: +Ultracold-atom analogs of electronic devices." Physical Review A 75.2 +(2007): 023615. +30. +S. Eckel, J. G. Lee, F. Jendrzejewski, N. Murray, C. W. Clark, C. J. +Lobb, W. D. Phillips, M. Edwards, and G. K. Campbell, Hysteresis in a +quantized superfluid atomtronic circuit, Nature 506, 200 (2014). +31. +Barredo, D., Lienhard, V., De Leseleuc, S., Lahaye, T. and Browaeys, +A., Synthetic three-dimensional atomic structures assembled atom by +atom. Nature 561, 79 (2018). + diff --git a/CdE3T4oBgHgl3EQfUQrt/content/tmp_files/load_file.txt b/CdE3T4oBgHgl3EQfUQrt/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..7f1df2e7ecfd931fc5abed4b9ce69d2cb04c7c52 --- /dev/null +++ b/CdE3T4oBgHgl3EQfUQrt/content/tmp_files/load_file.txt @@ -0,0 +1,545 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf,len=544 +page_content='Letter Optica 1 Subnanometer confinement and bundling of atoms in a Rydberg empowered optical lattice MOHAMMADSADEGH KHAZALI Institute for Research in Fundamental Sciences (IPM), Tehran 19538-33511, Iran Department of Physics, University of Tehran, North Kargar Ave.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Tehran P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Box 14395-547, Iran Compiled January 12, 2023 Optical lattices are the basic blocks of atomic quan- tum technology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The scale and resolution of these lattices are diffraction-limited to the light wavelength.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Tight confinement of single sites in conventional lat- tices requires excessive laser intensity which in turn suppresses the coherence due to enhanced scattering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This article proposes a new scheme for atomic opti- cal lattice with sub-wavelength spatial structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The potential is formed by the nonlinear optical response of the three-level Rydberg dressed atoms, which is not constrained by the diffraction limit of the driv- ing fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The lattice consists of a 3D array of ultra- narrow Lorentzian wells with sub-nanometer widths.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The scheme allows moving adjacent sites to close dis- tances with sub-nanometer resolution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' These extreme scales are now optically accessible by a hybrid scheme deploying the dipolar interaction and optical twist of atomic eigenstates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The interaction-induced two-body resonance that forms the trapping potential, only oc- curs at a peculiar laser intensity, localizing the trap sites to ultra-narrow regions over the standing-wave driving field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The Lorentzian trapping potentials with 2Å width and 30MHz depth are realizable with scatter- ing rates as low as 1Hz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The mentioned improvements allow quantum logic operations with Rydberg-Fermi interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' These techniques are particularly demand- ing for the realization of atomtronics, quantum walks, Hubbard models, and neutral-atom quantum simula- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' © 2023 Optical Society of America http://dx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='1364/ao.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='XX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='XXXXXX 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' INTRODUCTION The primary enabling technology in atomic quantum proces- sors is the coherent control of the position and motion of atoms by lasers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The underlying mechanism in conventional optical lattices is the ac-Stark shift of atomic levels formed by far-off- resonant laser fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The diffraction limit, which is about the wavelength of the light, is what determines the scale and spatial resolution of such optical potential landscapes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This fundamen- tally limits the optical manipulation of atoms, affecting some of ⟩ |𝟓𝒔𝟐, 𝟏𝑺𝟎 ⟩ |𝟓𝒔𝟓𝒑, 𝟑𝑷𝟏 ⟩ |𝟓𝒔𝒏𝒔, 𝟑𝑺𝟏 ⟩ |𝒈 ⟩ |𝒑 ⟩ |𝒆 V −∆ 𝛀𝟏 𝛀𝟐(x) (a) 𝜃 (b) 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Ultra-tight confinement of atoms in an interaction- induced atomic lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (a) Rydberg dressing of ground state Sr atoms with standing-wave laser field (blue line) results in an interaction-induced periodic trapping potential that features sharp resonance at a narrow range of laser intensity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This would form ultra-narrow trapping wells (green line) that results in sub-nanometer atomic confinement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (b) The level scheme presents in-resonance two-photon Rydberg excitation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' the quantum technology applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' For instance, in the re- cently proposed Rydberg-Fermi quantum simulator [2, 3], ultra- tight confinement of atoms within the single lobe of the Rydberg wave-function is required for high-fidelity scalable quantum processing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Tight confinement is also demanding for applica- tions that are based on distance selective interaction [4, 5] and controlled Rydberg anti-blockade operations [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Finally, tight confinement is demanded for improving the fidelity of neutral atom processors [7–10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This article presents the first scheme for ultra-tight sub- nanometer confinement of atoms in an optical lattice with dy- namic features to move pair of lattice sites close to each other at extreme scales of about 4Å which is the realm of solid-state crystals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The dynamic control of the lattice separation allows a new type of quantum gate operations powered by the remote Rydberg-Fermi spin-flip.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The Rydberg-Fermi spin-flip has been observed in Bose Einstein Condensate (BEC) at inter-atomic distances of about 30nm [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' However, the real application of this phenomenon in an atomic lattice quantum processor was elusive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The real application requires dynamic maneuver of the interatomic distance from micrometer scale during the laser-addressing of individual sites [12, 13] to 30nm over the interaction stage [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Furthermore, since the Rydberg-Fermi interaction is proportional to the Rydberg wave-function proba- bility amplitude, the interatomic distance must be fixed within nanometer-scale precision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The tight confinement and ultra-high precision of interatomic distances in the current lattice proposal arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='04450v1 [quant-ph] 11 Jan 2023 Letter Optica 2 opens new opportunities to develop nano-scale quantum tech- nologies of this type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Tight confinement of atoms in conventional optical lattices requires an extensive power i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' the spatial width is inversely proportional to the quadruple root of the laser intensity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The drawback is the loss of coherence due to the enhanced scattering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In an alternative approach, this article deploys the nonlinear response of Rydberg-dressed atoms to the intensity of standing- wave driving field, as a means to form a lattice of ultra-narrow trapping potentials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The sub-wavelength resolution arises when the composition of eigenstates on two-atom basis twists rapidly at a specific light intensity to form interaction-induced reso- nance over a short length scale of the standing-wave.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Unlike the conventional ac-Stark shift potentials, this interaction-induced potential is a quantum effect, with magnitude proportional to ¯h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This effect forms 3D lattices with potential widths as small as the neutral atom radius.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The proposed lattice features dynamic terms that bundle pairs of atoms and draw them near to sub- nanometer distances;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' the realm that used to be exclusive to solid state crystals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The recent advances in optical control of Rydberg atoms have opened a wide range of applications in quantum technology [14– 21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The required dipolar interaction in this proposal is formed by the in-resonance dressing of ground-state atoms with the highly excited Rydberg state [22, 23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rydberg dressing of a BEC with homogeneous laser lights could form triangular and quasi-ordered droplet crystals [23, 24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' However, this periodic structure would not be fixed in space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The spatial pattern of the driving field and intensity dependence of the potential would spatially pin the lattice sites to the nodes of standing wave.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Therefore, the lattice structure would be fixed in the space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This feature is required for addressing individual sites in atomic processors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The atomic lattice scheme is based on dressing 88Sr atoms with the highly excited Rydberg level, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In the two- photon in-resonance dressing scheme [22, 23], the single atom Hamiltonian is given by Hi/¯h = Ω1 2 (σi gp + σi pg) + Ω2(x) 2 (σi ep + σi pe) − ∆σpp, (1) where σα,β = |α⟩⟨β| is the transition operator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The two Rabi frequencies Ω1,2 are applied by 689nm and 318nm lasers that are detuned from the intermediate state |p⟩ by ∆.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' With negligible Rydberg decay rates, the system would follow the dark eigen- state |d⟩ ∝ Ω2|g⟩ − Ω1|e⟩ with zero light-shift.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In the limit of Ω1 ≪ Ω2, ground state atoms will be partially dressed by Ryd- berg states with the population of Pe = (Ω1/Ω2)2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The van-der Waals interaction between Rydberg atoms Vij = ¯hC6/r6 ijσieeσj ee is a function of interatomic distance rij.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The interaction of |5sns 3S1⟩ Rydberg atoms is repulsive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This strong interaction could exceed atom-light coupling over several micrometers of interatomic separations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The dynamic of the system under Rydberg interaction is gov- erned by the master equation of two-body density matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The two-body density matrices ρij = Tr¯i,¯jρ are obtained by tracing over all but i and j particles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The corresponding master equation would be given by ∂tρij = − i ¯h [Hi + Hj + Vij, ρij] + Li(ρij) + Lj(ρij) (2) The internal state dynamics are governed by single-particle dissi- pation described by Li operator acting on ith atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The Liouvil- lian term Li(ρ) = ∑β D(cβ)ρi with D(c)ρi = cρic† − 1/2(c†cρi + 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 x/Rc 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 0 U/|U0| 2/2 (a) 5 1 0 1 5 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 0 Num Ana (b) 5 1 0 1 5 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 0 Num Ana (c) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Interaction-induced atomic lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (a) The red line shows the spatial profile of the Rabi frequency Ω2(x) = Ω2c + Ω2sw| sin(kx sin(θ/2))|, where k = 2π/λ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The blue line shows the interaction of two atoms as a function of in- teratomic distance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' When two atoms are within the soft-core radius and are both located at the nodes of the standing wave they experience a strong trapping potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' With a single atom per lattice site, the effective trapping interaction would be the sum of two-body interactions of all the sites within the ±Rc distance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (b) The interaction induced resonance occurs at Ω2 = 2|∆|, with maximum depth of U0 = 3¯hΩ4 1 8∆γ2 and a HWHM of Ω2 − 2|∆| = ±γp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The analytical form of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 5 and numerical calculation of the interaction potential (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3) presents a per- fect match.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (c) Spatial form of the interaction-induced trap at the position of the ith standing wave node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The Lorentzian po- tential of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 6 with the width w (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 7) shows a perfect match with the numeric calculation of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Chosen parameters in (a) are Ω2c = 2∆ = 2π × 10MHz, Ω2sw = ∆/2 loss limited to 1Hz, n = 100, θ = π.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' ρic†c) in the Lindblad form governs the dissipative time evolu- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lindblad terms encounter spontaneous emission from Ry- dberg cpe = √γe|p⟩⟨e| and intermediate level cgp = √γp|g⟩⟨p|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The spontaneous emission rates are γp/2π = 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='6kHz and γe can be found in [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Considering the steady state ¯ρij of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2, the effective interac- tion would be given by U(rij) = Tr[ ¯ρij(Hi + Hj + Vij)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (3) For homogeneous lasers, a plateau-type interaction profile would be formed with constant interaction within the soft-core as depicted by the dotted line in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In Rydberg-dressing the interaction region is defined by Rc;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' the interatomic distance within which interaction-induced laser detuning equals the ef- fective laser bandwidth PrV(Rc) = Ω1Ω2/2∆ [23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The soft- core interaction features a sharp peak at Ω2 = 2|∆|, due to an interaction-induced resonance, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' To form the optical lattice with the mentioned interaction- induced resonance, a space-dependent variation of the upper laser is deployed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Using different intensities for the counter- propagating 318nm lasers, results in the desired spatial pattern Letter Optica 3 |𝛽"〉 𝛽$ |𝛽$〉 |𝑔𝑔〉 |𝜆"〉 (b) ∆ −2∆ 2Ω1 ⟩ |𝑒𝑒 2Ω1 2Ω1 𝑉 2Ω2 2Ω2 ∆ ⟩ |𝑔𝑒$ ⟩ |𝑔𝑝$ ⟩ |𝑔𝑔 ⟩ |𝑝𝑝 ⟩ |𝑝𝑒$ (a) |𝜆$〉 |𝜆0〉 𝛽" Ω2 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The origin of the trapping potential is the interaction- induced resonance at Ω2 = 2|∆|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (a) With Ω1 ≪ Ω2 the two- atom Hilbert space would be organized in three subspaces of ground state, single-excitation (green box) and double- excitations (yellow box) that are coupled by weak Ω1 laser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (b) The effects of strong coupling Ω2 and interaction V could be observed by diagonalizing the green and yellow subspaces with eigen-states of |β±⟩ and |λ0,±⟩ respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' At the nodes of the standing wave Ω2(x) = −2∆, the interaction-induced level shift, makes the |λ−⟩ in-resonance with the ground state, significantly enhancing the interaction and forming the trap- ping potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' of the Rabi frequency Ω2(x) = Ω2c + Ω2sw| sin(kx sin(θ/2))|, (4) where k = 2π/λ is the laser wave-vector and θ is the angle between counter propagating lights, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 1a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Adjusting Ω2c = −2∆ in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 4 forms periodic trapping potentials at the nodes of standing wave upon the presence of at least two atoms within the core radius Rc, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2a,c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In a 1D lattice with single atom ocupation per site, the effective potential experienced by each site is the sum of two-body interactions of neighboring lattice sites within the interacting range of ±Rc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Considering the isotropic Rydberg interaction of the S orbital, extension to the 3D lattice is trivial.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Here we analytically formulate the interaction-induced reso- nance peak around Ω2(x) = 2|∆|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Considering the level scheme of Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3a in two-atom basis, the doubly excited Rydberg state asymptotically decouples within the interaction region Rc as V → ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Taking into account the remaining states, in the limit of Ω1 ≪ Ω2c the steady state density could be obtained by adding three orders of perturbative corrections to the initial ground state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In the limit of γp ≪ ∆ the dressing interaction of the steady state simplifies to U(x) = ¯hΩ4 1 Ω2(x)2 4∆[2∆2 + Ω2(x)2] [4∆2 − Ω2(x)2]2 + 16γ2p∆2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (5) The maximum interaction occurs at Ω2 = 2|∆| with the value of U0 = 3¯hΩ4 1 8∆γ2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Note that the attractive or repulsive nature of the potential peak is determined by the sign of detuning ∆.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The half-width at half-maximum of interaction peak occurs at Ω2(x) − |2∆| = ±γp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The presented analytic model of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 5 perfectly resemble the numerical results, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Considering the spatial variation of the Ω2 in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 4 over the narrow area of the potential peak k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='(x − xi) ≪ 1 with Ω2c = −2∆, the spatial profile of the ith trapping site has a Lorentzian form Ui(x) = U0 1 + (x − xi)2/w2 (6) where the half-width at half-maximum and the depth of the spatial trap well are given by w = γp k sin(θ/2)Ω2sw ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' U0 = 3¯hΩ4 1 8∆γ2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (7) Figure 2c compares this analytical form of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 6 with the numer- ical results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The scale of the trap width as a function of Ω2sw is plotted in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 4c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Remarkably, with Ω2sw/2π = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='7MHz the trap width would be as tight as the radius of 88Sr atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 0 5 10 2c/2 (MHz) 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 2 U0 (kHz) (a) 0 2 4 6 8 10 2c/2 (MHz) 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='01 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='02 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='03 1/ 2c (b) 102 104 2sw/2 (kHz) 100 101 trap width w (nm) (c) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (a) The scale of trap depth U0 for two atoms located within the core distance of Rc is plotted as a function of Ω2c for the constant scattering rate of 1Hz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Having N lattice sites within the interaction distance Rc, the trapping potential ex- perienced by an atom would add up to NU0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (b) The decoher- ence rate is adjusted to 1Hz by controlling the ratio of Ω1/Ω2c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (c) The width of Lorentzian traps w (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 7) is plotted as a func- tion of Ω2sw for θ = π.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The origin of the enhanced interaction at Ω2 = 2|∆| can be traced to two-atom resonance that occurs in the presence of strong interaction [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Considering the laser coupling on the two-atom basis, for Ω1 ≪ Ω2 the two-atom Hilbert space would be organized in three subspaces that are coupled by weak Ω1 laser, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' These subspaces are the ground state |gg⟩, one atom excitation {|gp⟩+, |ge⟩+}, and two atom excitation states {|pp⟩, |pe⟩+, |ee⟩}, with |αβ⟩+ = (|α⟩ + |β⟩)/ √ 2 represents sym- metric two-particle states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The strong coupling Ω2, mixes the states in each subspace.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pre-diagonalizing the subsystems quan- tifies the light-shifts experienced by the eigen-states, see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' For the second subspace with single excitation, the coupling Hamiltonian in the {|gp⟩+, |ge⟩+} basis is given by S2/¯h = � � −∆ Ω2/2 Ω2/2 0 � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (8) The eigen-energies in this subspace β±/¯h = −∆/2 ± 1/2 � ∆2 + Ω2 2 does not get resonant with the ground state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In the third subspace with double excitations, the coupling Hamil- Letter Optica 4 tonian in the {|pp⟩, |pe⟩+, |ee⟩} basis is given by S3/¯h = � � � � � −2∆ Ω2/ √ 2 0 Ω2/ √ 2 −∆ Ω2/ √ 2 0 Ω2/ √ 2 V � � � � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' (9) For large interaction inside the softcore V → ∞, the eigen- energies are λ0 = V and λ± = − 3¯h 2 ∆ ± ¯h/2 � ∆2 + 2Ω2 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The doubly excited Rydberg state |λ0⟩ ≈ |ee⟩ decouples asymp- totically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' At Ω2 = 2|∆| one of |λ±⟩ eigen-states couples res- onantly with the ground state, generating an enhanced light- shift.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' As discussed above, small deviation of laser intensity Ω2 − 2|∆| = ±γp makes the λ− eigen-state out of resonance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Hence, the interaction-induced resonant peaks would be local- ized at very narrow areas of the Ω2(x) standing-wave.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The main source of decoherence in this system is the spon- taneous emission from the intermediate state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rydberg inter- action disturbs individual atom’s dark state, populating the intermediate state |p⟩, and hence increases the loss rate per atom Γ = Tr[ρi(γpσpp + γeσee)] at the trapping potential teeth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The loss rate spatial profile is approximately given by γpU(x)/∆.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The maximum loss for a given Ω2(x) profile could be controlled by adjusting the intensity of Ω1 laser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The scale of trap depth for two atoms located within the soft-core is plotted in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 4a as a function of Ω2c for the constant scattering rate of 1Hz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The interaction-to-loss ratio enhances by applying stronger laser driving of Ω2c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Having N single-atom-occupied trapping sites within the ±Rc interaction distance, the trapping potential expe- rienced by an atom would add up to NU0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In a case study, con- sidering the lattice constant of λ/2, dressing ground state atoms to |5s100s 3S1⟩ with Ω2c/2π = 10MHz and limiting the loss rate to 1Hz, the collective trapping potential experienced by a single atom in 1D (3D) lattice interacting by neighboring cites within the soft-core would be N1DU0 = 77kHz, (N3DU0=37MHz).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' An important feature of the proposed Rydberg-empowered optical lattice is the possibility to move the pair lattice sites close to each other by manipulating the laser intensity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The inter- atomic distance could approach the extreme scales that used to be limited to solid-state crystals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' As mentioned above, the resonance trapping potential occurs at the points of standing waves that fulfill the Ω2c + Ω2sw| sin(kx sin(θ/2))| = 2|∆|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Ac- cordingly, by adjusting the Ω2c and Ω2sw the resonance could occur at positions other than the nodes of the standing-wave driving-field, making a lattice of dimers as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 5b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' To find the precision in adjusting the minimum interatomic distance, consider the case that the pair sites are very close to the position of the nodes |x − xn| ≪ (k sin(θ/2))−1 where xn is the position of a random node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' At this regime, the sep- aration of two lattice sites from a node would be given by x − xn = ±(2|∆| − Ω2c)/(Ω2swk sin(θ/2)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Therefore, having larger Ω2sw would enhance the adjustment precision of intra- dimer lattice spacing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A sample laser parameters for trapping two lattice sites at 4Å distance are Ω2sw/2π = 10MHz and 2|∆| − Ω2c = 2π × 25kHz and θ = π which are experimentally realizable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Outlook- The sub-nanoscale resolution in trapping and bundling of pair sites demonstrated here extends the toolbox of neutral atom quantum technology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Ultra-narrow wells in this proposal allow significant suppression of the lattice constant with a time-sharing approach [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' In this approach, the applied standing wave is stroboscopically shifted in space by λ/2N and 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='02 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='04 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='06 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='08 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='12 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='14 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='16 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 (a) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='02 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='04 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='06 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='08 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='12 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='14 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='16 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='5 (b) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Moving lattice sites by adjusting the relative intensity of counter-propagating lasers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The lattice constant is originally Λ/4 in (a).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' By changing the laser intensity, the resonance con- dition Ω2(x) = 2|∆| fulfills at positions other than the nodes of standing-wave, forming a lattice of dimers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The distance of the atomic pairs could be made as small as 4Å.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' hence the effective lattice constant would be smaller by a factor of N−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' This compaction of the atomic lattice is quite demanding for scaling the lattice sites with the current limited laser powers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Furthermore, in quantum simulation with optical lattices, the en- ergy scale of Hubbard models for both hopping and interaction of atoms is set by the minimum lattice constant which used to be limited to λ/2, leading to challenging temperature requirements to observe quantum phases of interest [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A distinct research avenue looks at the applications of the presented scheme with ultra-narrow repulsive peaks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Equation 5 shows that changing the detuning sign would preserve the interaction profile but only flip the potential sign from attrac- tive to repulsive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' These ultra-narrow barriers are ideal for the realization of the Kronig-Penney (KP) lattice model [27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Fur- thermore, the three dimensional repulsive δ-function peaks form nearly perfect box-traps [28].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' These repulsive narrow peaks also realize thin tunnel junctions for atomtronic devices [29, 30].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' The potential is easily generalizable to other geometries in three dimensions using the holographically designed laser intensity [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' REFERENCES 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Gross and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Bloch, Science 357, 995 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M Khazali, W Lechner, Electron cloud design for Rydberg multi-qubit gates, arXiv:2111.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='01581 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, Mohammadsadegh, Universal terminal for mobile edge- quantum computing, arXiv:2204.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='08522 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='Discrete-Time Quantum-Walk & Floquet Topological In- sulators via Distance-Selective Rydberg-Interaction?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Quantum 6, 664 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Hollerith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Srakaew, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Wei, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Rubio-Abadal, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Adler, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Weckesser, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Kruckenhauser, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Walther, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', van Bijnen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Rui, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' and Gross, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Realizing distance-selective interactions in a Rydberg- dressed atom array.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', 128, 113602 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Ates, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pohl, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pattard, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rost, "Antiblockade in Rydberg excitation of an ultracold lattice gas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='" Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 98, 023002 (2007).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Graham, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Kwon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Grinkemeyer, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Marra, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Jiang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Licht- man, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Ebert, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' and Saffman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Rydberg-mediated entanglement in a two-dimensional neutral atom qubit array.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', 123, 230501 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pagano, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Weber, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Jaschke, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pfau, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Meinert, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Montangero, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Büchler, Error budgeting for a controlled-phase gate with strontium-88 Rydberg atoms, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Research 4, 033019 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Cetina, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Egan, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Noel, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Goldman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Biswas, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Risinger, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Zhu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' and Monroe, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Control of transverse motion for quan- Letter Optica 5 tum gates on individually addressed atomic qubits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' PRX Quantum, 3, 010334 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Brennen, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Caves, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Jessen, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' and Deutsch, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Quantum logic gates in optical lattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 82, 1060 (1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Niederprüm, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Thomas, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Eichert, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', and Ott, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rydberg molecule- induced remote spin flips.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 117, 123002 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Weitenberg, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Endres, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Sherson, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Cheneau, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Schauß, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Fukuhara, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Bloch & S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Kuhr, "Single-spin addressing in an atomic Mott insulator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='" Nature 471 319 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Saffman, "Quantum computing with atomic qubits and Rydberg interactions: progress and challenges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='" J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' B 49, 202001 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Saffman, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Walker, and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Mølmer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' "Quantum information with Rydberg atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='" Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 82, 2313 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Mølmer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Fast multiqubit gates by adiabatic evolution in interacting excited-state manifolds of rydberg atoms and supercon- ducting circuits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' X, 10, 021054, (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' R Murray, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pohl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Polariton exchange interactions in multichannel optical networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Physical Review Letters, 123 113605, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Heshami, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Simon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Photon-photon gate via the in- teraction between two collective Rydberg excitations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Physical Review A, 91 030301, (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, All optical quantum information processing via a single-step Rydberg blockade gate, arXiv:2211.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='06998 (2022) 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Heshami, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Simon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Single-photon source based on Rydberg exciton blockade.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' B: At.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Mol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Opt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', 50, 215301, (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lau, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Humeniuk, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Simon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Large energy superpositions via Rydberg dressing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A, 94, 023408, (2016);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Progress towards macroscopic spin and mechanical su- perposition via Rydberg interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A, 98, 043836, (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, "Quantum information and computation with Rydberg atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='" Iranian Journal of Applied Physics 10 19 (2021);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khaz- ali, Applications of Atomic Ensembles for Photonic Quantum Informa- tion Processing and Fundamental Tests of Quantum Physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Diss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' University of Calgary (Canada), (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Gaul, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' DeSalvo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Aman, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Dunning, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Killian, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Pohl, Resonant Rydberg Dressing of Alkaline-Earth Atoms via Elec- tromagnetically Induced Transparency, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 116, 243001 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Khazali, Rydberg noisy dressing and applications in making soliton molecules and droplet quasicrystals, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Research 3, L032033 (2021) 24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Henkel, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Cinti, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Jain, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Pupillo, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' and Pohl, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Supersolid vor- tex crystals in Rydberg-dressed Bose-Einstein condensates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Physical Review Letters 108,265301 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Kunze, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Hohmann, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Kluge, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lantzsch, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Monz, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Stenner, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Stratmann, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Wendt, and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Zimmer, Lifetime measurements of highly excited Rydberg states of strontium I, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' D 27, 111 (1993).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Nascimbene, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Goldman, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Cooper, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Dalibard, Dynamic optical lattices of sub-wavelength spacing for ultracold atoms, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 115, 140401 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' de L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Kronig and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Penney, Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Roy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A 130, 499 (1931).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Gaunt, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Schmidutz, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Gotlibovych, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Smith, and Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Hadz- ibabic, Bose-Einstein condensation of atoms in a uniform potential, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 110, 200406 (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' BT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Seaman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Krämer, DZ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Anderson, MJ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Holland, "Atomtronics: Ultracold-atom analogs of electronic devices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='" Physical Review A 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content='2 (2007): 023615.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Eckel, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lee, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Jendrzejewski, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Murray, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Clark, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Lobb, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Phillips, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Edwards, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Campbell, Hysteresis in a quantized superfluid atomtronic circuit, Nature 506, 200 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Barredo, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Lienhard, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', De Leseleuc, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Lahaye, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' and Browaeys, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=', Synthetic three-dimensional atomic structures assembled atom by atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} +page_content=' Nature 561, 79 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/CdE3T4oBgHgl3EQfUQrt/content/2301.04450v1.pdf'} diff --git a/DNE5T4oBgHgl3EQfUA_Z/content/2301.05541v1.pdf b/DNE5T4oBgHgl3EQfUA_Z/content/2301.05541v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9cf8e3f7972a64f5baf5851d7cf849dfc20741c8 --- /dev/null +++ b/DNE5T4oBgHgl3EQfUA_Z/content/2301.05541v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:508c4c6eb329b6870843b8718af0c02903e74af0c04201589b4b70b7c1417eb0 +size 3985787 diff --git a/DNE5T4oBgHgl3EQfUA_Z/vector_store/index.faiss b/DNE5T4oBgHgl3EQfUA_Z/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..36c5b4b83af4d588d3077ffba08c5621dd77bbed --- /dev/null +++ b/DNE5T4oBgHgl3EQfUA_Z/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bea867519c4af923aba0437d83baf7c023c97f5375041022e6b16a4d836d1fd0 +size 3342381 diff --git a/DtE5T4oBgHgl3EQfUQ_q/content/tmp_files/2301.05543v1.pdf.txt b/DtE5T4oBgHgl3EQfUQ_q/content/tmp_files/2301.05543v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..b4586b3fcdbeef878072556daa21e0c244468c16 --- /dev/null +++ b/DtE5T4oBgHgl3EQfUQ_q/content/tmp_files/2301.05543v1.pdf.txt @@ -0,0 +1,570 @@ +Classification of Cross-cultural News Events +Abdul Sittar +∗ +abdul.sittar@ijs.si +Jožef Stefan Institute and Jožef Stefan International +Postgraduate School +Jamova cesta 39 +Ljubljana, Slovenia +ABSTRACT +We present a methodology to support the analysis of culture +from text such as news events and demonstrate its usefulness +on categorising news events from different categories (society, +business, health, recreation, science, shopping, sports, arts, com- +puters, games and home) across different geographical locations +(different places in 117 countries). We group countries based on +the culture that they follow and then filter the news events based +on their content category. The news events are automatically +labelled with the help of Hofstede’s cultural dimensions. We +present combinations of events across different categories and +check the performances of different classification methods. We +also presents experimental comparison of different number of +features in order to find a suitable set to represent the culture. +KEYWORDS +cultural barrier, news events, text classification + +1 INTRODUCTION +Culture is defined as a collective programming of the mind which +distinguishes the members of one group or category of people +from another [9]. It has a huge impact on the lives of people and +in result it influences events that involve cross-cultural stake- +holders. News spreading is one of the most effective mechanisms +for spreading information across the borders. The news to be +spread wider cross multiple barriers such as linguistic, economic, +geographical, political, time zone, and cultural barriers. Due to +rapidly growing number of events with significant international +impact, cross-cultural analytics gain increased importance for +professionals and researchers in many disciplines, including digi- +tal humanities, media studies, and journalism. The most recent +examples of such events include COVID-19 and Brexit [1]. There +are few determinants that have significant influence on the pro- +cess of information selection, analysis and propagation. These +include cultural values and differences, economic conditions and +association between countries. For instance, if two countries are +culturally more similar, there are more chances that there will +be a heavier news flow between them [10], [3]. In this paper, +we focus on classification of news events across different cul- +tures. We select some of the most read daily newspapers and +collect information using Event Registry about the news they +have published. Event Registry is a system which analyzes news +articles, identifies groups of articles that describe the same event +and represent them as a single event [7]. The description of the + +Permission to make digital or hard copies of part or all of this work for personal +or classroom use is granted without fee provided that copies are not made or +distributed for profit or commercial advantage and that copies bear this notice and +the full citation on the first page. Copyrights for third-party components of this +work must be honored. For all other uses, contact the owner/author(s). +Information Society 2021, 4–8 October 2021, Ljubljana, Slovenia +© 2021 Copyright held by the owner/author(s). +Dunja Mladenić +dunja.mladenic@ijs.si +Jožef Stefan Institute and Jožef Stefan International +Postgraduate School +Jamova cesta 39 +Ljubljana, Slovenia +meta data of an event is shown in the Table 1. The main scientific +contributions of this paper are the following: +(1) A novel perspective of aligning news events across dif- +ferent cultures through categorising countries and news +events. +(2) A cross-cultural automatically annotated dataset in several +different domains (Business, Science, Sports, Health etc.). +(3) Experimental comparison of several classification mod- +els adopting different set of features (character ngrams, +GLOVE embeddings and word ngrams). +Table 1: The description of the meta data of an event. + Attributes Description + +title +title of the event +summary +summary of the event +source +event reported by a news source +categories +list of DMOZ categories +location +location of the event + + + +2 RELATED WORK +In this section, we review the related literature about the influ- +ence of culture, its representation and classification in different +fields. +Countries that share a common culture are expected to have +heavier news flows between them when reporting on similar +events [10]. There are many quantitative studies that found de- +mographic, psychological, socio-cultural, source, system, and +content-related aspects [2]. +Cross-cultural research and understanding the cultural influences +in different fields have competitive advantages. The goal of re- +searching the impact of culture might be to draw conclusions +in which way the cultural factors influence a specific corporate +action. There are many type of cultures such as societal, organi- +zational, and business culture etc [8]. +The hidden nature of cultural behavior causes some difficulties +in measurement and defining these. To cope with difficulties, +researchers have developed measurements that measure culture +on a general scale to compare differences among cultures and +management styles. These results can be used to find similarities +within a region and differences to other regions. There are many +models that have tried to explain cultural differences between +societies. Hofstede’s national culture dimensions (HNCD) have +been widely used and cited in different disciplines [6, 5]. Hofst- +ede’s dimensions are the result of a factor analysis at the level +of country means of comprehensive survey instrument, aimed +at identifying systematic differences in national cultural. Their +purpose is to measure culture in countries, societies, sub-groups, +and organizations; they are not meant to be regarded as psycho- +logical traits. +There is a plethora of research studies that were conducted to un- +derstand the cultural influences such as cross-culture privacy and + +Information Society 2021, 4–8 October 2021, Ljubljana, Slovenia +Abdul and Dunja, et al. + + +attitude prediction, and cultural influences on today’s business. +[4] explores how culture affects the technological, organizational, +and environmental determinants of machine learning adoption +by conducting a comparative case study between Germany and +US. Rather than looking at the influence of cultural differences +within one domain, we intend to understand association between +news events belonging to different domains (society, business, +health, recreation, science, shopping, sports, arts, computers, +games and home) and different cultures (117 countries from all +the continents). We conduct this research to find an appropriate +representation and classification of culture across different do- +mains. + +3 DATA DESCRIPTION +3.1 Dataset Statistics +We choose the top 10 daily read newspapers in the world in 2020 1 +and collect the events reported by these newspapers using Event +Registry [7] over the time period of 2016-2020. Approximately +8000 events belongs to each newspaper with exception of “Za- +man” that has only 900 events. Figure 1 shows the number of +events reported by the selected newspapers on a yearly basis. +This dataset can be found on the Zenodo repository (version +1.0.0) 2 + + + +Figure 1: Each color in a bar represents the total number +of events per year by a daily newspaper and a complete +bar shows the total number of events per year by all the +newspapers. + +The attributes of an event with description are displayed in +Table 1. Few attributes are self-explanatory such as title, summary, +date, and source. DMOZ-categories are used to represent topics +of the content. The DMOZ project is a hierarchical collection of +web page links organized by subject matters 3. Event Registry use +top 3 levels of DMoz taxonomy which amount to about 50,000 +categories 4. +4 MATERIAL AND METHODS +4.1 Problem Definition +There are two main parts of the problem that we are addressing. +The first part is to label the examples by assigning a culture C to a +news event E using its location L. The second part is a multi-class +classification task where we predict the culture C of a news event +E using its summary description S and its content category G as + +1 https://www.trendrr.net/ +2 https://zenodo.org/record/5225053 +3 https://dmoz-odp.org/ +4 https://eventregistry.org/documentation?tab=terminology + +provided by the Event Registry. This task can be formulated as: +𝐶 = 𝑓 (𝑆, 𝐺) +C donates the culture of the news event, f is the learning function, +S donates summary of a news event and G donates category of a +news event (see Table 1). +4.2 Methodology + + + + + +Figure 2: Classification of cross-cultural news events. + +4.2.1 +Data labeling. Each news event has information about the +type of categories to which it belongs and the location where it +happened (see Table 1). Each event has many categories and each +category has a weight reflecting its relevance for the event. We +only keep the most relevant categories and group the news events +based on their categories. For each group of events, we estimate +the cultural characteristic of each event through the country of +the place where the event occurred. We cluster the countries +based on their culture. We utilize the Hofstede’s national culture +dimensions (HNCD) to represent the culture of a country. We take +average of cultural dimensions and call it average cultural score. +Based on this score, we find optimal number of clusters using +popular clustering algorithm k-means (see Figure 4). Finally, we +label each news event with one of the six cultural clusters. + + + +Figure 3: The pie chart depicts the percentage of the news +events that occurred in six different clusters (each cluster +consists of a list of countries with similar culture). + +4.2.2 +Data representation. Each news event in Event Registry +has associated categories with it along with a weight (see Table +1), we take the top categories based on their weight. In case of +multiple categories with equal weight, we sort them alphabeti- +cally and keep the first one. We represent each news event by a +short summary S and a set of content categories G. +Clusters of Countries +Char Ngrams +News Events +Dataset Annotation +Glove Embeddings +Classification +Category of Events +Word Ngrams + +Newspaper +14k +asahi.com +chinadally.com.cn +dawn.com +12k +nytimes.com +smh.com.au +10k +theguardian.com +timesofindia.indiatimes.com +8k +washingtonpost.com +wsj.com +6k +zaman.com.tr +4k +2k +0 +2016 +2017 +2018 +2019 +2020 +2021ranhnza +ZambiaThaland,Jord +Bangladesh, Lithuaria, Indones +Dominican Republic +Estonia,IndiaChina,Buga +MoccoAigria, +Romania, Serbia, Azerbajan, +Croatia, +Portugat,Boivia, Liby +Chile, Sovenia,Philippin +Amenia, Belarus,SouthKorea, +Malaysia, Uruguay, +24- +Taiwan,Abania,Urae,Mod +Georgia, Argentina, lraq +25% +Montenegro, Czesh Republic +Spain, +Kazakhstan, Bosnia and Herzegovina +Turkey, Brazi, Grece +C3 +C1 +Russia, Slovakia, Japan +Saudi Arabia, Poland +Trindad and Tobago, Mczambique +Colonbia +Ghana,AngolaPueoRi +heral,emkceia +C5 +C2 +Noay,SwnLat +C6 +8.0496 +Niger, Salvador, Venezuela +Finland, ireland,New +Zealand,Nethertands, +21:4% +C4 +Cota ca,cuadrunisia,Egy +Canada, SouthAfrica, +Kuwait,Panaa,Guaa +Australia. +UnitdAbes,Ca +United States,Aistia +8.93% +Suriname +Kingdom,Gemany,taly +Luovembourg,Unitd +12.5% +Malaw,Jamaica,Nepal,SierraLeon +Swtt +Fij,HonurasKya,Bhuta +Hungary, Belgium +Nambia Sianka,Senegal,Burki +FasoSvnaLebanonClassification of Cross-cultural News Events +Information Society 2021, 4–8 October 2021, Ljubljana, Slovenia + + + + + +Figure 4: In word cloud, the color of each word shows cluster to whom it belongs (see Figure 3). Radial dendrograms +illustrate the shared categories of news events between the pair of six clusters. + +4.2.3 +Data Modeling. For multi-class classification task, we use +simple classification models (SVM, Decision Tree, KNN, Naive +Bayes, Logistic Regression) as well as neural network. For sim- +ple classification models, we input character and word ngrams +varying the number of ngrams and compare the results. We also +use pre-trained Glove embeddings. +5 EXPERIMENTAL EVALUATION +5.1 Evaluation Metric +For multi-class classification task, we use following most com- +monly used evaluation measures: accuracy, precision, recall, and +F1 score. +6 RESULTS AND ANALYSIS +6.1 Annotation Results +The results of annotation are six clusters where almost 50% news +events belong to the two clusters (shown with red and blue colors) +and remaining 50% belong to the other four clusters 3. Looking +in each group, we find that clusters do not lies in a specific +geographic area or a continent. Rather all the countries in a +cluster belong to the different continents. Similarly, these clusters +do not have all the countries that are economically rich or poor. +There are more categories in green and red colors in the word +cloud (see Figure 4) which represent to the cluster with that colors. +Radial dendrograms in Figure 4 present the shared categories +between the clusters. In the figure, root of the tree is data and +then there are ten pair of clusters that share the same categories. +The objective of this whole process was to keep news events +according to the category to whom they belongs. Moreover, we +can only observe the cultural differences when we have same +type of news events from different places. +6.2 Classification Results +Fro the experimental results we can see that the best performance +is achieved by Logistic Regression, kNN and Decision Tree. The +performance of SVM varies depending on the number of selected +features: the highest F1-score is achieved with the top 10K or 20K +word ngrams using 1 to 3 word ngrams (see Figure 5). Looking at +the character ngrams, the highest F1-score is achieved when we +select the top 15K characters for all the tested algorithms except +Naive Bayes which declines in performance with the growing +set of features. Based on these settings, we achieve the highest +accuracy (0.85) using Logistic Regression. Using Glove embed- +dings, we experiment with and without using the category of +event. The highest F1-score with and without the category is 0.80 +and 0.79 respectively. + + + +7 CONCLUSIONS AND FUTURE WORK +For researchers and professionals, it is very important to anal- +yse the cross-cultural differences in different disciplines. As the +international impact is increasing and international events are +becoming popular, the need to develop some automatic methods +is significantly increasing and leaving a blank space. We con- +ducted experiments on news events related to different fields +to have a broader look on data and machine learning methods. +Further research would be helpful in examining the impact of +specific socio-cultural factors on news events. In this research +work, we estimate the culture of a specific place by its country, +use basic features and simple classification models. To continue +this work further, we would like to improve feature set such as +by including part of speech tagging (POS) as well as other state +of the art embeddings. + + + +ACKNOWLEDGMENTS +The research described in this paper was supported by the Slove- +nian research agency under the project J2-1736 Causalify and +by the European Union’s Horizon 2020 research and innovation +programme under the Marie Skłodowska-Curie grant agreement +No 812997. + +Shopping.Clothing +echnolog +condt +oistic: +SOrtS +Softwar +Recreation Collecting +ortsBow +Shopping +sports +Societylssues +science +Environment +Business Financial +Services +ArtsMoviess +portsGolfArts_Music +soclety_M +SocietyIssue +Sports +Socce +Society +Eguestrian +Society +Manai +Society GayELesbian andBisexua +creation +Business Food and Related Products +orts-Mart +Soorts +ycilno +Home +am +Collectibles +soorts +BasketbalColads +Envr +"20 +Sports +sety +Equestrian +Strength.Sports. +Team Spinit- +Sports +-Cycling +Rope Skipping +Soccer- +C3C4 +Sports+ +MartialAts +Society* +-Religion and Spintualty +Hockey ++Sports +C2C +Science- +Astronomy +Equestrian +Cycing +Soci +ntsInformation Society 2021, 4–8 October 2021, Ljubljana, Slovenia +Abdul and Dunja, et al. + + + + +Figure 5: First two line charts illustrate the variations in +F1 score by simple classification models after varying the +number of features. The first line chart depicts the results +of word ngrams whereas the second one shows the results +for character ngrams. The last line graph presents com- +parison between Glove embeddings (with and without cat- +egory feature). + +REFERENCES +[1] +Sara Abdollahi, Simon Gottschalk, and Elena Demidova. +2020. Eventkg+ click: a dataset of language-specific event- +centric user interaction traces. arXiv preprint arXiv:2010.12370. +[2] +Hosam Al-Samarraie, Atef Eldenfria, and Husameddin +Dawoud. 2017. The impact of personality traits on users’ +information-seeking behavior. Information Processing & +Management, 53, 1, 237–247. +[3] +Tsan-Kuo Chang and Jae-Won Lee. 1992. Factors affecting +gatekeepers’ selection of foreign news: a national survey +of newspaper editors. Journalism Quarterly, 69, 3, 554–561. +[4] +Verena Eitle and Peter Buxmann. 2020. Cultural differences +in machine learning adoption: an international compari- +son between germany and the united states. +[5] +Meihan He and Jongsu Lee. 2020. Social culture and in- +novation diffusion: a theoretically founded agent-based +model. Journal of Evolutionary Economics, 1–41. +[6] +Mahmood Khosrowjerdi, Anneli Sundqvist, and Katriina +Byström. 2020. Cultural patterns of information source use: +a global study of 47 countries. Journal of the Association +for Information Science and Technology, 71, 6, 711–724. +[7] +Gregor Leban, Blaz Fortuna, Janez Brank, and Marko Gro- +belnik. 2014. Event registry: learning about world events +from news. In Proceedings of the 23rd International Confer- +ence on World Wide Web, 107–110. +[8] +Björn Preuss. 2017. Text mining and machine learning to +capture cultural data. Technical report. working paper, 2. +doi: 10.13140/RG. 2.2. 30937.42080. +[9] +Giselle Rampersad and Turki Althiyabi. 2020. Fake news: +acceptance by demographics and culture on social media. +Journal of Information Technology & Politics, 17, 1, 1–11. +[10] +H Denis Wu. 2007. A brave new world for international +news? exploring the determinants of the coverage of for- +eign news on us websites. International Communication +Gazette, 69, 6, 539–551. + +Top K features versus Accuracy (Word Ngrams, 1-3) +1.0 +SVM +Decision +0.8 +Tree +KNN +Naive +0.6 +Bayes +Logistic +0.4 +Regres.. +0.2 +0.0 +5,000 +10,000 +15,000 +20,000 +Top K +Top K features versus F1 (Char Ngrams, 2-6) +1.0 +SVM +Decision +0.8 +Tree +KNN +Naive +0.6 +Bayes +Logistic +0.4 +Regres.. +0.2 +0.0 +2,500 +5,000 +7,500 +10,000 +12,500 +15,000 +Top K +Categories vs. Without Categories +0.85 +Glove +(with +category) +0.80 +Glove +(without +category) +0.75 +0.70 +0.65 +2 +4 +6 +8 +10 +Epochs \ No newline at end of file diff --git a/DtE5T4oBgHgl3EQfUQ_q/content/tmp_files/load_file.txt b/DtE5T4oBgHgl3EQfUQ_q/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..f57d788d253c8da1a6728cf54b6dd502b1510113 --- /dev/null +++ b/DtE5T4oBgHgl3EQfUQ_q/content/tmp_files/load_file.txt @@ -0,0 +1,262 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf,len=261 +page_content='Classification of Cross-cultural News Events Abdul Sittar ∗ abdul.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='sittar@ijs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='si Jožef Stefan Institute and Jožef Stefan International Postgraduate School Jamova cesta 39 Ljubljana, Slovenia ABSTRACT We present a methodology to support the analysis of culture from text such as news events and demonstrate its usefulness on categorising news events from different categories (society, business, health, recreation, science, shopping, sports, arts, com- puters, games and home) across different geographical locations (different places in 117 countries).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We group countries based on the culture that they follow and then filter the news events based on their content category.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The news events are automatically labelled with the help of Hofstede’s cultural dimensions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We present combinations of events across different categories and check the performances of different classification methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We also presents experimental comparison of different number of features in order to find a suitable set to represent the culture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' KEYWORDS cultural barrier, news events, text classification 1 INTRODUCTION Culture is defined as a collective programming of the mind which distinguishes the members of one group or category of people from another [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' It has a huge impact on the lives of people and in result it influences events that involve cross-cultural stake- holders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' News spreading is one of the most effective mechanisms for spreading information across the borders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The news to be spread wider cross multiple barriers such as linguistic, economic, geographical, political, time zone, and cultural barriers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Due to rapidly growing number of events with significant international impact, cross-cultural analytics gain increased importance for professionals and researchers in many disciplines, including digi- tal humanities, media studies, and journalism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The most recent examples of such events include COVID-19 and Brexit [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' There are few determinants that have significant influence on the pro- cess of information selection, analysis and propagation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' These include cultural values and differences, economic conditions and association between countries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' For instance, if two countries are culturally more similar, there are more chances that there will be a heavier news flow between them [10], [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' In this paper, we focus on classification of news events across different cul- tures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We select some of the most read daily newspapers and collect information using Event Registry about the news they have published.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Event Registry is a system which analyzes news articles, identifies groups of articles that describe the same event and represent them as a single event [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The description of the Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Copyrights for third-party components of this work must be honored.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' For all other uses, contact the owner/author(s).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Information Society 2021, 4–8 October 2021, Ljubljana, Slovenia © 2021 Copyright held by the owner/author(s).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Dunja Mladenić dunja.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='mladenic@ijs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='si Jožef Stefan Institute and Jožef Stefan International Postgraduate School Jamova cesta 39 Ljubljana, Slovenia meta data of an event is shown in the Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The main scientific contributions of this paper are the following: (1) A novel perspective of aligning news events across dif- ferent cultures through categorising countries and news events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' (2) A cross-cultural automatically annotated dataset in several different domains (Business, Science, Sports, Health etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' (3) Experimental comparison of several classification mod- els adopting different set of features (character ngrams, GLOVE embeddings and word ngrams).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Table 1: The description of the meta data of an event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Attributes Description title title of the event summary summary of the event source event reported by a news source categories list of DMOZ categories location location of the event 2 RELATED WORK In this section, we review the related literature about the influ- ence of culture, its representation and classification in different fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Countries that share a common culture are expected to have heavier news flows between them when reporting on similar events [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' There are many quantitative studies that found de- mographic, psychological, socio-cultural, source, system, and content-related aspects [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Cross-cultural research and understanding the cultural influences in different fields have competitive advantages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The goal of re- searching the impact of culture might be to draw conclusions in which way the cultural factors influence a specific corporate action.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' There are many type of cultures such as societal, organi- zational, and business culture etc [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The hidden nature of cultural behavior causes some difficulties in measurement and defining these.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' To cope with difficulties, researchers have developed measurements that measure culture on a general scale to compare differences among cultures and management styles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' These results can be used to find similarities within a region and differences to other regions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' There are many models that have tried to explain cultural differences between societies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Hofstede’s national culture dimensions (HNCD) have been widely used and cited in different disciplines [6, 5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Hofst- ede’s dimensions are the result of a factor analysis at the level of country means of comprehensive survey instrument, aimed at identifying systematic differences in national cultural.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Their purpose is to measure culture in countries, societies, sub-groups, and organizations;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' they are not meant to be regarded as psycho- logical traits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' There is a plethora of research studies that were conducted to un- derstand the cultural influences such as cross-culture privacy and Information Society 2021, 4–8 October 2021, Ljubljana, Slovenia Abdul and Dunja, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' attitude prediction, and cultural influences on today’s business.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [4] explores how culture affects the technological, organizational, and environmental determinants of machine learning adoption by conducting a comparative case study between Germany and US.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Rather than looking at the influence of cultural differences within one domain, we intend to understand association between news events belonging to different domains (society, business, health, recreation, science, shopping, sports, arts, computers, games and home) and different cultures (117 countries from all the continents).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We conduct this research to find an appropriate representation and classification of culture across different do- mains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 3 DATA DESCRIPTION 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='1 Dataset Statistics We choose the top 10 daily read newspapers in the world in 2020 1 and collect the events reported by these newspapers using Event Registry [7] over the time period of 2016-2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Approximately 8000 events belongs to each newspaper with exception of “Za- man” that has only 900 events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Figure 1 shows the number of events reported by the selected newspapers on a yearly basis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' This dataset can be found on the Zenodo repository (version 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0) 2 Figure 1: Each color in a bar represents the total number of events per year by a daily newspaper and a complete bar shows the total number of events per year by all the newspapers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The attributes of an event with description are displayed in Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Few attributes are self-explanatory such as title, summary, date, and source.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' DMOZ-categories are used to represent topics of the content.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The DMOZ project is a hierarchical collection of web page links organized by subject matters 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Event Registry use top 3 levels of DMoz taxonomy which amount to about 50,000 categories 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 4 MATERIAL AND METHODS 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='1 Problem Definition There are two main parts of the problem that we are addressing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The first part is to label the examples by assigning a culture C to a news event E using its location L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The second part is a multi-class classification task where we predict the culture C of a news event E using its summary description S and its content category G as 1 https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='trendrr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='net/ 2 https://zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='org/record/5225053 3 https://dmoz-odp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='org/ 4 https://eventregistry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='org/documentation?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='tab=terminology provided by the Event Registry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' This task can be formulated as: 𝐶 = 𝑓 (𝑆, 𝐺) C donates the culture of the news event, f is the learning function, S donates summary of a news event and G donates category of a news event (see Table 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2 Methodology Figure 2: Classification of cross-cultural news events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='1 Data labeling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Each news event has information about the type of categories to which it belongs and the location where it happened (see Table 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Each event has many categories and each category has a weight reflecting its relevance for the event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We only keep the most relevant categories and group the news events based on their categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' For each group of events, we estimate the cultural characteristic of each event through the country of the place where the event occurred.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We cluster the countries based on their culture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We utilize the Hofstede’s national culture dimensions (HNCD) to represent the culture of a country.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We take average of cultural dimensions and call it average cultural score.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Based on this score, we find optimal number of clusters using popular clustering algorithm k-means (see Figure 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Finally, we label each news event with one of the six cultural clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Figure 3: The pie chart depicts the percentage of the news events that occurred in six different clusters (each cluster consists of a list of countries with similar culture).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2 Data representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Each news event in Event Registry has associated categories with it along with a weight (see Table 1), we take the top categories based on their weight.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' In case of multiple categories with equal weight, we sort them alphabeti- cally and keep the first one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We represent each news event by a short summary S and a set of content categories G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Clusters of Countries Char Ngrams News Events Dataset Annotation Glove Embeddings Classification Category of Events Word Ngrams Newspaper 14k asahi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com chinadally.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='cn dawn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com 12k nytimes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com smh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='au 10k theguardian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com timesofindia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='indiatimes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com 8k washingtonpost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com wsj.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com 6k zaman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='com.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='tr 4k 2k 0 2016 2017 2018 2019 2020 2021ranhnza ZambiaThaland,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Jord Bangladesh,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Lithuaria,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Indones Dominican Republic Estonia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='IndiaChina,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Buga MoccoAigria,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Romania,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Serbia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Azerbajan,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Croatia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Portugat,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Boivia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Liby Chile,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Sovenia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Philippin Amenia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Belarus,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='SouthKorea,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Malaysia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Uruguay,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 24- Taiwan,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Abania,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Urae,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Mod Georgia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Argentina,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' lraq 25% Montenegro,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Czesh Republic Spain,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Kazakhstan,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Bosnia and Herzegovina Turkey,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Brazi,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Grece C3 C1 Russia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Slovakia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Japan Saudi Arabia,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Poland Trindad and Tobago,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Mczambique Colonbia Ghana,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='AngolaPueoRi heral,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='emkceia C5 C2 Noay,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='SwnLat C6 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0496 Niger, Salvador, Venezuela Finland, ireland,New Zealand,Nethertands, 21:4% C4 Cota ca,cuadrunisia,Egy Canada, SouthAfrica, Kuwait,Panaa,Guaa Australia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' UnitdAbes,Ca United States,Aistia 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='93% Suriname Kingdom,Gemany,taly Luovembourg,Unitd 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='5% Malaw,Jamaica,Nepal,SierraLeon Swtt Fij,HonurasKya,Bhuta Hungary, Belgium Nambia Sianka,Senegal,Burki FasoSvnaLebanonClassification of Cross-cultural News Events Information Society 2021, 4–8 October 2021, Ljubljana, Slovenia Figure 4: In word cloud, the color of each word shows cluster to whom it belongs (see Figure 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Radial dendrograms illustrate the shared categories of news events between the pair of six clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='3 Data Modeling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' For multi-class classification task, we use simple classification models (SVM, Decision Tree, KNN, Naive Bayes, Logistic Regression) as well as neural network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' For sim- ple classification models, we input character and word ngrams varying the number of ngrams and compare the results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We also use pre-trained Glove embeddings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 5 EXPERIMENTAL EVALUATION 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='1 Evaluation Metric For multi-class classification task, we use following most com- monly used evaluation measures: accuracy, precision, recall, and F1 score.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 6 RESULTS AND ANALYSIS 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='1 Annotation Results The results of annotation are six clusters where almost 50% news events belong to the two clusters (shown with red and blue colors) and remaining 50% belong to the other four clusters 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Looking in each group, we find that clusters do not lies in a specific geographic area or a continent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Rather all the countries in a cluster belong to the different continents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Similarly, these clusters do not have all the countries that are economically rich or poor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' There are more categories in green and red colors in the word cloud (see Figure 4) which represent to the cluster with that colors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Radial dendrograms in Figure 4 present the shared categories between the clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' In the figure, root of the tree is data and then there are ten pair of clusters that share the same categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The objective of this whole process was to keep news events according to the category to whom they belongs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Moreover, we can only observe the cultural differences when we have same type of news events from different places.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2 Classification Results Fro the experimental results we can see that the best performance is achieved by Logistic Regression, kNN and Decision Tree.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The performance of SVM varies depending on the number of selected features: the highest F1-score is achieved with the top 10K or 20K word ngrams using 1 to 3 word ngrams (see Figure 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Looking at the character ngrams, the highest F1-score is achieved when we select the top 15K characters for all the tested algorithms except Naive Bayes which declines in performance with the growing set of features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Based on these settings, we achieve the highest accuracy (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='85) using Logistic Regression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Using Glove embed- dings, we experiment with and without using the category of event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The highest F1-score with and without the category is 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='80 and 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='79 respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 7 CONCLUSIONS AND FUTURE WORK For researchers and professionals, it is very important to anal- yse the cross-cultural differences in different disciplines.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' As the international impact is increasing and international events are becoming popular, the need to develop some automatic methods is significantly increasing and leaving a blank space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' We con- ducted experiments on news events related to different fields to have a broader look on data and machine learning methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Further research would be helpful in examining the impact of specific socio-cultural factors on news events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' In this research work, we estimate the culture of a specific place by its country, use basic features and simple classification models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' To continue this work further, we would like to improve feature set such as by including part of speech tagging (POS) as well as other state of the art embeddings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' ACKNOWLEDGMENTS The research described in this paper was supported by the Slove- nian research agency under the project J2-1736 Causalify and by the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 812997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Shopping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Clothing echnolog condt oistic: SOrtS Softwar Recreation Collecting ortsBow Shopping sports Societylssues science Environment Business Financial Services ArtsMoviess portsGolfArts_Music soclety_M SocietyIssue Sports Socce Society Eguestrian Society Manai Society GayELesbian andBisexua creation Business Food and Related Products orts-Mart Soorts ycilno Home am Collectibles soorts BasketbalColads Envr "20 Sports sety Equestrian Strength.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='Sports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Team Spinit- Sports Cycling Rope Skipping Soccer- C3C4 Sports+ MartialAts Society* Religion and Spintualty Hockey +Sports C2C Science- Astronomy Equestrian Cycing Soci ntsInformation Society 2021, 4–8 October 2021, Ljubljana, Slovenia Abdul and Dunja, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Figure 5: First two line charts illustrate the variations in F1 score by simple classification models after varying the number of features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The first line chart depicts the results of word ngrams whereas the second one shows the results for character ngrams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The last line graph presents com- parison between Glove embeddings (with and without cat- egory feature).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' REFERENCES [1] Sara Abdollahi, Simon Gottschalk, and Elena Demidova.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Eventkg+ click: a dataset of language-specific event- centric user interaction traces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' arXiv preprint arXiv:2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='12370.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [2] Hosam Al-Samarraie, Atef Eldenfria, and Husameddin Dawoud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' The impact of personality traits on users’ information-seeking behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Information Processing & Management, 53, 1, 237–247.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [3] Tsan-Kuo Chang and Jae-Won Lee.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 1992.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Factors affecting gatekeepers’ selection of foreign news: a national survey of newspaper editors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Journalism Quarterly, 69, 3, 554–561.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [4] Verena Eitle and Peter Buxmann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Cultural differences in machine learning adoption: an international compari- son between germany and the united states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [5] Meihan He and Jongsu Lee.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Social culture and in- novation diffusion: a theoretically founded agent-based model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Journal of Evolutionary Economics, 1–41.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [6] Mahmood Khosrowjerdi, Anneli Sundqvist, and Katriina Byström.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Cultural patterns of information source use: a global study of 47 countries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Journal of the Association for Information Science and Technology, 71, 6, 711–724.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [7] Gregor Leban, Blaz Fortuna, Janez Brank, and Marko Gro- belnik.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Event registry: learning about world events from news.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' In Proceedings of the 23rd International Confer- ence on World Wide Web, 107–110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [8] Björn Preuss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Text mining and machine learning to capture cultural data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Technical report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' working paper, 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='13140/RG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 30937.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='42080.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [9] Giselle Rampersad and Turki Althiyabi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Fake news: acceptance by demographics and culture on social media.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Journal of Information Technology & Politics, 17, 1, 1–11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' [10] H Denis Wu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' A brave new world for international news?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' exploring the determinants of the coverage of for- eign news on us websites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' International Communication Gazette, 69, 6, 539–551.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Top K features versus Accuracy (Word Ngrams, 1-3) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0 SVM Decision 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='8 Tree KNN Naive 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='6 Bayes Logistic 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='4 Regres.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='. 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0 5,000 10,000 15,000 20,000 Top K Top K features versus F1 (Char Ngrams, 2-6) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0 SVM Decision 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='8 Tree KNN Naive 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='6 Bayes Logistic 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='4 Regres.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='. 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='0 2,500 5,000 7,500 10,000 12,500 15,000 Top K Categories vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content=' Without Categories 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='85 Glove (with category) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='80 Glove (without category) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='75 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='70 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} +page_content='65 2 4 6 8 10 Epochs' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/DtE5T4oBgHgl3EQfUQ_q/content/2301.05543v1.pdf'} diff --git a/EtAzT4oBgHgl3EQfG_tj/content/tmp_files/2301.01037v1.pdf.txt b/EtAzT4oBgHgl3EQfG_tj/content/tmp_files/2301.01037v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..1ff8718a7b25a44f8b8cef5f5f111ebea4f9848b --- /dev/null +++ b/EtAzT4oBgHgl3EQfG_tj/content/tmp_files/2301.01037v1.pdf.txt @@ -0,0 +1,377 @@ +Uptrendz: API-Centric Real-time +Recommendations in Multi-Domain Settings +Emanuel Lacic1, Tomislav Duricic1,2, Leon Fadljevic1, +Dieter Theiler1, and Dominik Kowald(�)1,2 +1 Know-Center GmbH, Graz, Austria +{elacic,tduricic,lfadljevic,dtheiler,dkowald}@know-center.at +2 Graz University of Technology, Graz, Austria +Abstract. In this work, we tackle the problem of adapting a real-time +recommender system to multiple application domains, and their underly- +ing data models and customization requirements. To do that, we present +Uptrendz, a multi-domain recommendation platform that can be cus- +tomized to provide real-time recommendations in an API-centric way. +We demonstrate (i) how to set up a real-time movie recommender us- +ing the popular MovieLens-100k dataset, and (ii) how to simultaneously +support multiple application domains based on the use-case of recom- +mendations in entrepreneurial start-up founding. For that, we differenti- +ate between domains on the item- and system-level. We believe that our +demonstration shows a convenient way to adapt, deploy and evaluate a +recommender system in an API-centric way. The source-code and doc- +umentation that demonstrates how to utilize the configured Uptrendz +API is available on GitHub. +Keywords: Uptrendz, API-centric recommendations, multi-domain rec- +ommendations, real-time recommendations +1 +Introduction +Utilizing recommender systems is nowadays recognized as a necessary feature to +help users discover relevant content [15,14]. Most industry practitioners [3], when +they build a recommender system, adapt existing algorithms to the underlying +data and customization requirements of the respective application domain (e.g., +movies, music, news, etc.). However, the focus of the research community has +recently shifted towards building recommendation systems that simultaneously +support multiple application domains [4,7,16] in an API-centric way. +In this work, we demonstrate Uptrendz3, an API-centric recommendation +platform, which can be configured to simultaneously provide real-time recom- +mendations in an API-centric way to multiple domains. Uptrendz supports pop- +ular recommendation algorithms, e.g., Collaborative Filtering (CF), Content- +based Filtering (CBF, or Most Popular (MP), that are applied across different +3 https://uptrendz.ai/ +arXiv:2301.01037v1 [cs.IR] 3 Jan 2023 + +2 +E. Lacic, T. Duricic, L. Fadljevic, D. Theiler, and D. Kowald +RECOMMENDER +CUSTOMIZATION +SERVICE +ISOLATION +DATA +HETEROGENEITY +FAULT +TOLERANCE +MULTI-DOMAIN RECOMMENDER SYSTEM +Fig. 1. Aspects that need to be addressed when building a recommender system for a +multi-domain environment [10]. +application domains. The focus of this demonstration is to show how domain- +specific data-upload APIs can be created to support the customization of the +respective recommendation algorithms. Using the MovieLens-100k dataset [6] +and a real-world use-case of entrepreneurial start-up founding4, we show how +such an approach allows for a highly customized recommendation system that +can be used in an API-centric way. The source-code and documentation for this +demonstration is available via GitHub5. +2 +The Uptrendz Platform +The Uptrendz platform is built on top of the ScaR recommendation framework +[11]. As shown in [10] and Figure 1, the microservice-based system architecture +addresses four distinctive requirements of a multi-domain recommender system, +i.e., (i) service isolation, (ii) data heterogeneity, (iii) recommender customization, +and (iv) fault tolerance. Uptrendz provides a layer on top of the framework to +dynamically configure an application domain and to instantly provide an API +to (i) upload item, user and interaction data, and (ii) request recommendations. +Domain-specific data model. As discussed by [1], different domains may em- +ploy the same recommender algorithm but can differ with respect to what kind +of data is utilized to build the model (e.g., interaction types, context, etc.). Given +an API-centric approach, we show that in order to support the customization of +recommender algorithms with domain-specific parameters, the underlying plat- +form needs to unambiguously know which source of information should be used +to calculate the recommendations. To do that, the Uptrendz platform first allows +generating customized data upload APIs for multiple item and user entities (see +Table 1). Second, with respect to interaction data, both user-item and user-user +interactions can be configured. The interaction API is further customized in ac- +cordance to what kind of interactions the respective application domain actually +supports, i.e., (i) registered users, anonymous sessions or both, (ii) interaction +timestamp tracking, and (iii) type of interaction (explicit or implicit). +4 https://cogsteps.com/ +5 https://github.com/lacic/ECIR2023Demo + +API-Centric Real-time Recommendations in Multi-Domain Settings +3 +Table 1. Supported attributes to configure the data upload API for items and users. +Type +Sub-Type +Description +Categorical +Text +Single +Value +String value, which usually represents a category. Used +for post-filtering recommendation results. +Multiple +Values +List of string values, which usually represent an array +of categories. Used for post-filtering recommendation +results. +Free Text +English +English +text, which is processed and utilized for +content-based recommendations. +German +German text, which is processed and utilized for +content-based recommendations. +Numeric +Integer +Used for post-filtering recommendations (e.g., user +age). +Real +Used for post-filtering recommendations (e.g., price). +Date +- +Date information for the respective entity (e.g., creation +date) +Recommender customization. The Uptrendz platform fosters the notion of +defining personalization scenarios (i.e., use-cases) when creating recommenda- +tion APIs. The available selection of real-time recommendation models [11] for +a given scenario depends on (i) what should be recommended (e.g., item or user +entities), (ii) for whom the recommendations are targeted (e.g., registered or +anonymous users) and, (iii) what kind of context is given [2] (e.g., item ID to +recommend relevant content for). As we adopt a non-restricted configuration +with respect to the number of freely defined user interaction types, algorithms +that use this kind of data (e.g., Collaborative Filtering) can be customized to +utilize any subset of the list of available interactions as well as to define how +much weight a particular interaction type should have. With respect to post- +filtering recommendation results, each model can use categorical (e.g., tags [12] +or other semantic representations [8]) or numerical data attributes to ensure that +the resulting recommendations either contain or exclude a particular value (see +Table 1 for complete list of attributes). +3 +Multi-Domain Support +In order to provide a multi-domain recommender platform, we support the no- +tions of a system-level and item-level domain in accordance with [5]. For the +former, items and users belong to distinct systems (e.g., Netflix and Amazon). +For the latter, individual domains have different types of items and users which +may share some common attributes (e.g., movies and books). +Demo Walkthrough: System-level domain. When a domain is created on +a system level, the underlying data is physically stored in a different location +than the data of other domains. Hence, domains do not share any data between +themselves and the underlying services are isolated so that the performance of + +4 +E. Lacic, T. Duricic, L. Fadljevic, D. Theiler, and D. Kowald +Fig. 2. Example of supporting multiple domains on the item-level (up) and configuring +a hybrid recommendation algorithm (below) with previously created APIs. +one domain does not impact the performance of another domain (e.g., during +request load peaks). We demonstrate how to create a movie recommender on a +system level. To utilize the MovieLens-100k dataset [6], we first need to configure +the respective data services to upload (i) movie, (ii) user, and (iii) interaction +data. Each entity needs to be separately created in the Uptrendz platform in +order to generate an API that can be used to upload the MovieLens-specific +data attributes. This allows creating recommendation scenarios for (i) similar +movies (CBF), (ii) popular horror movies (MP with post-filtering), (iii) movies +based on ratings (CF), (iv) their weighted hybrid combination (e.g., for cold-start +settings [13], and (v) a user recommender for a given movie. +Demo Walkthrough: Item-level domain. To showcase how to configure Up- +trendz to support multiple-domains on an item-level, we present the use-case of +entrepreneurial start-up founding. Here, we recommend experts that can provide + +Available attributes for entity: +news +Field Name +Field Type +Field Subtype +id +→ +Categorical Text +[ Single Value +content +→ +Free Text +→ +English +name +Free Text +→ +English +active +Categorical Text +→ +Single Value +categories +→ +Categorical Text +→ +Multiple Values +Available attributes for entity: +innovation +Field Name +Field Type +Field Subtype +id +→ +Categorical Text +Single Value +author +→ + Categorical Text +→ +Single Value +description +→ +Free Text +→ +[English +name +Free Text +English +headline + Free Text +→ +English +location +→ + Categorical Text +Single Value +development_phase +→ + Categorical Text +→ +Single Value +patent_description +Free Text +→ +English +help_time +→ +Numeric +→ +Integer +active +→ + Categorical Text +→ +Single Value +categories + Categorical Text +Multiple Values +Mutiple Values +fields_of_interest +→ + Categorical Text +→General Settings +Scenario name + discover innovations +Scenario ID: discover-innovations +What will be recommended? +Recommendation Model +innovation +> +HybridRoundRobinWeightedSum +V +Items +innovation +ItemContext +institution +Choose Context +education +news +Model Specific Settings +Users +user +Select all desired scenarios which you would like to include into this hybrid scenario. +In order to prioritize between reference scenarios, for each selected scenario you must assign a proper weight with an integer value +Available profiles + Connect People Innovation Content + Invite People Brainstorm Content + Discover Innovations Personalized +10 + Discover Innovations Popular +1 + Discover Innovations Content History +5API-Centric Real-time Recommendations in Multi-Domain Settings +5 +Fig. 3. Uptrendz requires the specification of (i) the item types that should be recom- +mended (e.g., products or users, depending on the domain - left figure), and (ii) the +user types for which recommendations should be generated (e.g., registered users or +session users - right figure). +feedback to an innovation idea, support co-founder matching, help incubators, +innovation hubs and accelerators to discover innovations but also provide rel- +evant educational materials until the innovation idea matures enough to form +a start-up. In this case, each recommendable entity has a separate data model +and can be viewed as part of a standalone application domain. Figure 2 depicts +how adding multiple item entities in the data catalog allows customizing data +attributes for the respective domain. While configuring a recommendation al- +gorithm, the respective item-level domain can be selected to be recommended. +Here, via the example of a hybrid algorithm, only pre-configured algorithms can +be utilized that belong to the same domain (i.e., innovation recommendations). +Finally, in Figure 3, we show how Uptrendz allows the specification of (i) +different item types that can be recommended, and (ii) different user types for +which recommendations should be generated. Our demo application includes +different specification examples. +4 +Conclusion +In this paper, we present Uptrendz, an API-centric recommendation platform +that can be customized to provide real-time recommendations for multiple do- +mains. To do that, we support the notions of a system-level and item-level do- +main. We demonstrate Uptrendz using the popular MovieLens-100k dataset and +the use-case of entrepreneurial start-up founding. +In future work, we plan to support even more use cases from other domains, +e.g., music recommendations [9]. Here, we also want to integrate fairness-aware +recommendation algorithms for mitigating e.g., popularity bias effects. +Acknowledgements. This research was funded by CogSteps and the “DDAI” +COMET Module within the COMET – Competence Centers for Excellent Tech- +nologies Programme, funded by the Austrian Federal Ministry for Transport, +Innovation and Technology (bmvit), the Austrian Federal Ministry for Digital +and Economic Affairs (bmdw), the Austrian Research Promotion Agency (FFG), +the province of Styria (SFG) and partners from industry and academia. + +What would you like to +围 +recommend? +product, article, job... +Add Item Entity +网What kind of users do +you have? +user +Add User Entity +T6 +E. Lacic, T. Duricic, L. Fadljevic, D. Theiler, and D. Kowald +References +1. Adomavicius, G., Tuzhilin, A.: Context-aware recommender systems. In: Proceed- +ings of the 2008 ACM Conference on Recommender Systems. pp. 335–336. Rec- +Sys ’08, ACM (2008). https://doi.org/10.1145/1454008.1454068, http://doi.acm. +org/10.1145/1454008.1454068 +2. Adomavicius, G., Tuzhilin, A.: Context-aware recommender systems. In: Recom- +mender systems handbook, pp. 217–253. Springer (2011) +3. Amatriain, X., Basilico, J.: Past, present, and future of recommender systems: An +industry perspective. In: Proceedings of the 10th ACM conference on recommender +systems. pp. 211–214 (2016) +4. Bonab, H., Aliannejadi, M., Vardasbi, A., Kanoulas, E., Allan, J.: Cross-market +product recommendation. In: Proceedings of the 30th ACM International Confer- +ence on Information & Knowledge Management. pp. 110–119 (2021) +5. Cantador, I., Fern´andez-Tob´ıas, I., Berkovsky, S., Cremonesi, P.: Cross-domain +recommender systems. In: Recommender Systems Handbook. Springer (2015) +6. Harper, F.M., Konstan, J.A.: The movielens datasets: History and context. Acm +transactions on interactive intelligent systems (tiis) 5(4), 1–19 (2015) +7. Im, I., Hars, A.: Does a one-size recommendation system fit all? the effectiveness +of collaborative filtering based recommendation systems across different domains +and search modes. ACM Transactions on Information Systems (TOIS) 26(1), 4–es +(2007) +8. Kowald, D., Dennerlein, S.M., Theiler, D., Walk, S., Trattner, C.: The social se- +mantic server a framework to provide services on social semantic network data. In: +Proceedings of I-SEMANTICS 2013). pp. 50–54 (2013) +9. Kowald, D., Muellner, P., Zangerle, E., Bauer, C., Schedl, M., Lex, E.: Support +the underground: characteristics of beyond-mainstream music listeners. EPJ Data +Science 10(1), 1–26 (2021) +10. Lacic, E., Kowald, D., Lex, E.: Tailoring recommendations for a multi-domain en- +vironment. Workshop on Intelligent Recommender Systems by Knowledge Transfer +& Learning (RecSysKTL’2017) co-located with the 11th ACM Conference on Rec- +ommender Systems (RecSys’2017) (2017) +11. Lacic, E., Kowald, D., Parra, D., Kahr, M., Trattner, C.: Towards a scalable social +recommender engine for online marketplaces: The case of apache solr. In: Workshop +Proceedings of WWW’2014. pp. 817–822 (2014) +12. Lacic, E., Kowald, D., Seitlinger, P., Trattner, C., Parra, D.: Recommending items +in social tagging systems using tag and time information. Proceedings of the 1st +International Workshop on Social Personalisation (SP’2014) co-located with Hy- +pertext’2014 (2014) +13. Lacic, E., Kowald, D., Traub, M., Luzhnica, G., Simon, J.P., Lex, E.: Tackling cold- +start users in recommender systems with indoor positioning systems. In: Poster +Proceedings of the 9th {ACM} Conference on Recommender Systems. Association +of Computing Machinery (2015) +14. McNee, S.M., Riedl, J., Konstan, J.A.: Being accurate is not enough: how accuracy +metrics have hurt recommender systems. In: CHI’06 extended abstracts on Human +factors in computing systems. pp. 1097–1101. ACM (2006) +15. Resnick, P., Varian, H.R.: Recommender systems. Communications of the ACM +40(3), 56–58 (1997) +16. Roitero, K., Carterette, B., Mehrotra, R., Lalmas, M.: Leveraging behavioral het- +erogeneity across markets for cross-market training of recommender systems. In: +Companion Proceedings of the Web Conference 2020. pp. 694–702 (2020) + diff --git a/EtAzT4oBgHgl3EQfG_tj/content/tmp_files/load_file.txt b/EtAzT4oBgHgl3EQfG_tj/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..1099c86f31d9979c8f940839eb761c70be09d269 --- /dev/null +++ b/EtAzT4oBgHgl3EQfG_tj/content/tmp_files/load_file.txt @@ -0,0 +1,366 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf,len=365 +page_content='Uptrendz: API-Centric Real-time Recommendations in Multi-Domain Settings Emanuel Lacic1, Tomislav Duricic1,2, Leon Fadljevic1, Dieter Theiler1, and Dominik Kowald(�)1,2 1 Know-Center GmbH, Graz, Austria {elacic,tduricic,lfadljevic,dtheiler,dkowald}@know-center.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='at 2 Graz University of Technology, Graz, Austria Abstract.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In this work, we tackle the problem of adapting a real-time recommender system to multiple application domains, and their underly- ing data models and customization requirements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' To do that, we present Uptrendz, a multi-domain recommendation platform that can be cus- tomized to provide real-time recommendations in an API-centric way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' We demonstrate (i) how to set up a real-time movie recommender us- ing the popular MovieLens-100k dataset, and (ii) how to simultaneously support multiple application domains based on the use-case of recom- mendations in entrepreneurial start-up founding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' For that, we differenti- ate between domains on the item- and system-level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' We believe that our demonstration shows a convenient way to adapt, deploy and evaluate a recommender system in an API-centric way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' The source-code and doc- umentation that demonstrates how to utilize the configured Uptrendz API is available on GitHub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Keywords: Uptrendz, API-centric recommendations, multi-domain rec- ommendations, real-time recommendations 1 Introduction Utilizing recommender systems is nowadays recognized as a necessary feature to help users discover relevant content [15,14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Most industry practitioners [3], when they build a recommender system, adapt existing algorithms to the underlying data and customization requirements of the respective application domain (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', movies, music, news, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' However, the focus of the research community has recently shifted towards building recommendation systems that simultaneously support multiple application domains [4,7,16] in an API-centric way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In this work, we demonstrate Uptrendz3, an API-centric recommendation platform, which can be configured to simultaneously provide real-time recom- mendations in an API-centric way to multiple domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Uptrendz supports pop- ular recommendation algorithms, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Collaborative Filtering (CF), Content- based Filtering (CBF, or Most Popular (MP), that are applied across different 3 https://uptrendz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='ai/ arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='01037v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='IR] 3 Jan 2023 2 E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Duricic, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Fadljevic, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Theiler, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Kowald RECOMMENDER CUSTOMIZATION SERVICE ISOLATION DATA HETEROGENEITY FAULT TOLERANCE MULTI-DOMAIN RECOMMENDER SYSTEM Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Aspects that need to be addressed when building a recommender system for a multi-domain environment [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' application domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' The focus of this demonstration is to show how domain- specific data-upload APIs can be created to support the customization of the respective recommendation algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Using the MovieLens-100k dataset [6] and a real-world use-case of entrepreneurial start-up founding4, we show how such an approach allows for a highly customized recommendation system that can be used in an API-centric way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' The source-code and documentation for this demonstration is available via GitHub5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 2 The Uptrendz Platform The Uptrendz platform is built on top of the ScaR recommendation framework [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' As shown in [10] and Figure 1, the microservice-based system architecture addresses four distinctive requirements of a multi-domain recommender system, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', (i) service isolation, (ii) data heterogeneity, (iii) recommender customization, and (iv) fault tolerance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Uptrendz provides a layer on top of the framework to dynamically configure an application domain and to instantly provide an API to (i) upload item, user and interaction data, and (ii) request recommendations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Domain-specific data model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' As discussed by [1], different domains may em- ploy the same recommender algorithm but can differ with respect to what kind of data is utilized to build the model (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', interaction types, context, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Given an API-centric approach, we show that in order to support the customization of recommender algorithms with domain-specific parameters, the underlying plat- form needs to unambiguously know which source of information should be used to calculate the recommendations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' To do that, the Uptrendz platform first allows generating customized data upload APIs for multiple item and user entities (see Table 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Second, with respect to interaction data, both user-item and user-user interactions can be configured.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' The interaction API is further customized in ac- cordance to what kind of interactions the respective application domain actually supports, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', (i) registered users, anonymous sessions or both, (ii) interaction timestamp tracking, and (iii) type of interaction (explicit or implicit).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 4 https://cogsteps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='com/ 5 https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='com/lacic/ECIR2023Demo API-Centric Real-time Recommendations in Multi-Domain Settings 3 Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Supported attributes to configure the data upload API for items and users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Type Sub-Type Description Categorical Text Single Value String value, which usually represents a category.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Used for post-filtering recommendation results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Multiple Values List of string values, which usually represent an array of categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Used for post-filtering recommendation results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Free Text English English text, which is processed and utilized for content-based recommendations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' German German text, which is processed and utilized for content-based recommendations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Numeric Integer Used for post-filtering recommendations (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', user age).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Real Used for post-filtering recommendations (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', price).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Date Date information for the respective entity (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', creation date) Recommender customization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' The Uptrendz platform fosters the notion of defining personalization scenarios (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', use-cases) when creating recommenda- tion APIs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' The available selection of real-time recommendation models [11] for a given scenario depends on (i) what should be recommended (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', item or user entities), (ii) for whom the recommendations are targeted (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', registered or anonymous users) and, (iii) what kind of context is given [2] (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', item ID to recommend relevant content for).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' As we adopt a non-restricted configuration with respect to the number of freely defined user interaction types, algorithms that use this kind of data (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Collaborative Filtering) can be customized to utilize any subset of the list of available interactions as well as to define how much weight a particular interaction type should have.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' With respect to post- filtering recommendation results, each model can use categorical (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', tags [12] or other semantic representations [8]) or numerical data attributes to ensure that the resulting recommendations either contain or exclude a particular value (see Table 1 for complete list of attributes).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 3 Multi-Domain Support In order to provide a multi-domain recommender platform, we support the no- tions of a system-level and item-level domain in accordance with [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' For the former, items and users belong to distinct systems (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Netflix and Amazon).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' For the latter, individual domains have different types of items and users which may share some common attributes (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', movies and books).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Demo Walkthrough: System-level domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' When a domain is created on a system level, the underlying data is physically stored in a different location than the data of other domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Hence, domains do not share any data between themselves and the underlying services are isolated so that the performance of 4 E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Duricic, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Fadljevic, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Theiler, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Kowald Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Example of supporting multiple domains on the item-level (up) and configuring a hybrid recommendation algorithm (below) with previously created APIs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' one domain does not impact the performance of another domain (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', during request load peaks).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' We demonstrate how to create a movie recommender on a system level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' To utilize the MovieLens-100k dataset [6], we first need to configure the respective data services to upload (i) movie, (ii) user, and (iii) interaction data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Each entity needs to be separately created in the Uptrendz platform in order to generate an API that can be used to upload the MovieLens-specific data attributes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' This allows creating recommendation scenarios for (i) similar movies (CBF), (ii) popular horror movies (MP with post-filtering), (iii) movies based on ratings (CF), (iv) their weighted hybrid combination (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', for cold-start settings [13], and (v) a user recommender for a given movie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Demo Walkthrough: Item-level domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' To showcase how to configure Up- trendz to support multiple-domains on an item-level, we present the use-case of entrepreneurial start-up founding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Here,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' we recommend experts that can provide ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Available attributes for entity: ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='news ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Field Name ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Field Type ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Field Subtype ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='id ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='[ Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='content ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Free Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='English ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='name ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Free Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='English ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='active ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='categories ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Multiple Values ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Available attributes for entity: ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='innovation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Field Name ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Field Type ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Field Subtype ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='id ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='author ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='description ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Free Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='[English ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='name ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Free Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='English ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='headline ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Free Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='English ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='location ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='development_phase ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='patent_description ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Free Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='English ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='help_time ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Numeric ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Integer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='active ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Single Value ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='categories ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Multiple Values ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Mutiple Values ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='fields_of_interest ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Categorical Text ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='→General Settings ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Scenario name ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='discover innovations ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='Scenario ID: discover-innovations ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='What will be recommended?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Recommendation Model innovation > HybridRoundRobinWeightedSum V Items innovation ItemContext institution Choose Context education news Model Specific Settings Users user Select all desired scenarios which you would like to include into this hybrid scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In order to prioritize between reference scenarios, for each selected scenario you must assign a proper weight with an integer value Available profiles Connect People Innovation Content Invite People Brainstorm Content Discover Innovations Personalized 10 Discover Innovations Popular 1 Discover Innovations Content History 5API-Centric Real-time Recommendations in Multi-Domain Settings 5 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Uptrendz requires the specification of (i) the item types that should be recom- mended (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', products or users, depending on the domain - left figure), and (ii) the user types for which recommendations should be generated (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', registered users or session users - right figure).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' feedback to an innovation idea, support co-founder matching, help incubators, innovation hubs and accelerators to discover innovations but also provide rel- evant educational materials until the innovation idea matures enough to form a start-up.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In this case, each recommendable entity has a separate data model and can be viewed as part of a standalone application domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Figure 2 depicts how adding multiple item entities in the data catalog allows customizing data attributes for the respective domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' While configuring a recommendation al- gorithm, the respective item-level domain can be selected to be recommended.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Here, via the example of a hybrid algorithm, only pre-configured algorithms can be utilized that belong to the same domain (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', innovation recommendations).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Finally, in Figure 3, we show how Uptrendz allows the specification of (i) different item types that can be recommended, and (ii) different user types for which recommendations should be generated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Our demo application includes different specification examples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 4 Conclusion In this paper, we present Uptrendz, an API-centric recommendation platform that can be customized to provide real-time recommendations for multiple do- mains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' To do that, we support the notions of a system-level and item-level do- main.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' We demonstrate Uptrendz using the popular MovieLens-100k dataset and the use-case of entrepreneurial start-up founding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In future work, we plan to support even more use cases from other domains, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', music recommendations [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Here, we also want to integrate fairness-aware recommendation algorithms for mitigating e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', popularity bias effects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Acknowledgements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' This research was funded by CogSteps and the “DDAI” COMET Module within the COMET – Competence Centers for Excellent Tech- nologies Programme, funded by the Austrian Federal Ministry for Transport, Innovation and Technology (bmvit), the Austrian Federal Ministry for Digital and Economic Affairs (bmdw), the Austrian Research Promotion Agency (FFG), the province of Styria (SFG) and partners from industry and academia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' What would you like to 围 recommend?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' product, article, job.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Add Item Entity 网What kind of users do you have?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' user Add User Entity T6 E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Duricic, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Fadljevic, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Theiler, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Kowald References 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Adomavicius, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Tuzhilin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Context-aware recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Proceed- ings of the 2008 ACM Conference on Recommender Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 335–336.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Rec- Sys ’08, ACM (2008).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='1145/1454008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='1454068, http://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='acm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='1145/1454008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='1454068 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Adomavicius, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Tuzhilin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Context-aware recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Recom- mender systems handbook, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 217–253.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Springer (2011) 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Amatriain, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Basilico, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Past, present, and future of recommender systems: An industry perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Proceedings of the 10th ACM conference on recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 211–214 (2016) 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Bonab, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Aliannejadi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Vardasbi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Kanoulas, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Allan, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Cross-market product recommendation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Proceedings of the 30th ACM International Confer- ence on Information & Knowledge Management.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 110–119 (2021) 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Cantador, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Fern´andez-Tob´ıas, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Berkovsky, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Cremonesi, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Cross-domain recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Recommender Systems Handbook.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Springer (2015) 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Harper, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Konstan, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' : The movielens datasets: History and context.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Acm transactions on interactive intelligent systems (tiis) 5(4), 1–19 (2015) 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Im, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Hars, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Does a one-size recommendation system fit all?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' the effectiveness of collaborative filtering based recommendation systems across different domains and search modes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' ACM Transactions on Information Systems (TOIS) 26(1), 4–es (2007) 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Kowald, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Dennerlein, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Theiler, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Walk, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Trattner, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': The social se- mantic server a framework to provide services on social semantic network data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Proceedings of I-SEMANTICS 2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 50–54 (2013) 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Kowald, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Muellner, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Zangerle, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Bauer, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Schedl, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Lex, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Support the underground: characteristics of beyond-mainstream music listeners.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' EPJ Data Science 10(1), 1–26 (2021) 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Kowald, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Lex, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Tailoring recommendations for a multi-domain en- vironment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Workshop on Intelligent Recommender Systems by Knowledge Transfer & Learning (RecSysKTL’2017) co-located with the 11th ACM Conference on Rec- ommender Systems (RecSys’2017) (2017) 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Kowald, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Parra, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Kahr, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Trattner, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Towards a scalable social recommender engine for online marketplaces: The case of apache solr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Workshop Proceedings of WWW’2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 817–822 (2014) 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Kowald, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Seitlinger, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Trattner, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Parra, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Recommending items in social tagging systems using tag and time information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Proceedings of the 1st International Workshop on Social Personalisation (SP’2014) co-located with Hy- pertext’2014 (2014) 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Lacic, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Kowald, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Traub, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Luzhnica, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Simon, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Lex, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Tackling cold- start users in recommender systems with indoor positioning systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Poster Proceedings of the 9th {ACM} Conference on Recommender Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Association of Computing Machinery (2015) 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' McNee, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Riedl, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Konstan, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' : Being accurate is not enough: how accuracy metrics have hurt recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: CHI’06 extended abstracts on Human factors in computing systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 1097–1101.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' ACM (2006) 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Resnick, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Varian, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' : Recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Communications of the ACM 40(3), 56–58 (1997) 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' Roitero, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Carterette, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Mehrotra, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=', Lalmas, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=': Leveraging behavioral het- erogeneity across markets for cross-market training of recommender systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' In: Companion Proceedings of the Web Conference 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} +page_content=' 694–702 (2020)' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/EtAzT4oBgHgl3EQfG_tj/content/2301.01037v1.pdf'} diff --git a/EtFLT4oBgHgl3EQfFS_K/content/2301.11987v1.pdf b/EtFLT4oBgHgl3EQfFS_K/content/2301.11987v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..993c14a56bfe9915c60dadc1c84b0a3fa9465b4a --- /dev/null +++ b/EtFLT4oBgHgl3EQfFS_K/content/2301.11987v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3569e785672fc482c15bfd1ff8fed001c9fcaf29c6fe5a958a13fe6e3156451d +size 709796 diff --git a/EtFLT4oBgHgl3EQfFS_K/vector_store/index.pkl b/EtFLT4oBgHgl3EQfFS_K/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..504d9678a6f34c040ba26f25aa1a3b371f6c22f2 --- /dev/null +++ b/EtFLT4oBgHgl3EQfFS_K/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6d7a1310490ddff96369ceec5293466bf1c3f13df350b36ac5da46cf3225e29 +size 144407 diff --git a/GdFKT4oBgHgl3EQfby5L/content/2301.11813v1.pdf b/GdFKT4oBgHgl3EQfby5L/content/2301.11813v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c4caf1db0a96a7e6607de61b1a841248788f2d9a --- /dev/null +++ b/GdFKT4oBgHgl3EQfby5L/content/2301.11813v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7f72f4ff72da9c45fcea3862867a48d7a98741945b357f6d8921a32474c5c6b +size 3304692 diff --git a/GdFKT4oBgHgl3EQfby5L/vector_store/index.faiss b/GdFKT4oBgHgl3EQfby5L/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..13e94a6006e7329a7e9448f59a64682a0e9ae190 --- /dev/null +++ b/GdFKT4oBgHgl3EQfby5L/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c420fe2ae7da5ac7fb269a31f373e3d5b705a1aac0990b5b2cc6d0105a9aca20 +size 6619181 diff --git a/INAyT4oBgHgl3EQf5vp6/content/tmp_files/2301.00810v1.pdf.txt b/INAyT4oBgHgl3EQf5vp6/content/tmp_files/2301.00810v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..c89e1cc019557d8abdbb14f01f5250e1df5c1515 --- /dev/null +++ b/INAyT4oBgHgl3EQf5vp6/content/tmp_files/2301.00810v1.pdf.txt @@ -0,0 +1,1566 @@ +SIRL: Similarity-based Implicit Representation Learning +Andreea Bobu∗ +abobu@berkeley.edu +University of California, Berkeley +United States of America +Yi Liu∗ +yiliu77@berkeley.edu +University of California, Berkeley +United States of America +Rohin Shah +rohinmshah@deepmind.com +DeepMind Research +United Kingdom +Daniel S. Brown +dsbrown@berkeley.edu +University of California, Berkeley +United States of America +Anca D. Dragan +anca@berkeley.edu +University of California, Berkeley +United States of America +ɸ +Similarity +query +Preference +query H1 +Preference +query H2 +Preference +query H3 +Optimize 𝑅!! +Optimize 𝑅!" +Optimize 𝑅!# +𝑅!! +𝑅!" +𝑅!# +Figure 1: Our goal is to learn representations for robot behavior that capture what is salient to people, and, thus, support generalizable pref- +erence learning with low sample complexity. We propose to extract this representation by asking people trajectory similarity queries (left), +where they judge which two out of three trajectories are most similar to each other. We then use the representation to learn reward functions +corresponding to different people’s preferences on different tasks (right). +ABSTRACT +When robots learn reward functions using high capacity models +that take raw state directly as input, they need to both learn a +representation for what matters in the task — the task “features" — +as well as how to combine these features into a single objective. If +they try to do both at once from input designed to teach the full +reward function, it is easy to end up with a representation that +contains spurious correlations in the data, which fails to generalize +to new settings. Instead, our ultimate goal is to enable robots to +identify and isolate the causal features that people actually care +about and use when they represent states and behavior. Our idea +is that we can tune into this representation by asking users what +behaviors they consider similar: behaviors will be similar if the +features that matter are similar, even if low-level behavior is dif- +ferent; conversely, behaviors will be different if even one of the +∗Both authors contributed equally to this research. +Permission to make digital or hard copies of part or all of this work for personal or +classroom use is granted without fee provided that copies are not made or distributed +for profit or commercial advantage and that copies bear this notice and the full citation +on the first page. Copyrights for third-party components of this work must be honored. +For all other uses, contact the owner/author(s). +HRI ’23, March 13–16, 2023, Stockholm, Sweden +© 2023 Copyright held by the owner/author(s). +ACM ISBN 978-1-4503-9964-7/23/03. +https://doi.org/10.1145/3568162.3576989 +features that matter differs. This, in turn, is what enables the robot +to disambiguate between what needs to go into the representation +versus what is spurious, as well as what aspects of behavior can be +compressed together versus not. The notion of learning representa- +tions based on similarity has a nice parallel in contrastive learning, +a self-supervised representation learning technique that maps vi- +sually similar data points to similar embeddings, where similarity +is defined by a designer through data augmentation heuristics. By +contrast, in order to learn the representations that people use, so we +can learn their preferences and objectives, we use their definition +of similarity. In simulation as well as in a user study, we show that +learning through such similarity queries leads to representations +that, while far from perfect, are indeed more generalizable than +self-supervised and task-input alternatives. +KEYWORDS +robot reward learning, human-aligned representation learning +ACM Reference Format: +Andreea Bobu, Yi Liu, Rohin Shah, Daniel S. Brown, and Anca D. Dragan. +2023. SIRL: Similarity-based Implicit Representation Learning. In Proceedings +of the 2023 ACM/IEEE International Conference on Human-Robot Interaction +(HRI ’23), March 13–16, 2023, Stockholm, Sweden. ACM, New York, NY, USA, +12 pages. https://doi.org/10.1145/3568162.3576989 +arXiv:2301.00810v1 [cs.RO] 2 Jan 2023 + +GNParams +Preference query1 +ReplayRed +Replay Blue +Red better than Blue +Blue better than Red +Side View 1 +Side View 2 +FrontView1 +Front View 2 +Top ViewHRI ’23, March 13–16, 2023, Stockholm, Sweden +TBD +1 +INTRODUCTION +Imagine waking up in the morning and your home robot assistant +wants to place a steaming mug of fresh coffee on the table exactly +where it knows you will sit. Depending on the context, you will +have a different preference for how the robot should be doing its +task. Some days it carries your favorite mug close to the table to +prevent it from breaking in the case of a slip (so that it will remain +your favorite mug); other days the steam from your delicious meal +is difficult to handle for the robot’s perception, so you’d want it to +keep a large clearance from the table to avoid collisions. Similarly, +some days you want the robot to keep your mug away from your +laptop to avoid spilling on it; other days the mug only has an +espresso shot so you want the robot keep it close to the laptop to +prevent clutter and leave the rest of the table open for you. +The reward function the robot should optimize changes — whether +due to variations in the task, having different users, or, as in the +examples above, different contexts that are not always part of the +robot state (e.g. holding the user’s favorite mug and not just a regu- +lar mug). However, the representation on top of which the reward is +built, i.e the features that are important (like the distance from the +table, being above the laptop, etc.), are shared. If the robot learns +this representation correctly, it can use it to obtain the right reward +function, even if the task, user, and context changes. Meta-learning +and multi-task learning methods [30, 46, 56] learn the representa- +tion from user input meant to teach the full reward, like preference +queries or demonstrations. By contrast, we propose that if learning +generalizable representations is the goal, then we should ask the +user for input that is specifically meant to teach the representation +itself, rather than input meant to teach the full reward and hoping +to extract a good representation along the way. +But asking people to teach robots representations, rather than +tasks, is not so easy. What are the features that they care about? +While some techniques advocate for people enabling users to teach +each feature separately [11], people may not always be able to +explicate their representation and break it down into concepts that +are individually teachable. In this work, our idea is that we can +implicitly tune into the representations people use by asking them +to do a proxy task of evaluating similarity of behaviors. Behaviors +will be similar if the features that matter are similar, even if low-level +behavior is different; conversely, behaviors will be different if even +one of the features that matter differs. This, in turn, should enable +the robot to arrive at the features that matter — we want robots that +can disambiguate between what needs to go into the representation +versus what is spurious, as well as what aspects of behavior can be +compressed together into a feature embedding versus kept separate. +We thus introduce a novel type of human input to help the robot +extract the person’s representation: trajectory similarity queries. +A trajectory similarity query is a triplet of trajectories that the +person answers by picking the two more similar trajectories. In +Figure 1 (left), the person chooses the two trajectories that are close +to the table and far from the laptop, even though visually they +look dissimilar. This results in an (anchor, positive, negative) triplet +that can be used for training a feature representation. We call this +process Similarity-based Implicit Representation Learning (SIRL). +Our method has a parallel in self-supervised learning work, espe- +cially contrastive learning, where the goal is to learn a good visual +representation by training from (anchor, positive, negative) triplets +generated via data augmentation techniques [19]. However, this +notion of similarity is purely visual, driven by manually designed +heuristics for data augmentation, and is not necessarily reflective of +what users would consider similar. For instance, two images might +be labeled as visually different, when in fact their difference is only +with respect to some low-level aspects that are not really relevant +to the distribution of tasks people care about. This would result +in representations that contain too many distractor features that +are not present in the human’s representation. Our method uses +similarity too, but we defer to the user’s judgement of similarity, +with the goal of reconstructing the user’s representation. +Of course, our method is not the full answer to learning causally +aligned representations. But our experiments suggest that it out- +performs methods that are self-supervised, or that learn from input +meant to teach the full tasks. In simulation, where we know the +causal features, we show that SIRL learns representations better +aligned with them, which in turn leads to learning multiple more +generalizable reward functions downstream (Figure 1). We also +present a user study where we crowdsource similarity queries from +different people to learn a shared SIRL representation that better re- +covers each of their individual preferences. While the study results +do show a significant effect, the effect size is much lower than in +simulation. This is attributable in part to the interface difficulty of +analyzing the robot trajectories, which means more work is needed +to determine the best interfaces that enable users to accurately +answer similarity queries. Moreover, some users reported strug- +gling to trade off the different features, which means that similarity +queries might not be entirely preference-agnostic. Nonetheless, our +results underscore that there are gains by explicitly aligning robot +and human representations, rather than hoping it will happen as a +byproduct of learning rewards from standard queries. +2 +RELATED WORK +Learning from Human Input. Human-in-the-loop learning is a +well-established paradigm where the robot uses human input to +infer a policy or reward function capturing the desired behavior. +In imitation learning, the robot learns a policy that essentially +copies human demonstrations [47], a strategy that typically doesn’t +generalize well outside the training regime [40]. Meanwhile, inverse +reinforcement learning (IRL) uses the demonstrations to extract +a reward function capturing why a specific behavior is desirable, +thus better generalizing to unseen scenarios [1]. Recent research +goes beyond demonstrations, utilizing other types of human input +for reward learning, such as corrections [6], comparisons [55], or +rankings [14]. Unless explicitly designed for, these methods learn a +latent representation implicitly from the respective human input. +We seek to instead explicitly learn a preference-agnostic latent space +that can be used for downstream tasks like reward learning. We +focus on learning models of human reward functions via pairwise +preference queries [55], but we believe the latent space we learn +can be useful for learning from any of the above types of feedback. +Representation and Similarity Learning. Common represen- +tation learning approaches are unsupervised [18, 20, 32] or self- +supervised [4, 27, 39, 48], but because they are purposefully de- +signed to bypass human supervision, the learned embedding does + +SIRL: Similarity-based Implicit Representation Learning +HRI ’23, March 13–16, 2023, Stockholm, Sweden +not necessarily correspond to features the person cares about. Prior +work leverages task labels [17] or trajectory rankings [13] to learn +latent spaces that help identify specific goals or preferences. By +contrast, we focus on learning task-agnositic measures of feature +similarity that are useful for learning multiple preferences. Some +work looks at having people interactively select features from a +pre-defined set [15, 16, 42] or teach task-agnostic features sequen- +tially via kinesthetic feature demonstrations [10] or active learning +techniques [9, 31, 37]. We instead focus on fully learning a lower- +dimensional feature representation all-at-once, rather than one at +a time. Furthermore, rather than relying on the human to provide +physical demonstrations for learning a good feature space [10, 11], +we propose a more accessible and general form of human feedback: +showing the user triplets of trajectories and simply asking them +to label which two trajectories are the most similar. Triplet losses +have been widely used to learn similarity models that capture how +humans perceive objects [2, 3, 25, 44, 54]; however, to the best of our +knowledge, we are the first to use a triplet loss to learn a general, +task-agnostic similarity model of how humans perceive trajectories. +Meta- and Multi-Task Reward Function Learning. To learn +multiple models of human reward functions, prior work has pro- +posed clustering demonstrations and learning a different reward +function for each cluster [5, 21, 26]; however, these methods require +a large number of demonstrations and do not adapt to new reward +functions. Meta-learning [29] seeks to learn a reward function ini- +tialization that enables fast fine-tuning at test time [33, 51, 57, 58]. +Multi-task reward learning approaches pretrain a reward function +on multiple human intents and then fine-tune the reward function +at test time [30, 46]. This has been shown to be more stable and scal- +able than meta-learning approaches [43], but still requires curating +a large set of training environments. By contrast, we do not assume +any knowledge of the test-time task distribution a priori and do not +require access to a population of different reward functions dur- +ing training. Rather, we focus on learning a task-agnostic feature +representation that can be utilized for down-stream reward learn- +ing tasks. In particular, we test our learned representation on the +down-stream task of learning models of human reward functions +via pairwise preference queries over trajectories [8, 41, 50, 55]. +3 +METHOD +We present our method for learning preference-agnostic represen- +tations from trajectory similarity queries. Our intuition is that if a +human judges two behaviors to be similar, then their representa- +tions should also be similar. Since directly asking if two trajectories +are similar is difficult without an explicit threshold, we instead +present the human with a triplet of trajectories and ask them to +pick the two most similar (or, equivalently, the most dissimilar one). +We use the human’s answers to train the representation such that +similar trajectories have embeddings that are close and dissimilar +trajectories map to embeddings far apart. The robot then uses this +latent space as a shared representation for downstream preference +learning tasks with multiple people, each with different preferences. +3.1 +Preliminaries +We define a trajectory 𝜉 as a sequence of states, and denote the +space of all possible trajectories by Ξ. The human’s preference +over trajectories is given by a reward function 𝑅 : Ξ ↦→ R that is +unobserved by the robot and must be learned from human inter- +action. The robot reasons over a parameterized approximation of +the reward function 𝑅𝜃, where 𝜃 represents the parameters of a +neural network. To learn 𝜃, the robot collects human preference +labels over trajectories [22, 55] and seeks to find parameters 𝜃 that +maximize the likelihood of the human input. The robot can then +use the learned reward function to score trajectories during motion +planning in order to align its behavior with a particular human’s +preferences. We focus on explicitly using human input to first learn +a good representation and then use that representation for down- +stream reward learning, rather than using reward-specific human +input (e.g., preferences or demonstrations) to implicitly learn the +representation at the same time as the reward function. +3.2 +Training the Feature Representation via +Trajectory Similarity Queries +We seek to train a latent space that is useful for multiple down- +stream preference learning tasks. To do this, we propose learning a +preference-agnostic model of human similarity. One way to learn +such a model would be to ask users to judge whether two trajecto- +ries are similar or not; however, humans are better at giving relative +rather than binary or quantitative assessments of similarity [35, 53]. +Thus, rather than asking users to use some internal threshold or +scoring mechanism to quantitatively measure similarity, we instead +focus on qualitative trajectory similarity queries. We present the +user with a visualization of three trajectories and ask them to pick +the two most similar ones (equivalently the most dissimilar one). +The human’s queries form a data set D𝑠𝑖𝑚 = {(𝜉𝑖 +𝑃1, 𝜉𝑖 +𝑃2, 𝜉𝑖 +𝑁 )}, where +𝜉𝑖 +𝑃1 and 𝜉𝑖 +𝑃2 are the trajectories that are most similar and 𝜉𝑖 +𝑁 is the +trajectory most dissimilar to the other two. +We can interpret similarity (or dissimilarity) as a distance func- +tion, so we define the distance between two trajectories as the 𝐿2 +feature distance: 𝑑(𝜉1, 𝜉2) = ∥𝜙(𝜉1) − 𝜙(𝜉2)∥2 +2. Given a dataset of +trajectory similarity queries D𝑠𝑖𝑚, we use the triplet loss [7]: +L𝑡𝑟𝑖𝑝 (𝜉𝐴, 𝜉𝑃, 𝜉𝑁 ) = max(𝑑(𝜉𝐴, 𝜉𝑃) − 𝑑(𝜉𝐴, 𝜉𝑁 ) + 𝛼, 0) , +(1) +a form of contrastive learning where 𝜉𝐴 is the anchor, 𝜉𝑃 is the +positive example, 𝜉𝑁 is the negative example, and 𝛼 ≥ 0 is a margin +between positive and negative pairs. However, because our queries +do not contain an explicit anchor, our final loss is as follows: +L𝑠𝑖𝑚(𝜙) = +|D𝑠𝑖𝑚 | +∑︁ +𝑖=1 +L𝑡𝑟𝑖𝑝 (𝜉𝑖 +𝑃1, 𝜉𝑖 +𝑃2, 𝜉𝑖 +𝑁 ) + L𝑡𝑟𝑖𝑝 (𝜉𝑖 +𝑃2, 𝜉𝑖 +𝑃1, 𝜉𝑖 +𝑁 ) . (2) +We train a similarity embedding function 𝜙 : Ξ ↦→ R𝑑 that mini- +mizes the above similarity loss, where 𝑑 is the representation di- +mensionality. The intuition is that optimizing this loss should push +together the embeddings of similar trajectories and push apart the +embeddings of dissimilar trajectories. Before training the repre- +sentation with the loss in Eq. (2), we may also pre-train it using +unsupervised learning [36], which we experiment with in Sec. 4. +3.3 +Using SIRL for Reward Learning +Given a learned embedding 𝜙, we can use it for learning models of +specific user preferences. While we focus on learning from pairwise + +HRI ’23, March 13–16, 2023, Stockholm, Sweden +TBD +preferences, we note that 𝜙 can in principle be used in downstream +tasks that learn from many types of human feedback [34]. When +learning a reward function from human preferences, we show the +human two trajectories, 𝜉𝐴 and 𝜉𝐵, and then ask which of these +two the human prefers. We collect a data set of such preferences +D𝑝𝑟𝑒𝑓 = {(𝜉𝑖 +𝐴, 𝜉𝑖 +𝐵, ℓ𝑖)} where ℓ𝑖 = 1 if 𝜉𝑖 +𝐴 is preferred to 𝜉𝑖 +𝐵, denoted +𝜉𝑖 +𝐴 ≻ 𝜉𝑖 +𝐵, and ℓ𝑖 = 0 otherwise. We interpret the human’s prefer- +ences through the lens of the Bradley-Terry preference model [12]: +𝑃𝜃 (𝜉𝐴 ≻ 𝜉𝐵) = +𝑒𝑅𝜃 (𝜙 (𝜉𝐴)) +𝑒𝑅𝜃 (𝜙 (𝜉𝐴)) + 𝑒𝑅𝜃 (𝜙 (𝜉𝐵)) . +(3) +We learn the reward function with a simple cross-entropy loss: +L𝑝𝑟𝑒𝑓 (𝜃) = − +|D𝑝𝑟𝑒𝑓 | +∑︁ +𝑖=0 +ℓ𝑖 log 𝑃𝜃 (𝜉𝑖 +𝐴 ≻ 𝜉𝑖 +𝐵)+(1−ℓ𝑖) log 𝑃𝜃 (𝜉𝑖 +𝐵 ≻ 𝜉𝑖 +𝐴) . +(4) +3.4 +Adapting to Different User Preferences +We want robots that can adapt to changes to an individual user’s +preferences depending on the context as well as quickly adapt to +new users’ preferences. Rather than learn each preference inde- +pendently by collecting a new set of human data and training a +completely new reward function 𝑅𝜃, we study whether we can +leverage the latent space learned by SIRL to perform more accurate +and sample-efficient multi-preference learning. When learning a +new user’s preference model, 𝑅𝜃 the robot can use 𝜙 to more quickly +learn the reward function 𝑅𝜃 (𝜙(𝜉)). Our main idea is that because +this shared latent representation𝜙 is trained via preference-agnostic +similarity queries, it is more transferable than using a multi-task or +meta-learning approach, where the pre-trained network is trained +using multiple, specific task objectives. Furthermore, because SIRL +uses human input to train 𝜙, we hypothesize that the learned fea- +ture space will be better suited for learning human reward functions +than a latent space learned via unsupervised training. +4 +EXPERIMENTS IN SIMULATION +We first investigate the quality of SIRL-trained representations and +their benefits for preference learning using simulated human input +in two environments with ground truth rewards and features. +4.1 +Environments +GridRobot (Figure 2a) is a 5-by-5 gridworld with two obstacles +and a laptop (the blue, green, and black boxes). Trajectories are +sequences of 9 states with the start and end in opposite corners. The +19-dimensional input consists of the 𝑥 and 𝑦 coordinates of each +state and a discretized angle in {−90◦, −60◦, −30◦, 0◦, 30◦, 60◦, 90◦} +at the end state. The simulated human answers queries based on 4 +features 𝜙∗ in this world: Euclidean distances to each object, and +the absolute value of the angle orientation. +JacoRobot (Figure 2b) is a pybullet [24] simulated environment +with a 7-DoF Jaco robot arm on a tabletop, with a human and +laptop in the environment. Trajectories are length 21, and each +state consists of 97 dimensions: the 𝑥𝑦𝑧 positions of all robot joints +and objects, and their rotation matrices. This results in a 2037- +dimensional input space, much larger than for GridRobot. The +4 features of interest 𝜙∗ for the simulated human are: a) table — +(a) GridRobot environment. +Preprogrammed Views +Query Answers +Trajectory Replay +Query ID +𝜉! +𝜉" +𝜉# +(b) JacoRobot environment and user study interface. +Figure 2: Visualization of the experimental environments. +distance of the robot’s End-Effector (EE) to the table; b) upright +— EE orientation relative to upright, to consider whether objects +are carried upright; c) laptop — 𝑥𝑦-plane distance of the EE to a +laptop, to consider whether the EE passes over the laptop at any +height; d) proxemics [45] — proxemic 𝑥𝑦-plane distance of the EE to +the human, where the EE is considered closer to the human when +moving in front of the human that to their side. +In GridRobot the state space is discretized, so the trajectory space +Ξ can be enumerated; however, the JacoRobot state space is contin- +uous, so we construct Ξ by smoothly perturbing the shortest path +trajectories from 10,000 randomly sampled start-goal pairs (see App. +A.1). We generate similarity and preference queries by randomly +sampling from Ξ. The simulated human answers similarity queries +by computing the 4 feature values for each of the three trajecto- +ries and choosing the two that were closest in the feature space. +For preference queries, the simulated human computes the ground +truth reward and samples the trajectory with the higher reward. +The space of true reward functions (used to simulate preference +labels) is defined as linear combinations of the 4 features described +above. The robot is not given access to the ground-truth features +nor the ground-truth reward function but must learn them from +similarity and preference labels over raw trajectory observations. +4.2 +Qualitative Examples +In Figure 3 we show similar and dissimilar trajectories learned by +SIRL in a simplified GridRobot environment with only the laptop +and the joint angle. Top: the given trajectory stays far from the +laptop and holds the cup on its side; SIRL learns that trajectories +that share those features are similar, despite being dissimilar in + +30° +30° +-60° +:09 +-90° +90Params +Triplet query5 +Replay Red +ReplayGreen +Replay Blue +Redmore similar toGreen +Red more similar to Blue +Green more similar to Blue +Side View 1 +Side View 2 +Front View 1 +FrontView2 +Top ViewSIRL: Similarity-based Implicit Representation Learning +HRI ’23, March 13–16, 2023, Stockholm, Sweden +Figure 3: SIRL picks the two most and least similar trajectories to +a query trajectory. Top: trajectories are similar in features despite +being dissimilar in states. Bottom: trajectories are dissimilar in fea- +tures despite being close in states. +the state-space. Bottom: the trajectory stays close to the laptop and +holds the cup at an angle; SIRL learns that trajectories that hold +the cup upright and stay far from the laptop are dissimilar, despite +being similar in the state-space (going up and then right). +4.3 +Experimental Setup +Manipulated Variables. We test the importance of user input that +is designed to teach the representation by comparing SIRL with +multi-task learning techniques from generic preference queries, +and unsupervised representation learning. We have 4 baselines: a) +VAE, which learns a representation with a variational reconstruc- +tion loss [36]; b) MultiPref , a multi-task baseline [30], where we +learn the representation 𝜙 implicitly by training multiple reward +functions (each with shared initial layers) via preference learning; +c) SinglePref , a hypothetical method that learns from an ideal user +who weighs all features equally; d) Random, a randomly initial- +ized embedding, which does not benefit from human data but is +also immune from any spurious correlations that might be learned +from biased data. For MultiPref, we trained versions with 10 and +50 simulated human preference rewards for good coverage of the +reward space. All embeddings have the same network size: for +GridRobot we used MLPs with 2 layers, 128 units each, mapping +to 6 output neurons, while for JacoRobot we used 1024 units to +handle the larger input space (see App. A.2). For a fair compar- +ison, we gave SIRL, SinglePref, and MultiPref equal amounts of +human data for pre-training: 𝑁 similarity queries for SIRL, and +𝑁 preference queries (used for a single human for SinglePref or +equally distributed amongst humans for MultiPref). We also per- +formed ablations with and without VAE pre-training and found that +SinglePref and MultiPref are better without VAE (see App. A.3). +Dependent Measures. To test the quality of the learned repre- +sentations, we use two metrics: Feature Prediction Error (FPE) and +Test Preference Accuracy (TPA). The FPE metric is inspired by prior +work that argues that good representations are linearly separa- +ble [23, 38, 49]. Our goal is to measure whether the embeddings +contain the necessary information to recover the 4 ground-truth +features in each environment. We generate data sets of sampled +trajectories labeled with their ground truth (normalized) feature +vector D𝐹𝑃𝐸 = {𝜉,𝜙∗}. We freeze each embedding and add a linear +regression layer on top to predict the feature vector for a given +trajectory. We split D𝐹𝑃𝐸 into 80% training and 20% test pairs, and +FPE is the mean squared error (MSE) on the test set between the +predicted feature vector and the ground truth feature vector. For +the human query methods, we report FPE with increasing number +of representation training queries 𝑁. +For TPA, we test whether good representations necessarily lead +to good learning of general preferences. We use the trained em- +beddings as the base for 20 randomly selected test preference re- +wards. For each 𝑅𝜃𝑖 , we generate a set of labeled preference queries +D𝜃𝑖 +𝑝𝑟𝑒𝑓 = {𝜉𝐴, 𝜉𝐵,𝑙}, which we split into 80% for training and 20% +for test. We train each reward model with 𝑀 preference queries +per test reward, and we vary 𝑀. All preference networks have the +same architecture: we take the embedding 𝜙 pre-trained with the +respective method, and add new fully connected layers to learn a +reward function from trajectory preference labels. For GridRobot +we used MLPs with 2 layers of 128 units, and for JacoRobot we +used 1024 units. We found that all methods apart from SIRL worked +better with unfrozen embeddings (App. A.3). We report TPA as +the preference accuracy for the learned reward models on the test +preference set, averaged across the test human preferences. +Hypotheses. We test two hypotheses: +H1. Using similarity queries specifically designed to teach the +representation (SIRL) leads to representations more predictive of +the true features (lower FPE) than unsupervised (VAE), implicit +(MultiPref, SinglePref), or random representations. +H2. The SIRL representations result in more generalizable re- +ward learning (higher TPA) than unsupervised (VAE), implicit (Mul- +tiPref, SinglePref), or random representations. +4.4 +Results +In Figure 4 we show the FPE score for both environments with +varying representation queries 𝑁 from 100 to 1000. For GridRobot, +both versions of SIRL (with or without VAE pre-training) perform +similarly and outperform all baselines, supporting H1. When pre- +training with preference queries, MultiPref with 10 humans per- +forms better than SinglePref or MultiPref with 50 humans: Sin- +glePref may be overfitting to the one human preference it has seen, +while when MultiPref has to split its data budget among 50 humans +it ends up learning a worse representation than Random. There +is a balance to be struck between the diversity in human training +rewards covered and the amount of pre-training data each reward +gets, a trade-off which SIRL avoids because similarity queries are +agnostic to the particular human reward. For the more complex +JacoRobot, both versions of SIRL outperform all baselines, in line +with H1, although SIRL without VAE scores better than with it. +In Figure 5 we present the TPA score for both environments with +a varying amount of test preference queries 𝑀 from 10 to 190, and +𝑁 = 100, 500, and 1000. For GridRobot, each respective method +performs comparably with different 𝑁s, suggesting that this is a +simple enough environment that low amounts of representation +data are sufficient. For JacoRobot, this is not the case: with just +100 queries, SIRL with VAE pre-training performs like VAE, SIRL +without pre-training has random performance (since it’s frozen), +and the preference baselines all perform close to Random, as if they + +Trajectory +I Most Similar Trajectories : +: Least Similar Trajectories +006. . +90° +00 +0° +30° +-30° +30° +. · 90° +0°HRI ’23, March 13–16, 2023, Stockholm, Sweden +TBD +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +Number of Representation Queries +0.0 +0.1 +0.2 +0.3 +0.4 +FPE +GridRobot Feature Prediction Error (FPE) +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +Number of Representation Queries +0.0 +0.1 +0.2 +0.3 +0.4 +FPE +JacoRobot Feature Prediction Error (FPE) +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +Figure 4: FPE for the GridRobot (left) and JacoRobot (right) environments with simulated human data. With enough data, SIRL learns repre- +sentations more predictive of the true features 𝜙∗ in both simple and complex environments. +weren’t trained with queries at all. For larger 𝑁, both versions of +SIRL start performing better than the baselines, suggesting that +with enough data a good representation can be learned. +Focusing on 𝑁 = 1000, our results support H2: both SIRLs out- +perform all baselines in both environments, although for JacoRobot +SIRL without VAE is better than with VAE. In the GridRobot en- +vironment VAE pre-training helps SIRL. However, while VAE per- +forms comparably to other baselines in GridRobot, it severely un- +derperforms in JacoRobot. This suggests that the reconstruction +loss struggles to recover a helpful starting representation when +the input space is higher dimensional and correlated. As a result, +using the VAE pre-training to warmstart SIRL hinders performance +when compared to starting from a blank slate. When comparing the +preference-based baselines, in GridRobot they all perform similarly +apart from MultiPref with 50 humans. In JacoRobot we see a trend +that more preference humans does not necessarily result in better +performance. This confirms our observation from Figure 4 that +deciding on an appropriate number of human preferences to use for +multi-task pretraining is challenging, a problem that SIRL bypasses. +Summary. With enough representation data, SIRL learns represen- +tations more predictive of the true features (H1), leading to learning +more generalizable rewards (H2). This does not necessarily mean +that SIRL representations are perfectly aligned with causal features +— they are just better aligned, so the learned rewards are also better. +When VAE pre-training recovers sensible starting representations +it further reduces the amount of human data SIRL needs, otherwise +it hurts performance. Lastly, surprisingly, Random is often better +than pre-training with preference queries: preference-based meth- +ods may learn features that correlate with the training data but are +not necessarily causal, and an incorrectly biased representation is +worse for learning downstream rewards than starting from scratch. +5 +USER STUDY +We now present a user study with novice users that provide simi- +larity queries via an interface for the JacoRobot environment. +5.1 +Experiment Design +We ran a user study in the JacoRobot environment, modified for +only two features: table and laptop (we removed the humanoid in +the environment). We designed an interface where people can click +and drag to change the view, and press buttons to replay trajectories +and record their query answer (Figure 2b). We chose to display the +Euclidean path of each trajectory in the query traces, as we found +that to help users more easily compare trajectories to one another. +The study has two phases: collecting similarity queries and col- +lecting preference queries. In the first phase, we introduce the user +to the interface and describe the two features of interest. Because +similarity queries are preference-agnostic, we describe examples of +possible preferences akin to the ones in Sec. 1, but we do not bias the +participant towards any specific preference yet. Each participant +practices answering a set of pre-selected, unrecorded similarity +queries, and then answers 100 recorded similarity queries. In the +second phase, we describe a scenario in the environment that has +a specific preference associated with it (e.g. “There’s smoke in the +kitchen, so the robot should stay high from the table” or “There is +smoke in the kitchen and the robot’s mug is empty, so you want to +stay far from the table and close to the laptop.”) and assign different +preference scenarios to each participant. Each person practices un- +recorded preference queries, then answers 100 preference queries. +Participants. We recruited 10 users (3 female, 6 male, 1 non-binary, +aged 20-28) from the campus community to give queries. Most users +had technical background, so we caution that our results will speak +to SIRL’s usability with this population rather than more generally. +Manipulated Variables. Guided by the results in Figure 5, we +compare our best performing method, SIRL without VAE, to Ran- +dom, the best performing realistic baseline. For SIRL we collect +100 similarity queries from each participant and train a shared +representation using all of their data. +Dependent Measures. We present the same two metrics from Sec. +4, FPE and TPA. For TPA, we collect 100 preference queries for each +user’s unique preference, we use 70% for training individual reward +networks which we evaluate on the remaining 30% queries (Real). +We compute TPA with cross-validation on 50 splits. To demonstrate +how well SIRL works for new people who don’t contribute to learn- +ing the similarity embedding, we also train SIRL on the similarity +queries of 9 of the users and compute TPA on the held-out user’s +preference data (Held-out), for each user, respectively. Lastly, be- +cause real data tends to be noisy, we also compute TPA with 70 + +SIRL: Similarity-based Implicit Representation Learning +HRI ’23, March 13–16, 2023, Stockholm, Sweden +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot with N = 100 +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot with N = 100 +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot with N = 500 +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot with N = 500 +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot with N = 1000 +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot with N = 1000 +SIRL+VAE (Ours) +SinglePref +SIRL (Ours) +MultiPref 10H +Random +MultiPref 50H +VAE +Figure 5: TPA for GridRobot (top) and JacoRobot (bottom) with simulated human data. With enough data, SIRL recovers more generalizable +rewards than unsupervised, preference-trained, or random representations. +simulated preference queries for 10 different rewards, which we +also evaluate on a simulated test set (Simulated). +Hypotheses. Our hypotheses for the study are: +H3. Similarity queries (SIRL) recover more salient features than +a random representation (lower FPE), even with novice user data. +H4. The SIRL representation results in more generalizable re- +ward learning (higher TPA), even with novice similarity queries. +5.2 +Analysis +Figure 6 summarizes the results. On the left, SIRL recovers a repre- +sentation twice as predictive of the true features, supporting H3. +A 2-sided t-test (p < .0001) confirms this. This suggests SIRL can +recover aspects of people’s feature representation even with noisy +similarity queries from novice users. On the right (Real), SIRL recov- +ers more generalizable rewards on average than Random, providing +evidence for H4. Furthermore, using the SIRL representation on +a novel user (Held-out) also performs better than Random, and +the result appears almost indistinguishable from Real. This sug- +gests that similarity queries can be effectively crowd-sourced and +the resulting representation works well for novel user preferences. +Lastly, training with simulated preference queries slightly improves +performance for both methods, suggesting that noise in the human +preference data can be substantial. Three ANOVAs with method as +a factor find a significant main effect (F(1, 18) = 6.0175, p = .0246, +F(1, 18) = 4.7547, p = .0427, and F(1, 18) = 16.1068, p < .001, respec- +tively). For each of the 3 cases, we also separated the 6 humans that +were assigned preferences pertaining to both features (e.g. “There is +smoke in the kitchen and the robot’s mug is empty, so stay far from +the table and close to the laptop.”). SIRL performance is slightly +better when using preference data from this subset of users, hinting +that perhaps the learned representation entangled the two features. + +HRI ’23, March 13–16, 2023, Stockholm, Sweden +TBD +All Humans +0.00 +0.02 +0.04 +0.06 +0.08 +0.10 +FPE +SIRL (Ours) +Random +All Humans +6 Humans +All Humans +6 Humans +All Humans +6 Humans +0.0 +0.1 +0.2 +0.3 +0.4 +0.5 +0.6 +0.7 +0.8 +TPA +Real +Held-out +Simulated +Figure 6: Study values for FPE, and TPA with real and simulated preferences. Even with novice similarity queries, SIRL recovers representations +both more predictive of the true features and more useful for learning different user rewards than the baseline. +Overall, the quantitative results support H3 and H4, providing +evidence that SIRL can recover more human-aligned representa- +tions. Subjectively, some users found the 2D interface deceiving at +times, as they would judge trajectory similarity differently based on +the viewpoint. This is a natural artifact of visualizing a 3D world in +2D, but future work should investigate better interfaces. Some users +reported struggling to trade off the two features when comparing +trajectories. This is in part due to the fact that we almost “engi- +neered” their internal representation, so a more in-the-wild study +could determine whether similarity queries are indeed preference- +agnostic. Lastly, some queries were easier than others: users’ time- +to-answer varied across triplets suggesting that future work could +use it as a confidence metric for how much to trust their answer. +6 +DISCUSSION AND LIMITATIONS +In this work, our goal was to tackle the problem of learning good +representations that capture (and disentangle) the features that +matter, while excluding spurious features. If we had such a repre- +sentation, then learning rewards that capture different preferences +and tasks on top of it would lead to generalizable models that reli- +ably incentivize the right behavior across different situations, rather +than picking up on correlates and being unable to distinguish good +from bad behaviors on new data. Our idea was to implicitly tap +into this representation by asking people what they find similar +versus not, because two behaviors will be similar if and only if +their representations are similar. We introduced a novel human +input type, trajectory similarity queries, and tested that it leads to +better representations than those learned through self-supervision +or via multi-task learning: it enables learning rewards from the +same training data that better rank behaviors on test data. +That said, we need to be explicit that this is not the be-all end-all +solution to our goal above. The representations learned, as we see +in simulation, are not perfectly aligned with causal features — they +are just better aligned. The learned rewards are not perfect — they +are just better than alternatives. Similarity queries do not solve the +problem fully, potentially because they suffer from the same issue +preference queries do: when multiple important features change, it +becomes harder to make a judgement call on what is more similar. +The advantage that we see from similarity queries, though, is that +rewards for particular tasks might ignore or down-weigh certain +features that matter in other tasks, while similarity queries are +task-agnostic and implicitly capture the distribution of tasks in the +human’s head. Rather than having to specify a task distribution for +multi-task learning, with similarity queries we are (implicitly) ask- +ing the user to leverage their more general-purpose representation +of the world. But thinking about how to overcome the challenge +that multiple changing features make these queries harder to an- +swer opens the door to exciting ideas for future work. For instance, +what if we iteratively built the space, and based similarity queries +on some current estimate of what are the important features; over +time, as the representation becomes more aligned with the human’s, +the queries would get better at honing in on specific features. +Another obvious limitation is that we did not do an in-the-wild +study. In theory, similarity queries should be used when people +already have a robot they are familiar with and, thus, have a distri- +bution of tasks they care about in their everyday contexts, but in +our study we needed to explain to users these contexts and what +might be important. In doing so, we almost “engineered” their inter- +nal representation. As robots become more prevalent, a follow-up +study where users are given much less structure and allowed to +actually tap into their unaltered representation might be possible. +In a sense, with SIRL we build a foundation model, and this some- +times requires hundreds of queries to learn a good representation. +While we don’t think having this much data when pre-training +is unreasonable, especially since it leads to significant desirable +performance improvement over baselines, sample-complexity is +crucial to address as we scale to more complex robotic tasks. Be- +cause similarity queries are task agnostic, we can crowd-source +the queries from multiple people (as we did in the user study) and +rely on this economy at scale to alleviate user burden. Future work +could also look at active querying methodologies to ask the person +for more informative similarity queries and reduce data amounts. +A further avenue of work is extending this beyond reward learn- +ing, using SIRL representations directly for policy learning or learn- +ing exploration functions. We also emphasize that similarity queries +are not a replacement for self-supervised learning; instead, we view +them as complementary — self-supervised learning might be able to +leverage expert-designed heuristics to eliminate many of the spuri- +ous features, while SIRL might serve to fine-tune the representation. +How to properly combined the two remains an open question. +Overall, similarity queries are a step towards recovering human- +aligned representations. They improve upon the state of the art, and +can benefit from further exploration in how to combine them with +other inputs and self-supervision, and how to make them easier +through better interfaces and query selection algorithms. + +SIRL: Similarity-based Implicit Representation Learning +HRI ’23, March 13–16, 2023, Stockholm, Sweden +REFERENCES +[1] Pieter Abbeel and Andrew Y Ng. 2004. Apprenticeship learning via inverse +reinforcement learning. In Machine Learning (ICML), International Conference on. +ACM. +[2] Sameer Agarwal, Josh Wills, Lawrence Cayton, Gert Lanckriet, David Kriegman, +and Serge Belongie. 2007. Generalized non-metric multidimensional scaling. In +Artificial Intelligence and Statistics. PMLR, 11–18. +[3] Ehsan Amid, Aristides Gionis, and Antti Ukkonen. 2015. A Kernel-Learning +Approach to Semi-supervised Clustering with Relative Distance Comparisons, +Vol. 9284. https://doi.org/10.1007/978-3-319-23528-8_14 +[4] Yusuf Aytar, Tobias Pfaff, David Budden, Tom Le Paine, Ziyu Wang, and Nando de +Freitas. 2018. Playing Hard Exploration Games by Watching YouTube. In Pro- +ceedings of the 32nd International Conference on Neural Information Processing +Systems (Montréal, Canada) (NIPS’18). Curran Associates Inc., Red Hook, NY, +USA, 2935–2945. +[5] Monica Babes, Vukosi N Marivate, Kaushik Subramanian, and Michael L Littman. +2011. Apprenticeship learning about multiple intentions. In ICML. +[6] Andrea Bajcsy, Dylan P. Losey, Marcia K. O’Malley, and Anca D. Dragan. 2017. +Learning Robot Objectives from Physical Human Interaction. In Proceedings of +the 1st Annual Conference on Robot Learning (Proceedings of Machine Learning +Research, Vol. 78), Sergey Levine, Vincent Vanhoucke, and Ken Goldberg (Eds.). +PMLR, 217–226. http://proceedings.mlr.press/v78/bajcsy17a.html +[7] Vassileios Balntas, Edgar Riba, Daniel Ponsa, and Krystian Mikolajczyk. 2016. +Learning local feature descriptors with triplets and shallow convolutional neural +networks. 119.1–119.11. https://doi.org/10.5244/C.30.119 +[8] Erdem Biyik and Dorsa Sadigh. 2018. Batch active preference-based learning of +reward functions. In Conference on robot learning. PMLR, 519–528. +[9] Andreea Bobu, Chris Paxton, Wei Yang, Balakumar Sundaralingam, Yu-Wei +Chao, Maya Cakmak, and Dieter Fox. 2022. Learning Perceptual Concepts by +Bootstrapping From Human Queries. IEEE Robotics Autom. Lett. 7, 4 (2022), +11260–11267. https://doi.org/10.1109/LRA.2022.3196164 +[10] Andreea Bobu, Marius Wiggert, Claire Tomlin, and Anca D. Dragan. 0. Inducing +Structure in Reward Learning by Learning Features. The International Jour- +nal of Robotics Research 0, 0 (0), 02783649221078031. +https://doi.org/10.1177/ +02783649221078031 arXiv:https://doi.org/10.1177/02783649221078031 +[11] Andreea Bobu, Marius Wiggert, Claire Tomlin, and Anca D. Dragan. 2021. Feature +Expansive Reward Learning: Rethinking Human Input. In Proceedings of the 2021 +ACM/IEEE International Conference on Human-Robot Interaction (Boulder, CO, +USA) (HRI ’21). Association for Computing Machinery, New York, NY, USA, +216–224. https://doi.org/10.1145/3434073.3444667 +[12] Ralph Allan Bradley and Milton E Terry. 1952. Rank analysis of incomplete block +designs: I. The method of paired comparisons. Biometrika 39, 3/4 (1952), 324–345. +[13] Daniel Brown, Russell Coleman, Ravi Srinivasan, and Scott Niekum. 2020. Safe +Imitation Learning via Fast Bayesian Reward Inference from Preferences. In +Proceedings of the 37th International Conference on Machine Learning (Proceedings +of Machine Learning Research, Vol. 119), Hal Daumé III and Aarti Singh (Eds.). +PMLR, 1165–1177. http://proceedings.mlr.press/v119/brown20a.html +[14] Daniel Brown, Wonjoon Goo, Prabhat Nagarajan, and Scott Niekum. 2019. Extrap- +olating beyond suboptimal demonstrations via inverse reinforcement learning +from observations. In International Conference on Machine Learning. PMLR, 783– +792. +[15] Kalesha Bullard, Sonia Chernova, and Andrea Lockerd Thomaz. 2018. Human- +Driven Feature Selection for a Robotic Agent Learning Classification Tasks from +Demonstration. In 2018 IEEE International Conference on Robotics and Automation, +ICRA 2018, Brisbane, Australia, May 21-25, 2018. IEEE, 6923–6930. https://doi. +org/10.1109/ICRA.2018.8461012 +[16] Maya Cakmak and Andrea Lockerd Thomaz. 2012. Designing robot learners +that ask good questions. In International Conference on Human-Robot Interaction, +HRI’12, Boston, MA, USA - March 05 - 08, 2012, Holly A. Yanco, Aaron Steinfeld, +Vanessa Evers, and Odest Chadwicke Jenkins (Eds.). ACM, 17–24. https://doi. +org/10.1145/2157689.2157693 +[17] Annie S. Chen, Suraj Nair, and Chelsea Finn. 2021. Learning Generalizable Robotic +Reward Functions from "In-The-Wild" Human Videos. CoRR abs/2103.16817 +(2021). arXiv:2103.16817 https://arxiv.org/abs/2103.16817 +[18] Ricky T. Q. Chen, Xuechen Li, Roger B Grosse, and David K Duvenaud. 2018. +Isolating Sources of Disentanglement in Variational Autoencoders. In Advances +in Neural Information Processing Systems, S. Bengio, H. Wallach, H. Larochelle, +K. Grauman, N. Cesa-Bianchi, and R. Garnett (Eds.), Vol. 31. Curran Associates, +Inc. +[19] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A +simple framework for contrastive learning of visual representations. In Interna- +tional conference on machine learning. PMLR, 1597–1607. +[20] Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter +Abbeel. 2016. InfoGAN: Interpretable Representation Learning by Information +Maximizing Generative Adversarial Nets. In Proceedings of the 30th International +Conference on Neural Information Processing Systems (Barcelona, Spain) (NIPS’16). +Curran Associates Inc., Red Hook, NY, USA, 2180–2188. +[21] Jaedeug Choi and Kee-Eung Kim. 2012. Nonparametric Bayesian inverse rein- +forcement learning for multiple reward functions. Advances in Neural Information +Processing Systems 25 (2012). +[22] Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario +Amodei. 2017. +Deep Reinforcement Learning from Human Preferences. In +Advances in Neural Information Processing Systems, I. Guyon, U. V. Luxburg, +S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.), +Vol. 30. Curran Associates, Inc. https://proceedings.neurips.cc/paper/2017/file/ +d5e2c0adad503c91f91df240d0cd4e49-Paper.pdf +[23] Adam Coates and A. Ng. 2012. Learning Feature Representations with K-Means. +In Neural Networks: Tricks of the Trade. +[24] Erwin Coumans and Yunfei Bai. 2016–2019. PyBullet, a Python module for physics +simulation for games, robotics and machine learning. http://pybullet.org. +[25] Cagatay Demiralp, Michael Bernstein, and Jeffrey Heer. 2014. Learning Perceptual +Kernels for Visualization Design. IEEE Transactions on Visualization and Computer +Graphics 20. https://doi.org/10.1109/TVCG.2014.2346978 +[26] Christos Dimitrakakis and Constantin A Rothkopf. 2011. Bayesian multitask +inverse reinforcement learning. In European workshop on reinforcement learning. +Springer, 273–284. +[27] Carl Doersch, Abhinav Kumar Gupta, and Alexei A. Efros. 2015. Unsupervised +Visual Representation Learning by Context Prediction. 2015 IEEE International +Conference on Computer Vision (ICCV) (2015), 1422–1430. +[28] A. D. Dragan, K. Muelling, J. Andrew Bagnell, and S. S. Srinivasa. 2015. Movement +primitives via optimization. In 2015 IEEE International Conference on Robotics and +Automation (ICRA). 2339–2346. https://doi.org/10.1109/ICRA.2015.7139510 +[29] Chelsea Finn, Pieter Abbeel, and Sergey Levine. 2017. Model-Agnostic Meta- +Learning for Fast Adaptation of Deep Networks. In Proceedings of the 34th Inter- +national Conference on Machine Learning - Volume 70 (Sydney, NSW, Australia) +(ICML’17). JMLR.org, 1126–1135. +[30] Adam Gleave and Oliver Habryka. 2018. Multi-task maximum entropy inverse +reinforcement learning. arXiv preprint arXiv:1805.08882 (2018). +[31] Bradley Hayes and Brian Scassellati. 2014. Discovering task constraints through +observation and active learning. In 2014 IEEE/RSJ International Conference on +Intelligent Robots and Systems, Chicago, IL, USA, September 14-18, 2014. IEEE, +4442–4449. https://doi.org/10.1109/IROS.2014.6943191 +[32] Irina Higgins, Loïc Matthey, Arka Pal, Christopher P. Burgess, Xavier Glorot, +Matthew M. Botvinick, Shakir Mohamed, and Alexander Lerchner. 2017. beta- +VAE: Learning Basic Visual Concepts with a Constrained Variational Framework. +In ICLR. +[33] Chao Huang, Wenhao Luo, and Rui Liu. 2021. Meta Preference Learning for +Fast User Adaptation in Human-Supervisory Multi-Robot Deployments. In 2021 +IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, +5851–5856. +[34] Hong Jun Jeon, Smitha Milli, and Anca Dragan. 2020. Reward-rational (implicit) +choice: A unifying formalism for reward learning. Advances in Neural Information +Processing Systems 33 (2020), 4415–4426. +[35] Maurice George Kendall. 1948. Rank correlation methods. (1948). +[36] Diederik P. Kingma and Max Welling. 2014. Auto-Encoding Variational Bayes. In +2nd International Conference on Learning Representations, ICLR 2014, Banff, AB, +Canada, April 14-16, 2014, Conference Track Proceedings, Yoshua Bengio and Yann +LeCun (Eds.). http://arxiv.org/abs/1312.6114 +[37] Johannes Kulick, Marc Toussaint, Tobias Lang, and Manuel Lopes. 2013. Active +Learning for Teaching a Robot Grounded Relational Symbols. In IJCAI 2013, +Proceedings of the 23rd International Joint Conference on Artificial Intelligence, +Beijing, China, August 3-9, 2013, Francesca Rossi (Ed.). IJCAI/AAAI, 1451–1457. +http://www.aaai.org/ocs/index.php/IJCAI/IJCAI13/paper/view/6706 +[38] Cheng-I Lai. 2019. Contrastive Predictive Coding Based Feature for Automatic +Speaker Verification. arXiv preprint arXiv:1904.01575 (2019). +[39] Michael Laskin, Aravind Srinivas, and Pieter Abbeel. 2020. CURL: Contrastive +Unsupervised Representations for Reinforcement Learning. In Proceedings of +the 37th International Conference on Machine Learning (Proceedings of Machine +Learning Research, Vol. 119), Hal Daumé III and Aarti Singh (Eds.). PMLR, 5639– +5650. https://proceedings.mlr.press/v119/laskin20a.html +[40] Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. 2020. Offline rein- +forcement learning: Tutorial, review, and perspectives on open problems. arXiv +preprint arXiv:2005.01643 (2020). +[41] Kejun Li, Maegan Tucker, Erdem Bıyık, Ellen Novoseller, Joel W Burdick, Yanan +Sui, Dorsa Sadigh, Yisong Yue, and Aaron D Ames. 2021. Roial: Region of interest +active learning for characterizing exoskeleton gait preference landscapes. In +2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, +3212–3218. +[42] Hoai Luu-Duc and Jun Miura. 2019. An Incremental Feature Set Refinement in a +Programming by Demonstration Scenario. In 4th IEEE International Conference +on Advanced Robotics and Mechatronics, ICARM 2019, Toyonaka, Japan, July 3-5, +2019. IEEE, 372–377. https://doi.org/10.1109/ICARM.2019.8833723 +[43] Zhao Mandi, Pieter Abbeel, and Stephen James. 2022. On the Effectiveness of +Fine-tuning Versus Meta-reinforcement Learning. arXiv preprint arXiv:2206.03271 +(2022). + +HRI ’23, March 13–16, 2023, Stockholm, Sweden +TBD +[44] Brian McFee, Gert Lanckriet, and Tony Jebara. 2011. Learning Multi-modal +Similarity. Journal of machine learning research 12, 2 (2011). +[45] Jonathan Mumm and Bilge Mutlu. 2011. Human-robot proxemics: physical and +psychological distancing in human-robot interaction. In Proceedings of the 6th +international conference on Human-robot interaction. 331–338. +[46] Kentaro Nishi and Masamichi Shimosaka. 2020. Fine-grained driving behavior +prediction via context-aware multi-task inverse reinforcement learning. In 2020 +IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2281– +2287. +[47] Takayuki Osa, Joni Pajarinen, Gerhard Neumann, J Andrew Bagnell, Pieter Abbeel, +Jan Peters, et al. 2018. An Algorithmic Perspective on Imitation Learning. Foun- +dations and Trends in Robotics 7, 1-2 (2018), 1–179. +[48] Deepak Pathak, Parsa Mahmoudieh, Guanghao Luo, Pulkit Agrawal, Dian Chen, +Fred Shentu, Evan Shelhamer, Jitendra Malik, Alexei A. Efros, and Trevor Darrell. +2018. Zero-Shot Visual Imitation. In 2018 IEEE/CVF Conference on Computer +Vision and Pattern Recognition Workshops (CVPRW). 2131–21313. https://doi.org/ +10.1109/CVPRW.2018.00278 +[49] C. J. Reed, X. Yue, A. Nrusimha, S. Ebrahimi, V. Vijaykumar, R. Mao, B. Li, S. +Zhang, D. Guillory, S. Metzger, K. Keutzer, and T. Darrell. 2022. Self-Supervised +Pretraining Improves Self-Supervised Pretraining. In 2022 IEEE/CVF Winter Con- +ference on Applications of Computer Vision (WACV). IEEE Computer Society, Los +Alamitos, CA, USA, 1050–1060. https://doi.org/10.1109/WACV51458.2022.00112 +[50] Dorsa Sadigh, Anca D Dragan, Shankar Sastry, and Sanjit A Seshia. 2017. Active +preference-based learning of reward functions. In Robotics: Science and systems. +[51] Seyed Kamyar Seyed Ghasemipour, Shixiang Shane Gu, and Richard Zemel. 2019. +Smile: Scalable meta inverse reinforcement learning through context-conditional +policies. Advances in Neural Information Processing Systems 32 (2019). +[52] Arjun Sripathy, Andreea Bobu, Zhongyu Li, Koushil Sreenath, Daniel S. Brown, +and Anca D. Dragan. 2022. Teaching Robots to Span the Space of Functional +Expressive Motion. https://doi.org/10.48550/ARXIV.2203.02091 +[53] Neil Stewart, Gordon DA Brown, and Nick Chater. 2005. Absolute identification +by relative judgment. Psychological review 112, 4 (2005), 881. +[54] Omer Tamuz, Ce Liu, Serge Belongie, Ohad Shamir, and Adam Tauman Kalai. +2011. Adaptively learning the crowd kernel. arXiv preprint arXiv:1105.1033 +(2011). +[55] Christian Wirth, Riad Akrour, Gerhard Neumann, Johannes Fürnkranz, et al. +2017. A survey of preference-based reinforcement learning methods. Journal of +Machine Learning Research 18, 136 (2017), 1–46. +[56] Kelvin Xu, Ellis Ratner, Anca Dragan, Sergey Levine, and Chelsea Finn. 2019. +Learning a Prior over Intent via Meta-Inverse Reinforcement Learning. In Pro- +ceedings of the 36th International Conference on Machine Learning (Proceedings of +Machine Learning Research, Vol. 97), Kamalika Chaudhuri and Ruslan Salakhutdi- +nov (Eds.). PMLR, 6952–6962. https://proceedings.mlr.press/v97/xu19d.html +[57] Kelvin Xu, Ellis Ratner, Anca Dragan, Sergey Levine, and Chelsea Finn. 2019. +Learning a prior over intent via meta-inverse reinforcement learning. In Interna- +tional Conference on Machine Learning. PMLR, 6952–6962. +[58] Lantao Yu, Tianhe Yu, Chelsea Finn, and Stefano Ermon. 2019. Meta-inverse +reinforcement learning with probabilistic context variables. Advances in Neural +Information Processing Systems 32 (2019). + +SIRL: Similarity-based Implicit Representation Learning +HRI ’23, March 13–16, 2023, Stockholm, Sweden +A +APPENDIX +A.1 +Trajectory generation +In GridRobot the state space is discretized, so the trajectory space Ξ +can be enumerated; however, the JacoRobot state space is continu- +ous, so we need to construct Ξ by sampling the infinite-dimensional +trajectory space. We randomly sample 10,000 start-goal pairs and +compute the shortest path in the robot’s configuration space for +each of them, 𝜉𝑆𝐺. Each trajectory has a horizon length 𝐻 and +consists of 𝑛-dimensional states. We then apply random torque de- +formations 𝑢 to each trajectory to obtain a deformed trajectory 𝜉𝑆𝐺 +𝐷 . +In particular, we randomly select up to 3 states along the trajectory, +and then deform each of the selected states with a different random +torque 𝑢. To deform a trajectory in the direction of 𝑢 we follow: +𝜉𝑆𝐺 +𝐷 += 𝜉𝑆𝐺 + 𝜇𝐴−1 ˜𝑢 , +(5) +where 𝐴 ∈ R𝑛(𝐻+1)×𝑛(𝐻+1) defines a norm on the Hilbert space of +trajectories and dictates the deformation shape [28], 𝜇 > 0 scales +the magnitude of the deformation, and ˜𝑢 ∈ R𝑛(𝐻+1) is 𝑢 at indices +𝑛𝑡 through 𝑛(𝑡 + 1) and 0 otherwise ( ˜𝑢 is 0 outside of the chosen +deformation state index). For each deformation, we randomly gen- +erated 𝜇 and the index of the state the deformation is applied to. +For smooth deformations, we used a norm 𝐴 based on acceleration, +but other norm choices are possible as well (see Dragan et al. [28] +for more details). We took inspiration for this deformation strategy +from Bajcsy et al. [6]. +A.2 +Training details +We present architecture and optimization details that can assist in +reproducing our training setup. +A.2.1 +Feature networks. All embeddings have the same network +size: for GridRobot we used MLPs with 2 hidden layers, 128 units +each, mapping to 6 output neurons, while for JacoRobot we used +1024 units to handle the larger input space. For both environments, +we used ReLU non-linearities after every linear layer. +We trained the VAE network with a standard variational recon- +struction loss [36] also including a KL-divergence-based regular- +ization term (to make the latent space regular). The regularization +part of the loss had a weight of 𝜆 = 0.01. For both environments, +we optimized the loss function using Adam for 2000 epochs with +an exponentially decaying learning rate of 0.01 (decay rate 0.99999) +and a batch-size of 32. +SinglePref and MultiPref with 10 and 50 humans are trained using +the standard preference loss in Eq. (4). Christiano et al. [22] ensured +that the rewards predicted by the preference network remain small +by normalizing them on the fly. We instead add an 𝑙2 regulatization +term on the predicted reward to the preference loss, with a weight +of 10 for GridRobot and 1 for JacoRobot. All three methods optimize +this final loss in the same way: for GridRobot, we use Adam for +5000 epochs with a learning rate of 0.01 and batch-size 32, while +for JacoRobot we found a lower learning rate of 0.001 to result in +more stable training. +Lastly, for SIRL we had the option to first pre-train with the above +VAE loss. Training with the similarity objective in Eq. (2) happens +disjointly, after pre-training. For both GridRobot and JacoRobot, we +optimized this loss function using Adam for 3000 epochs with an +exponentially decaying learning rate of 0.004 (decay rate 0.99999) +and batch-size 64. +We note that our current architectures assume fixed-length tra- +jectories but one could adopt an LSTM-based architecture for tra- +jectories of varying length [52]. +A.2.2 +Preference networks. For evaluating TPA, we used preference +networks on top of the embeddings for the respective methods we +evaluate. For GridRobot we used MLPs with 2 hidden layers of 128 +units, and for JacoRobot we used 1024 units for larger capacity. For +both environments, we used ReLU non-linearities after every linear +layer. We added the same 𝑙2 regularization to the loss in Eq. (4) +as before, with weight 10 for GridRobot and 1 for JacoRobot. For +GridRobot, we optimized our final loss function using Adam for +500 epochs with a learning rate of 0.001 and a batch-size of 64. For +JacoRobot, we increased the number of epochs to 1000. +A.3 +Ablations +Figure 5 illustrates results with frozen SIRL, and unfrozen baselines +without VAE pre-training, as these were the best configurations we +found for each method. In this section, we show the complete abla- +tion we performed to decide which methods benefit from frozen or +unfrozen embeddings, or VAE pre-training. Figure 7 showcases the +result of this ablation on both GridRobot and JacoRobot. Overall, +we see that SIRL does better when the learned representation is +frozen, while all the other methods do better when the representa- +tions is unfrozen. SinglePref and the MultiPref baselines perform +better without VAE pre-training, while SIRL sometimes benefits +from pre-training in simple environments like GridRobot. + +HRI ’23, March 13–16, 2023, Stockholm, Sweden +TBD +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot Test Preference Accuracy (TPA) with N = 1000 +SIRL (frozen) +SIRL (unfrozen) +SIRL+VAE (frozen) +SIRL+VAE (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot Test Preference Accuracy (TPA) with N = 1000 +SIRL (frozen) +SIRL (unfrozen) +SIRL+VAE (frozen) +SIRL+VAE (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot Test Preference Accuracy (TPA) with N = 1000 +SinglePref (frozen) +SinglePref (unfrozen) +SinglePref+VAE (frozen) +SinglePref+VAE (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot Test Preference Accuracy (TPA) with N = 1000 +SinglePref (frozen) +SinglePref (unfrozen) +SinglePref+VAE (frozen) +SinglePref+VAE (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot Test Preference Accuracy (TPA) with N = 1000 +MultiPref 10H (frozen) +MultiPref 10H (unfrozen) +MultiPref+VAE 10H (frozen) +MultiPref+VAE 10H (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot Test Preference Accuracy (TPA) with N = 1000 +MultiPref 10H (frozen) +MultiPref 10H (unfrozen) +MultiPref+VAE 10H (frozen) +MultiPref+VAE 10H (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot Test Preference Accuracy (TPA) with N = 1000 +MultiPref 50H (frozen) +MultiPref 50H (unfrozen) +MultiPref+VAE 50H (frozen) +MultiPref+VAE 50H (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot Test Preference Accuracy (TPA) with N = 1000 +MultiPref 50H (frozen) +MultiPref 50H (unfrozen) +MultiPref+VAE 50H (frozen) +MultiPref+VAE 50H (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +GridRobot Test Preference Accuracy (TPA) with N = 1000 +Random (frozen) +Random (unfrozen) +VAE (frozen) +VAE (unfrozen) +10 +30 +50 +70 +90 +110 +130 +150 +170 +190 +Number of Preference Queries +0.5 +0.6 +0.7 +0.8 +0.9 +1.0 +TPA +JacoRobot Test Preference Accuracy (TPA) with N = 1000 +Random (frozen) +Random (unfrozen) +VAE (frozen) +VAE (unfrozen) +Figure 7: Ablation results for GridRobot (left) and JacoRobot (right). Overall, SIRL does better when the learned representation is frozen, +while all other method do better when the representations is unfrozen. SinglePref and the MultiPref baselines perform better without VAE +pre-training, while SIRL sometimes benefits from pre-training in simple environments like GridRobot. + diff --git a/INAyT4oBgHgl3EQf5vp6/content/tmp_files/load_file.txt b/INAyT4oBgHgl3EQf5vp6/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..982de3b8ea92ad6d8b6cf160b5383ac1df2a4773 --- /dev/null +++ b/INAyT4oBgHgl3EQf5vp6/content/tmp_files/load_file.txt @@ -0,0 +1,951 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf,len=950 +page_content='SIRL: Similarity-based Implicit Representation Learning Andreea Bobu∗ abobu@berkeley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='edu University of California, Berkeley United States of America Yi Liu∗ yiliu77@berkeley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='edu University of California, Berkeley United States of America Rohin Shah rohinmshah@deepmind.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='com DeepMind Research United Kingdom Daniel S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Brown dsbrown@berkeley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='edu University of California, Berkeley United States of America Anca D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan anca@berkeley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='edu University of California, Berkeley United States of America ɸ Similarity query Preference query H1 Preference query H2 Preference query H3 Optimize 𝑅!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Optimize 𝑅!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='" Optimize 𝑅!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='# 𝑅!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 𝑅!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='" 𝑅!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='# Figure 1: Our goal is to learn representations for robot behavior that capture what is salient to people, and, thus, support generalizable pref- erence learning with low sample complexity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We propose to extract this representation by asking people trajectory similarity queries (left), where they judge which two out of three trajectories are most similar to each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We then use the representation to learn reward functions corresponding to different people’s preferences on different tasks (right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ABSTRACT When robots learn reward functions using high capacity models that take raw state directly as input, they need to both learn a representation for what matters in the task — the task “features" — as well as how to combine these features into a single objective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' If they try to do both at once from input designed to teach the full reward function, it is easy to end up with a representation that contains spurious correlations in the data, which fails to generalize to new settings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Instead, our ultimate goal is to enable robots to identify and isolate the causal features that people actually care about and use when they represent states and behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our idea is that we can tune into this representation by asking users what behaviors they consider similar: behaviors will be similar if the features that matter are similar, even if low-level behavior is dif- ferent;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' conversely, behaviors will be different if even one of the ∗Both authors contributed equally to this research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Copyrights for third-party components of this work must be honored.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For all other uses, contact the owner/author(s).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' HRI ’23, March 13–16, 2023, Stockholm, Sweden © 2023 Copyright held by the owner/author(s).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ACM ISBN 978-1-4503-9964-7/23/03.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1145/3568162.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3576989 features that matter differs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This, in turn, is what enables the robot to disambiguate between what needs to go into the representation versus what is spurious, as well as what aspects of behavior can be compressed together versus not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The notion of learning representa- tions based on similarity has a nice parallel in contrastive learning, a self-supervised representation learning technique that maps vi- sually similar data points to similar embeddings, where similarity is defined by a designer through data augmentation heuristics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' By contrast, in order to learn the representations that people use, so we can learn their preferences and objectives, we use their definition of similarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In simulation as well as in a user study, we show that learning through such similarity queries leads to representations that, while far from perfect, are indeed more generalizable than self-supervised and task-input alternatives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' KEYWORDS robot reward learning, human-aligned representation learning ACM Reference Format: Andreea Bobu, Yi Liu, Rohin Shah, Daniel S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Brown, and Anca D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2023.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SIRL: Similarity-based Implicit Representation Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 2023 ACM/IEEE International Conference on Human-Robot Interaction (HRI ’23), March 13–16, 2023, Stockholm, Sweden.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ACM, New York, NY, USA, 12 pages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1145/3568162.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3576989 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='00810v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='RO] 2 Jan 2023 GNParams Preference query1 ReplayRed Replay Blue Red better than Blue Blue better than Red Side View 1 Side View 2 FrontView1 Front View 2 Top ViewHRI ’23, March 13–16, 2023, Stockholm, Sweden TBD 1 INTRODUCTION Imagine waking up in the morning and your home robot assistant wants to place a steaming mug of fresh coffee on the table exactly where it knows you will sit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Depending on the context, you will have a different preference for how the robot should be doing its task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Some days it carries your favorite mug close to the table to prevent it from breaking in the case of a slip (so that it will remain your favorite mug);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' other days the steam from your delicious meal is difficult to handle for the robot’s perception, so you’d want it to keep a large clearance from the table to avoid collisions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Similarly, some days you want the robot to keep your mug away from your laptop to avoid spilling on it;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' other days the mug only has an espresso shot so you want the robot keep it close to the laptop to prevent clutter and leave the rest of the table open for you.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The reward function the robot should optimize changes — whether due to variations in the task, having different users, or, as in the examples above, different contexts that are not always part of the robot state (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' holding the user’s favorite mug and not just a regu- lar mug).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' However, the representation on top of which the reward is built, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='e the features that are important (like the distance from the table, being above the laptop, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ), are shared.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' If the robot learns this representation correctly, it can use it to obtain the right reward function, even if the task, user, and context changes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Meta-learning and multi-task learning methods [30, 46, 56] learn the representa- tion from user input meant to teach the full reward, like preference queries or demonstrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' By contrast, we propose that if learning generalizable representations is the goal, then we should ask the user for input that is specifically meant to teach the representation itself, rather than input meant to teach the full reward and hoping to extract a good representation along the way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' But asking people to teach robots representations, rather than tasks, is not so easy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' What are the features that they care about?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' While some techniques advocate for people enabling users to teach each feature separately [11], people may not always be able to explicate their representation and break it down into concepts that are individually teachable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In this work, our idea is that we can implicitly tune into the representations people use by asking them to do a proxy task of evaluating similarity of behaviors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Behaviors will be similar if the features that matter are similar, even if low-level behavior is different;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' conversely, behaviors will be different if even one of the features that matter differs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This, in turn, should enable the robot to arrive at the features that matter — we want robots that can disambiguate between what needs to go into the representation versus what is spurious, as well as what aspects of behavior can be compressed together into a feature embedding versus kept separate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We thus introduce a novel type of human input to help the robot extract the person’s representation: trajectory similarity queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A trajectory similarity query is a triplet of trajectories that the person answers by picking the two more similar trajectories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Figure 1 (left), the person chooses the two trajectories that are close to the table and far from the laptop, even though visually they look dissimilar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This results in an (anchor, positive, negative) triplet that can be used for training a feature representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We call this process Similarity-based Implicit Representation Learning (SIRL).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our method has a parallel in self-supervised learning work, espe- cially contrastive learning, where the goal is to learn a good visual representation by training from (anchor, positive, negative) triplets generated via data augmentation techniques [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' However, this notion of similarity is purely visual, driven by manually designed heuristics for data augmentation, and is not necessarily reflective of what users would consider similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For instance, two images might be labeled as visually different, when in fact their difference is only with respect to some low-level aspects that are not really relevant to the distribution of tasks people care about.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This would result in representations that contain too many distractor features that are not present in the human’s representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our method uses similarity too, but we defer to the user’s judgement of similarity, with the goal of reconstructing the user’s representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Of course, our method is not the full answer to learning causally aligned representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' But our experiments suggest that it out- performs methods that are self-supervised, or that learn from input meant to teach the full tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In simulation, where we know the causal features, we show that SIRL learns representations better aligned with them, which in turn leads to learning multiple more generalizable reward functions downstream (Figure 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We also present a user study where we crowdsource similarity queries from different people to learn a shared SIRL representation that better re- covers each of their individual preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' While the study results do show a significant effect, the effect size is much lower than in simulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This is attributable in part to the interface difficulty of analyzing the robot trajectories, which means more work is needed to determine the best interfaces that enable users to accurately answer similarity queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Moreover, some users reported strug- gling to trade off the different features, which means that similarity queries might not be entirely preference-agnostic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Nonetheless, our results underscore that there are gains by explicitly aligning robot and human representations, rather than hoping it will happen as a byproduct of learning rewards from standard queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2 RELATED WORK Learning from Human Input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Human-in-the-loop learning is a well-established paradigm where the robot uses human input to infer a policy or reward function capturing the desired behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In imitation learning, the robot learns a policy that essentially copies human demonstrations [47], a strategy that typically doesn’t generalize well outside the training regime [40].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Meanwhile, inverse reinforcement learning (IRL) uses the demonstrations to extract a reward function capturing why a specific behavior is desirable, thus better generalizing to unseen scenarios [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Recent research goes beyond demonstrations, utilizing other types of human input for reward learning, such as corrections [6], comparisons [55], or rankings [14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Unless explicitly designed for, these methods learn a latent representation implicitly from the respective human input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We seek to instead explicitly learn a preference-agnostic latent space that can be used for downstream tasks like reward learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We focus on learning models of human reward functions via pairwise preference queries [55], but we believe the latent space we learn can be useful for learning from any of the above types of feedback.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Representation and Similarity Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Common represen- tation learning approaches are unsupervised [18, 20, 32] or self- supervised [4, 27, 39, 48], but because they are purposefully de- signed to bypass human supervision, the learned embedding does SIRL: Similarity-based Implicit Representation Learning HRI ’23, March 13–16, 2023, Stockholm, Sweden not necessarily correspond to features the person cares about.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Prior work leverages task labels [17] or trajectory rankings [13] to learn latent spaces that help identify specific goals or preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' By contrast, we focus on learning task-agnositic measures of feature similarity that are useful for learning multiple preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Some work looks at having people interactively select features from a pre-defined set [15, 16, 42] or teach task-agnostic features sequen- tially via kinesthetic feature demonstrations [10] or active learning techniques [9, 31, 37].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We instead focus on fully learning a lower- dimensional feature representation all-at-once, rather than one at a time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Furthermore, rather than relying on the human to provide physical demonstrations for learning a good feature space [10, 11], we propose a more accessible and general form of human feedback: showing the user triplets of trajectories and simply asking them to label which two trajectories are the most similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Triplet losses have been widely used to learn similarity models that capture how humans perceive objects [2, 3, 25, 44, 54];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' however, to the best of our knowledge, we are the first to use a triplet loss to learn a general, task-agnostic similarity model of how humans perceive trajectories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Meta- and Multi-Task Reward Function Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' To learn multiple models of human reward functions, prior work has pro- posed clustering demonstrations and learning a different reward function for each cluster [5, 21, 26];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' however, these methods require a large number of demonstrations and do not adapt to new reward functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Meta-learning [29] seeks to learn a reward function ini- tialization that enables fast fine-tuning at test time [33, 51, 57, 58].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Multi-task reward learning approaches pretrain a reward function on multiple human intents and then fine-tune the reward function at test time [30, 46].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This has been shown to be more stable and scal- able than meta-learning approaches [43], but still requires curating a large set of training environments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' By contrast, we do not assume any knowledge of the test-time task distribution a priori and do not require access to a population of different reward functions dur- ing training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Rather, we focus on learning a task-agnostic feature representation that can be utilized for down-stream reward learn- ing tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In particular, we test our learned representation on the down-stream task of learning models of human reward functions via pairwise preference queries over trajectories [8, 41, 50, 55].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 3 METHOD We present our method for learning preference-agnostic represen- tations from trajectory similarity queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our intuition is that if a human judges two behaviors to be similar, then their representa- tions should also be similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Since directly asking if two trajectories are similar is difficult without an explicit threshold, we instead present the human with a triplet of trajectories and ask them to pick the two most similar (or, equivalently, the most dissimilar one).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We use the human’s answers to train the representation such that similar trajectories have embeddings that are close and dissimilar trajectories map to embeddings far apart.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The robot then uses this latent space as a shared representation for downstream preference learning tasks with multiple people, each with different preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 Preliminaries We define a trajectory 𝜉 as a sequence of states, and denote the space of all possible trajectories by Ξ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The human’s preference over trajectories is given by a reward function 𝑅 : Ξ ↦→ R that is unobserved by the robot and must be learned from human inter- action.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The robot reasons over a parameterized approximation of the reward function 𝑅𝜃, where 𝜃 represents the parameters of a neural network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' To learn 𝜃, the robot collects human preference labels over trajectories [22, 55] and seeks to find parameters 𝜃 that maximize the likelihood of the human input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The robot can then use the learned reward function to score trajectories during motion planning in order to align its behavior with a particular human’s preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We focus on explicitly using human input to first learn a good representation and then use that representation for down- stream reward learning, rather than using reward-specific human input (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=', preferences or demonstrations) to implicitly learn the representation at the same time as the reward function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 Training the Feature Representation via Trajectory Similarity Queries We seek to train a latent space that is useful for multiple down- stream preference learning tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' To do this, we propose learning a preference-agnostic model of human similarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' One way to learn such a model would be to ask users to judge whether two trajecto- ries are similar or not;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' however, humans are better at giving relative rather than binary or quantitative assessments of similarity [35, 53].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Thus, rather than asking users to use some internal threshold or scoring mechanism to quantitatively measure similarity, we instead focus on qualitative trajectory similarity queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We present the user with a visualization of three trajectories and ask them to pick the two most similar ones (equivalently the most dissimilar one).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The human’s queries form a data set D𝑠𝑖𝑚 = {(𝜉𝑖 𝑃1, 𝜉𝑖 𝑃2, 𝜉𝑖 𝑁 )}, where 𝜉𝑖 𝑃1 and 𝜉𝑖 𝑃2 are the trajectories that are most similar and 𝜉𝑖 𝑁 is the trajectory most dissimilar to the other two.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We can interpret similarity (or dissimilarity) as a distance func- tion, so we define the distance between two trajectories as the 𝐿2 feature distance: 𝑑(𝜉1, 𝜉2) = ∥𝜙(𝜉1) − 𝜙(𝜉2)∥2 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Given a dataset of trajectory similarity queries D𝑠𝑖𝑚, we use the triplet loss [7]: L𝑡𝑟𝑖𝑝 (𝜉𝐴, 𝜉𝑃, 𝜉𝑁 ) = max(𝑑(𝜉𝐴, 𝜉𝑃) − 𝑑(𝜉𝐴, 𝜉𝑁 ) + 𝛼, 0) , (1) a form of contrastive learning where 𝜉𝐴 is the anchor, 𝜉𝑃 is the positive example, 𝜉𝑁 is the negative example, and 𝛼 ≥ 0 is a margin between positive and negative pairs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' However, because our queries do not contain an explicit anchor, our final loss is as follows: L𝑠𝑖𝑚(𝜙) = |D𝑠𝑖𝑚 | ∑︁ 𝑖=1 L𝑡𝑟𝑖𝑝 (𝜉𝑖 𝑃1, 𝜉𝑖 𝑃2, 𝜉𝑖 𝑁 ) + L𝑡𝑟𝑖𝑝 (𝜉𝑖 𝑃2, 𝜉𝑖 𝑃1, 𝜉𝑖 𝑁 ) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (2) We train a similarity embedding function 𝜙 : Ξ ↦→ R𝑑 that mini- mizes the above similarity loss, where 𝑑 is the representation di- mensionality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The intuition is that optimizing this loss should push together the embeddings of similar trajectories and push apart the embeddings of dissimilar trajectories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Before training the repre- sentation with the loss in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (2), we may also pre-train it using unsupervised learning [36], which we experiment with in Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3 Using SIRL for Reward Learning Given a learned embedding 𝜙, we can use it for learning models of specific user preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' While we focus on learning from pairwise HRI ’23, March 13–16, 2023, Stockholm, Sweden TBD preferences, we note that 𝜙 can in principle be used in downstream tasks that learn from many types of human feedback [34].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' When learning a reward function from human preferences, we show the human two trajectories, 𝜉𝐴 and 𝜉𝐵, and then ask which of these two the human prefers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We collect a data set of such preferences D𝑝𝑟𝑒𝑓 = {(𝜉𝑖 𝐴, 𝜉𝑖 𝐵, ℓ𝑖)} where ℓ𝑖 = 1 if 𝜉𝑖 𝐴 is preferred to 𝜉𝑖 𝐵, denoted 𝜉𝑖 𝐴 ≻ 𝜉𝑖 𝐵, and ℓ𝑖 = 0 otherwise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We interpret the human’s prefer- ences through the lens of the Bradley-Terry preference model [12]: 𝑃𝜃 (𝜉𝐴 ≻ 𝜉𝐵) = 𝑒𝑅𝜃 (𝜙 (𝜉𝐴)) 𝑒𝑅𝜃 (𝜙 (𝜉𝐴)) + 𝑒𝑅𝜃 (𝜙 (𝜉𝐵)) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (3) We learn the reward function with a simple cross-entropy loss: L𝑝𝑟𝑒𝑓 (𝜃) = − |D𝑝𝑟𝑒𝑓 | ∑︁ 𝑖=0 ℓ𝑖 log 𝑃𝜃 (𝜉𝑖 𝐴 ≻ 𝜉𝑖 𝐵)+(1−ℓ𝑖) log 𝑃𝜃 (𝜉𝑖 𝐵 ≻ 𝜉𝑖 𝐴) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (4) 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='4 Adapting to Different User Preferences We want robots that can adapt to changes to an individual user’s preferences depending on the context as well as quickly adapt to new users’ preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Rather than learn each preference inde- pendently by collecting a new set of human data and training a completely new reward function 𝑅𝜃, we study whether we can leverage the latent space learned by SIRL to perform more accurate and sample-efficient multi-preference learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' When learning a new user’s preference model, 𝑅𝜃 the robot can use 𝜙 to more quickly learn the reward function 𝑅𝜃 (𝜙(𝜉)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our main idea is that because this shared latent representation𝜙 is trained via preference-agnostic similarity queries, it is more transferable than using a multi-task or meta-learning approach, where the pre-trained network is trained using multiple, specific task objectives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Furthermore, because SIRL uses human input to train 𝜙, we hypothesize that the learned fea- ture space will be better suited for learning human reward functions than a latent space learned via unsupervised training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4 EXPERIMENTS IN SIMULATION We first investigate the quality of SIRL-trained representations and their benefits for preference learning using simulated human input in two environments with ground truth rewards and features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 Environments GridRobot (Figure 2a) is a 5-by-5 gridworld with two obstacles and a laptop (the blue, green, and black boxes).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Trajectories are sequences of 9 states with the start and end in opposite corners.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The 19-dimensional input consists of the 𝑥 and 𝑦 coordinates of each state and a discretized angle in {−90◦, −60◦, −30◦, 0◦, 30◦, 60◦, 90◦} at the end state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The simulated human answers queries based on 4 features 𝜙∗ in this world: Euclidean distances to each object, and the absolute value of the angle orientation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' JacoRobot (Figure 2b) is a pybullet [24] simulated environment with a 7-DoF Jaco robot arm on a tabletop, with a human and laptop in the environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Trajectories are length 21, and each state consists of 97 dimensions: the 𝑥𝑦𝑧 positions of all robot joints and objects, and their rotation matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This results in a 2037- dimensional input space, much larger than for GridRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The 4 features of interest 𝜙∗ for the simulated human are: a) table — (a) GridRobot environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Preprogrammed Views Query Answers Trajectory Replay Query ID 𝜉!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 𝜉" 𝜉# (b) JacoRobot environment and user study interface.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Figure 2: Visualization of the experimental environments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' distance of the robot’s End-Effector (EE) to the table;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' b) upright — EE orientation relative to upright, to consider whether objects are carried upright;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' c) laptop — 𝑥𝑦-plane distance of the EE to a laptop, to consider whether the EE passes over the laptop at any height;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' d) proxemics [45] — proxemic 𝑥𝑦-plane distance of the EE to the human, where the EE is considered closer to the human when moving in front of the human that to their side.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In GridRobot the state space is discretized, so the trajectory space Ξ can be enumerated;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' however, the JacoRobot state space is contin- uous, so we construct Ξ by smoothly perturbing the shortest path trajectories from 10,000 randomly sampled start-goal pairs (see App.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We generate similarity and preference queries by randomly sampling from Ξ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The simulated human answers similarity queries by computing the 4 feature values for each of the three trajecto- ries and choosing the two that were closest in the feature space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For preference queries, the simulated human computes the ground truth reward and samples the trajectory with the higher reward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The space of true reward functions (used to simulate preference labels) is defined as linear combinations of the 4 features described above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The robot is not given access to the ground-truth features nor the ground-truth reward function but must learn them from similarity and preference labels over raw trajectory observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 Qualitative Examples In Figure 3 we show similar and dissimilar trajectories learned by SIRL in a simplified GridRobot environment with only the laptop and the joint angle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Top: the given trajectory stays far from the laptop and holds the cup on its side;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SIRL learns that trajectories that share those features are similar,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' despite being dissimilar in 30° 30° 60° :09 90° 90Params Triplet query5 Replay Red ReplayGreen Replay Blue Redmore similar toGreen Red more similar to Blue Green more similar to Blue Side View 1 Side View 2 Front View 1 FrontView2 Top ViewSIRL: Similarity-based Implicit Representation Learning HRI ’23,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' March 13–16,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2023,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Stockholm,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Sweden Figure 3: SIRL picks the two most and least similar trajectories to a query trajectory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Top: trajectories are similar in features despite being dissimilar in states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Bottom: trajectories are dissimilar in fea- tures despite being close in states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' the state-space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Bottom: the trajectory stays close to the laptop and holds the cup at an angle;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SIRL learns that trajectories that hold the cup upright and stay far from the laptop are dissimilar, despite being similar in the state-space (going up and then right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3 Experimental Setup Manipulated Variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We test the importance of user input that is designed to teach the representation by comparing SIRL with multi-task learning techniques from generic preference queries, and unsupervised representation learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We have 4 baselines: a) VAE, which learns a representation with a variational reconstruc- tion loss [36];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' b) MultiPref , a multi-task baseline [30], where we learn the representation 𝜙 implicitly by training multiple reward functions (each with shared initial layers) via preference learning;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' c) SinglePref , a hypothetical method that learns from an ideal user who weighs all features equally;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' d) Random, a randomly initial- ized embedding, which does not benefit from human data but is also immune from any spurious correlations that might be learned from biased data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For MultiPref, we trained versions with 10 and 50 simulated human preference rewards for good coverage of the reward space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' All embeddings have the same network size: for GridRobot we used MLPs with 2 layers, 128 units each, mapping to 6 output neurons, while for JacoRobot we used 1024 units to handle the larger input space (see App.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For a fair compar- ison, we gave SIRL, SinglePref, and MultiPref equal amounts of human data for pre-training: 𝑁 similarity queries for SIRL, and 𝑁 preference queries (used for a single human for SinglePref or equally distributed amongst humans for MultiPref).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We also per- formed ablations with and without VAE pre-training and found that SinglePref and MultiPref are better without VAE (see App.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dependent Measures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' To test the quality of the learned repre- sentations, we use two metrics: Feature Prediction Error (FPE) and Test Preference Accuracy (TPA).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The FPE metric is inspired by prior work that argues that good representations are linearly separa- ble [23, 38, 49].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our goal is to measure whether the embeddings contain the necessary information to recover the 4 ground-truth features in each environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We generate data sets of sampled trajectories labeled with their ground truth (normalized) feature vector D𝐹𝑃𝐸 = {𝜉,𝜙∗}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We freeze each embedding and add a linear regression layer on top to predict the feature vector for a given trajectory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We split D𝐹𝑃𝐸 into 80% training and 20% test pairs, and FPE is the mean squared error (MSE) on the test set between the predicted feature vector and the ground truth feature vector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For the human query methods, we report FPE with increasing number of representation training queries 𝑁.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For TPA, we test whether good representations necessarily lead to good learning of general preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We use the trained em- beddings as the base for 20 randomly selected test preference re- wards.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For each 𝑅𝜃𝑖 , we generate a set of labeled preference queries D𝜃𝑖 𝑝𝑟𝑒𝑓 = {𝜉𝐴, 𝜉𝐵,𝑙}, which we split into 80% for training and 20% for test.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We train each reward model with 𝑀 preference queries per test reward, and we vary 𝑀.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' All preference networks have the same architecture: we take the embedding 𝜙 pre-trained with the respective method, and add new fully connected layers to learn a reward function from trajectory preference labels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For GridRobot we used MLPs with 2 layers of 128 units, and for JacoRobot we used 1024 units.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We found that all methods apart from SIRL worked better with unfrozen embeddings (App.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We report TPA as the preference accuracy for the learned reward models on the test preference set, averaged across the test human preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Hypotheses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We test two hypotheses: H1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Using similarity queries specifically designed to teach the representation (SIRL) leads to representations more predictive of the true features (lower FPE) than unsupervised (VAE), implicit (MultiPref, SinglePref), or random representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' H2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The SIRL representations result in more generalizable re- ward learning (higher TPA) than unsupervised (VAE), implicit (Mul- tiPref, SinglePref), or random representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='4 Results In Figure 4 we show the FPE score for both environments with varying representation queries 𝑁 from 100 to 1000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For GridRobot, both versions of SIRL (with or without VAE pre-training) perform similarly and outperform all baselines, supporting H1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' When pre- training with preference queries, MultiPref with 10 humans per- forms better than SinglePref or MultiPref with 50 humans: Sin- glePref may be overfitting to the one human preference it has seen, while when MultiPref has to split its data budget among 50 humans it ends up learning a worse representation than Random.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' There is a balance to be struck between the diversity in human training rewards covered and the amount of pre-training data each reward gets, a trade-off which SIRL avoids because similarity queries are agnostic to the particular human reward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For the more complex JacoRobot, both versions of SIRL outperform all baselines, in line with H1, although SIRL without VAE scores better than with it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Figure 5 we present the TPA score for both environments with a varying amount of test preference queries 𝑀 from 10 to 190, and 𝑁 = 100, 500, and 1000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For GridRobot, each respective method performs comparably with different 𝑁s, suggesting that this is a simple enough environment that low amounts of representation data are sufficient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For JacoRobot, this is not the case: with just 100 queries, SIRL with VAE pre-training performs like VAE, SIRL without pre-training has random performance (since it’s frozen), and the preference baselines all perform close to Random, as if they Trajectory I Most Similar Trajectories : : Least Similar Trajectories 006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 90° 00 0° 30° 30° 30° .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' · 90° 0°HRI ’23, March 13–16, 2023, Stockholm, Sweden TBD 100 200 300 400 500 600 700 800 900 1000 Number of Representation Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='4 FPE GridRobot Feature Prediction Error (FPE) SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE 100 200 300 400 500 600 700 800 900 1000 Number of Representation Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='4 FPE JacoRobot Feature Prediction Error (FPE) SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE Figure 4: FPE for the GridRobot (left) and JacoRobot (right) environments with simulated human data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' With enough data, SIRL learns repre- sentations more predictive of the true features 𝜙∗ in both simple and complex environments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' weren’t trained with queries at all.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For larger 𝑁, both versions of SIRL start performing better than the baselines, suggesting that with enough data a good representation can be learned.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Focusing on 𝑁 = 1000, our results support H2: both SIRLs out- perform all baselines in both environments, although for JacoRobot SIRL without VAE is better than with VAE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In the GridRobot en- vironment VAE pre-training helps SIRL.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' However, while VAE per- forms comparably to other baselines in GridRobot, it severely un- derperforms in JacoRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This suggests that the reconstruction loss struggles to recover a helpful starting representation when the input space is higher dimensional and correlated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' As a result, using the VAE pre-training to warmstart SIRL hinders performance when compared to starting from a blank slate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' When comparing the preference-based baselines, in GridRobot they all perform similarly apart from MultiPref with 50 humans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In JacoRobot we see a trend that more preference humans does not necessarily result in better performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This confirms our observation from Figure 4 that deciding on an appropriate number of human preferences to use for multi-task pretraining is challenging, a problem that SIRL bypasses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Summary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' With enough representation data, SIRL learns represen- tations more predictive of the true features (H1), leading to learning more generalizable rewards (H2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This does not necessarily mean that SIRL representations are perfectly aligned with causal features — they are just better aligned, so the learned rewards are also better.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' When VAE pre-training recovers sensible starting representations it further reduces the amount of human data SIRL needs, otherwise it hurts performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Lastly, surprisingly, Random is often better than pre-training with preference queries: preference-based meth- ods may learn features that correlate with the training data but are not necessarily causal, and an incorrectly biased representation is worse for learning downstream rewards than starting from scratch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 5 USER STUDY We now present a user study with novice users that provide simi- larity queries via an interface for the JacoRobot environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 Experiment Design We ran a user study in the JacoRobot environment, modified for only two features: table and laptop (we removed the humanoid in the environment).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We designed an interface where people can click and drag to change the view, and press buttons to replay trajectories and record their query answer (Figure 2b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We chose to display the Euclidean path of each trajectory in the query traces, as we found that to help users more easily compare trajectories to one another.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The study has two phases: collecting similarity queries and col- lecting preference queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In the first phase, we introduce the user to the interface and describe the two features of interest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Because similarity queries are preference-agnostic, we describe examples of possible preferences akin to the ones in Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 1, but we do not bias the participant towards any specific preference yet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Each participant practices answering a set of pre-selected, unrecorded similarity queries, and then answers 100 recorded similarity queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In the second phase, we describe a scenario in the environment that has a specific preference associated with it (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' “There’s smoke in the kitchen, so the robot should stay high from the table” or “There is smoke in the kitchen and the robot’s mug is empty, so you want to stay far from the table and close to the laptop.”) and assign different preference scenarios to each participant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Each person practices un- recorded preference queries, then answers 100 preference queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Participants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We recruited 10 users (3 female, 6 male, 1 non-binary, aged 20-28) from the campus community to give queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Most users had technical background, so we caution that our results will speak to SIRL’s usability with this population rather than more generally.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Manipulated Variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Guided by the results in Figure 5, we compare our best performing method, SIRL without VAE, to Ran- dom, the best performing realistic baseline.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For SIRL we collect 100 similarity queries from each participant and train a shared representation using all of their data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dependent Measures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We present the same two metrics from Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 4, FPE and TPA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For TPA, we collect 100 preference queries for each user’s unique preference, we use 70% for training individual reward networks which we evaluate on the remaining 30% queries (Real).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We compute TPA with cross-validation on 50 splits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' To demonstrate how well SIRL works for new people who don’t contribute to learn- ing the similarity embedding, we also train SIRL on the similarity queries of 9 of the users and compute TPA on the held-out user’s preference data (Held-out), for each user, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Lastly, be- cause real data tends to be noisy, we also compute TPA with 70 SIRL: Similarity-based Implicit Representation Learning HRI ’23, March 13–16, 2023, Stockholm, Sweden 10 30 50 70 90 110 130 150 170 190 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot with N = 100 SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE 10 30 50 70 90 110 130 150 170 190 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot with N = 100 SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE 10 30 50 70 90 110 130 150 170 190 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot with N = 500 SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE 10 30 50 70 90 110 130 150 170 190 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot with N = 500 SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot with N = 1000 SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot with N = 1000 SIRL+VAE (Ours) SinglePref SIRL (Ours) MultiPref 10H Random MultiPref 50H VAE Figure 5: TPA for GridRobot (top) and JacoRobot (bottom) with simulated human data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' With enough data, SIRL recovers more generalizable rewards than unsupervised, preference-trained, or random representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' simulated preference queries for 10 different rewards, which we also evaluate on a simulated test set (Simulated).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Hypotheses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our hypotheses for the study are: H3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Similarity queries (SIRL) recover more salient features than a random representation (lower FPE), even with novice user data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' H4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The SIRL representation results in more generalizable re- ward learning (higher TPA), even with novice similarity queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 Analysis Figure 6 summarizes the results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' On the left, SIRL recovers a repre- sentation twice as predictive of the true features, supporting H3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A 2-sided t-test (p < .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0001) confirms this.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This suggests SIRL can recover aspects of people’s feature representation even with noisy similarity queries from novice users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' On the right (Real), SIRL recov- ers more generalizable rewards on average than Random, providing evidence for H4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Furthermore, using the SIRL representation on a novel user (Held-out) also performs better than Random, and the result appears almost indistinguishable from Real.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This sug- gests that similarity queries can be effectively crowd-sourced and the resulting representation works well for novel user preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Lastly, training with simulated preference queries slightly improves performance for both methods, suggesting that noise in the human preference data can be substantial.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Three ANOVAs with method as a factor find a significant main effect (F(1, 18) = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0175, p = .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0246, F(1, 18) = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7547, p = .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0427, and F(1, 18) = 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1068, p < .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='001, respec- tively).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For each of the 3 cases, we also separated the 6 humans that were assigned preferences pertaining to both features (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' “There is smoke in the kitchen and the robot’s mug is empty, so stay far from the table and close to the laptop.”).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SIRL performance is slightly better when using preference data from this subset of users, hinting that perhaps the learned representation entangled the two features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' HRI ’23, March 13–16, 2023, Stockholm, Sweden TBD All Humans 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='02 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='04 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='06 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='08 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='10 FPE SIRL (Ours) Random All Humans 6 Humans All Humans 6 Humans All Humans 6 Humans 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 TPA Real Held-out Simulated Figure 6: Study values for FPE, and TPA with real and simulated preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Even with novice similarity queries, SIRL recovers representations both more predictive of the true features and more useful for learning different user rewards than the baseline.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Overall, the quantitative results support H3 and H4, providing evidence that SIRL can recover more human-aligned representa- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Subjectively, some users found the 2D interface deceiving at times, as they would judge trajectory similarity differently based on the viewpoint.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This is a natural artifact of visualizing a 3D world in 2D, but future work should investigate better interfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Some users reported struggling to trade off the two features when comparing trajectories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' This is in part due to the fact that we almost “engi- neered” their internal representation, so a more in-the-wild study could determine whether similarity queries are indeed preference- agnostic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Lastly, some queries were easier than others: users’ time- to-answer varied across triplets suggesting that future work could use it as a confidence metric for how much to trust their answer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 6 DISCUSSION AND LIMITATIONS In this work, our goal was to tackle the problem of learning good representations that capture (and disentangle) the features that matter, while excluding spurious features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' If we had such a repre- sentation, then learning rewards that capture different preferences and tasks on top of it would lead to generalizable models that reli- ably incentivize the right behavior across different situations, rather than picking up on correlates and being unable to distinguish good from bad behaviors on new data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Our idea was to implicitly tap into this representation by asking people what they find similar versus not, because two behaviors will be similar if and only if their representations are similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We introduced a novel human input type, trajectory similarity queries, and tested that it leads to better representations than those learned through self-supervision or via multi-task learning: it enables learning rewards from the same training data that better rank behaviors on test data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' That said, we need to be explicit that this is not the be-all end-all solution to our goal above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The representations learned, as we see in simulation, are not perfectly aligned with causal features — they are just better aligned.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The learned rewards are not perfect — they are just better than alternatives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Similarity queries do not solve the problem fully, potentially because they suffer from the same issue preference queries do: when multiple important features change, it becomes harder to make a judgement call on what is more similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The advantage that we see from similarity queries, though, is that rewards for particular tasks might ignore or down-weigh certain features that matter in other tasks, while similarity queries are task-agnostic and implicitly capture the distribution of tasks in the human’s head.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Rather than having to specify a task distribution for multi-task learning, with similarity queries we are (implicitly) ask- ing the user to leverage their more general-purpose representation of the world.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' But thinking about how to overcome the challenge that multiple changing features make these queries harder to an- swer opens the door to exciting ideas for future work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For instance, what if we iteratively built the space, and based similarity queries on some current estimate of what are the important features;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' over time, as the representation becomes more aligned with the human’s, the queries would get better at honing in on specific features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Another obvious limitation is that we did not do an in-the-wild study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In theory, similarity queries should be used when people already have a robot they are familiar with and, thus, have a distri- bution of tasks they care about in their everyday contexts, but in our study we needed to explain to users these contexts and what might be important.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In doing so, we almost “engineered” their inter- nal representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' As robots become more prevalent, a follow-up study where users are given much less structure and allowed to actually tap into their unaltered representation might be possible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In a sense, with SIRL we build a foundation model, and this some- times requires hundreds of queries to learn a good representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' While we don’t think having this much data when pre-training is unreasonable, especially since it leads to significant desirable performance improvement over baselines, sample-complexity is crucial to address as we scale to more complex robotic tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Be- cause similarity queries are task agnostic, we can crowd-source the queries from multiple people (as we did in the user study) and rely on this economy at scale to alleviate user burden.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Future work could also look at active querying methodologies to ask the person for more informative similarity queries and reduce data amounts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A further avenue of work is extending this beyond reward learn- ing, using SIRL representations directly for policy learning or learn- ing exploration functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We also emphasize that similarity queries are not a replacement for self-supervised learning;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' instead, we view them as complementary — self-supervised learning might be able to leverage expert-designed heuristics to eliminate many of the spuri- ous features, while SIRL might serve to fine-tune the representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' How to properly combined the two remains an open question.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Overall, similarity queries are a step towards recovering human- aligned representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' They improve upon the state of the art, and can benefit from further exploration in how to combine them with other inputs and self-supervision, and how to make them easier through better interfaces and query selection algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SIRL: Similarity-based Implicit Representation Learning HRI ’23, March 13–16, 2023, Stockholm, Sweden REFERENCES [1] Pieter Abbeel and Andrew Y Ng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Apprenticeship learning via inverse reinforcement learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Machine Learning (ICML), International Conference on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ACM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [2] Sameer Agarwal, Josh Wills, Lawrence Cayton, Gert Lanckriet, David Kriegman, and Serge Belongie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Generalized non-metric multidimensional scaling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Artificial Intelligence and Statistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 11–18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [3] Ehsan Amid, Aristides Gionis, and Antti Ukkonen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A Kernel-Learning Approach to Semi-supervised Clustering with Relative Distance Comparisons, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 9284.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1007/978-3-319-23528-8_14 [4] Yusuf Aytar, Tobias Pfaff, David Budden, Tom Le Paine, Ziyu Wang, and Nando de Freitas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Playing Hard Exploration Games by Watching YouTube.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Pro- ceedings of the 32nd International Conference on Neural Information Processing Systems (Montréal, Canada) (NIPS’18).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Curran Associates Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=', Red Hook, NY, USA, 2935–2945.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [5] Monica Babes, Vukosi N Marivate, Kaushik Subramanian, and Michael L Littman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Apprenticeship learning about multiple intentions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In ICML.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [6] Andrea Bajcsy, Dylan P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Losey, Marcia K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' O’Malley, and Anca D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning Robot Objectives from Physical Human Interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 1st Annual Conference on Robot Learning (Proceedings of Machine Learning Research, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 78), Sergey Levine, Vincent Vanhoucke, and Ken Goldberg (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 217–226.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' http://proceedings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='mlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='press/v78/bajcsy17a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='html [7] Vassileios Balntas, Edgar Riba, Daniel Ponsa, and Krystian Mikolajczyk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning local feature descriptors with triplets and shallow convolutional neural networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 119.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1–119.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5244/C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='119 [8] Erdem Biyik and Dorsa Sadigh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Batch active preference-based learning of reward functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Conference on robot learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 519–528.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [9] Andreea Bobu, Chris Paxton, Wei Yang, Balakumar Sundaralingam, Yu-Wei Chao, Maya Cakmak, and Dieter Fox.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning Perceptual Concepts by Bootstrapping From Human Queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE Robotics Autom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 7, 4 (2022), 11260–11267.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/LRA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3196164 [10] Andreea Bobu, Marius Wiggert, Claire Tomlin, and Anca D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Inducing Structure in Reward Learning by Learning Features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The International Jour- nal of Robotics Research 0, 0 (0), 02783649221078031.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1177/ 02783649221078031 arXiv:https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1177/02783649221078031 [11] Andreea Bobu, Marius Wiggert, Claire Tomlin, and Anca D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Feature Expansive Reward Learning: Rethinking Human Input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 2021 ACM/IEEE International Conference on Human-Robot Interaction (Boulder, CO, USA) (HRI ’21).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Association for Computing Machinery, New York, NY, USA, 216–224.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1145/3434073.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3444667 [12] Ralph Allan Bradley and Milton E Terry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 1952.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Rank analysis of incomplete block designs: I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The method of paired comparisons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Biometrika 39, 3/4 (1952), 324–345.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [13] Daniel Brown, Russell Coleman, Ravi Srinivasan, and Scott Niekum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Safe Imitation Learning via Fast Bayesian Reward Inference from Preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 37th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 119), Hal Daumé III and Aarti Singh (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 1165–1177.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' http://proceedings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='mlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='press/v119/brown20a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='html [14] Daniel Brown, Wonjoon Goo, Prabhat Nagarajan, and Scott Niekum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Extrap- olating beyond suboptimal demonstrations via inverse reinforcement learning from observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In International Conference on Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 783– 792.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [15] Kalesha Bullard, Sonia Chernova, and Andrea Lockerd Thomaz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Human- Driven Feature Selection for a Robotic Agent Learning Classification Tasks from Demonstration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2018 IEEE International Conference on Robotics and Automation, ICRA 2018, Brisbane, Australia, May 21-25, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE, 6923–6930.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/ICRA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8461012 [16] Maya Cakmak and Andrea Lockerd Thomaz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Designing robot learners that ask good questions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In International Conference on Human-Robot Interaction, HRI’12, Boston, MA, USA - March 05 - 08, 2012, Holly A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Yanco, Aaron Steinfeld, Vanessa Evers, and Odest Chadwicke Jenkins (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ACM, 17–24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1145/2157689.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2157693 [17] Annie S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Chen, Suraj Nair, and Chelsea Finn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning Generalizable Robotic Reward Functions from "In-The-Wild" Human Videos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' CoRR abs/2103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='16817 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' arXiv:2103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='16817 https://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/abs/2103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='16817 [18] Ricky T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Chen, Xuechen Li, Roger B Grosse, and David K Duvenaud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Isolating Sources of Disentanglement in Variational Autoencoders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Advances in Neural Information Processing Systems, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Bengio, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Wallach, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Larochelle, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Grauman, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Cesa-Bianchi, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Garnett (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ), Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Curran Associates, Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [19] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A simple framework for contrastive learning of visual representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Interna- tional conference on machine learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 1597–1607.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [20] Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 30th International Conference on Neural Information Processing Systems (Barcelona, Spain) (NIPS’16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Curran Associates Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=', Red Hook, NY, USA, 2180–2188.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [21] Jaedeug Choi and Kee-Eung Kim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Nonparametric Bayesian inverse rein- forcement learning for multiple reward functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Advances in Neural Information Processing Systems 25 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [22] Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Deep Reinforcement Learning from Human Preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Advances in Neural Information Processing Systems, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Guyon, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Luxburg, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Bengio, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Wallach, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Fergus, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Vishwanathan, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Garnett (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' ), Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Curran Associates, Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://proceedings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='neurips.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='cc/paper/2017/file/ d5e2c0adad503c91f91df240d0cd4e49-Paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='pdf [23] Adam Coates and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Ng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning Feature Representations with K-Means.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Neural Networks: Tricks of the Trade.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [24] Erwin Coumans and Yunfei Bai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2016–2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PyBullet, a Python module for physics simulation for games, robotics and machine learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' http://pybullet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [25] Cagatay Demiralp, Michael Bernstein, and Jeffrey Heer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning Perceptual Kernels for Visualization Design.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE Transactions on Visualization and Computer Graphics 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/TVCG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2346978 [26] Christos Dimitrakakis and Constantin A Rothkopf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Bayesian multitask inverse reinforcement learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In European workshop on reinforcement learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Springer, 273–284.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [27] Carl Doersch, Abhinav Kumar Gupta, and Alexei A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Efros.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Unsupervised Visual Representation Learning by Context Prediction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2015 IEEE International Conference on Computer Vision (ICCV) (2015), 1422–1430.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [28] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Muelling, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Andrew Bagnell, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Srinivasa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Movement primitives via optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2015 IEEE International Conference on Robotics and Automation (ICRA).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2339–2346.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/ICRA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7139510 [29] Chelsea Finn, Pieter Abbeel, and Sergey Levine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Model-Agnostic Meta- Learning for Fast Adaptation of Deep Networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 34th Inter- national Conference on Machine Learning - Volume 70 (Sydney, NSW, Australia) (ICML’17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' JMLR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org, 1126–1135.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [30] Adam Gleave and Oliver Habryka.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Multi-task maximum entropy inverse reinforcement learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' arXiv preprint arXiv:1805.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='08882 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [31] Bradley Hayes and Brian Scassellati.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Discovering task constraints through observation and active learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems, Chicago, IL, USA, September 14-18, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE, 4442–4449.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/IROS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6943191 [32] Irina Higgins, Loïc Matthey, Arka Pal, Christopher P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Burgess, Xavier Glorot, Matthew M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Botvinick, Shakir Mohamed, and Alexander Lerchner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' beta- VAE: Learning Basic Visual Concepts with a Constrained Variational Framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In ICLR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [33] Chao Huang, Wenhao Luo, and Rui Liu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Meta Preference Learning for Fast User Adaptation in Human-Supervisory Multi-Robot Deployments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE, 5851–5856.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [34] Hong Jun Jeon, Smitha Milli, and Anca Dragan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Reward-rational (implicit) choice: A unifying formalism for reward learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Advances in Neural Information Processing Systems 33 (2020), 4415–4426.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [35] Maurice George Kendall.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 1948.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Rank correlation methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (1948).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [36] Diederik P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Kingma and Max Welling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Auto-Encoding Variational Bayes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2nd International Conference on Learning Representations, ICLR 2014, Banff, AB, Canada, April 14-16, 2014, Conference Track Proceedings, Yoshua Bengio and Yann LeCun (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/abs/1312.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6114 [37] Johannes Kulick, Marc Toussaint, Tobias Lang, and Manuel Lopes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Active Learning for Teaching a Robot Grounded Relational Symbols.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In IJCAI 2013, Proceedings of the 23rd International Joint Conference on Artificial Intelligence, Beijing, China, August 3-9, 2013, Francesca Rossi (Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IJCAI/AAAI, 1451–1457.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='aaai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/ocs/index.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='php/IJCAI/IJCAI13/paper/view/6706 [38] Cheng-I Lai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Contrastive Predictive Coding Based Feature for Automatic Speaker Verification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' arXiv preprint arXiv:1904.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='01575 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [39] Michael Laskin, Aravind Srinivas, and Pieter Abbeel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' CURL: Contrastive Unsupervised Representations for Reinforcement Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 37th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 119), Hal Daumé III and Aarti Singh (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 5639– 5650.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://proceedings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='mlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='press/v119/laskin20a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='html [40] Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Offline rein- forcement learning: Tutorial, review, and perspectives on open problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' arXiv preprint arXiv:2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='01643 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [41] Kejun Li, Maegan Tucker, Erdem Bıyık, Ellen Novoseller, Joel W Burdick, Yanan Sui, Dorsa Sadigh, Yisong Yue, and Aaron D Ames.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Roial: Region of interest active learning for characterizing exoskeleton gait preference landscapes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2021 IEEE International Conference on Robotics and Automation (ICRA).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE, 3212–3218.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [42] Hoai Luu-Duc and Jun Miura.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' An Incremental Feature Set Refinement in a Programming by Demonstration Scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 4th IEEE International Conference on Advanced Robotics and Mechatronics, ICARM 2019, Toyonaka, Japan, July 3-5, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE, 372–377.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/ICARM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8833723 [43] Zhao Mandi, Pieter Abbeel, and Stephen James.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' On the Effectiveness of Fine-tuning Versus Meta-reinforcement Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' arXiv preprint arXiv:2206.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='03271 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' HRI ’23, March 13–16, 2023, Stockholm, Sweden TBD [44] Brian McFee, Gert Lanckriet, and Tony Jebara.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning Multi-modal Similarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Journal of machine learning research 12, 2 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [45] Jonathan Mumm and Bilge Mutlu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Human-robot proxemics: physical and psychological distancing in human-robot interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Proceedings of the 6th international conference on Human-robot interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 331–338.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [46] Kentaro Nishi and Masamichi Shimosaka.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Fine-grained driving behavior prediction via context-aware multi-task inverse reinforcement learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2020 IEEE International Conference on Robotics and Automation (ICRA).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE, 2281– 2287.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [47] Takayuki Osa, Joni Pajarinen, Gerhard Neumann, J Andrew Bagnell, Pieter Abbeel, Jan Peters, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' An Algorithmic Perspective on Imitation Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Foun- dations and Trends in Robotics 7, 1-2 (2018), 1–179.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [48] Deepak Pathak, Parsa Mahmoudieh, Guanghao Luo, Pulkit Agrawal, Dian Chen, Fred Shentu, Evan Shelhamer, Jitendra Malik, Alexei A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Efros, and Trevor Darrell.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Zero-Shot Visual Imitation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2131–21313.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/ 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/CVPRW.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='00278 [49] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Reed, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Yue, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Nrusimha, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Ebrahimi, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Vijaykumar, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Mao, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Li, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Zhang, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Guillory, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Metzger, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Keutzer, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Darrell.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Self-Supervised Pretraining Improves Self-Supervised Pretraining.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In 2022 IEEE/CVF Winter Con- ference on Applications of Computer Vision (WACV).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' IEEE Computer Society, Los Alamitos, CA, USA, 1050–1060.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1109/WACV51458.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='00112 [50] Dorsa Sadigh, Anca D Dragan, Shankar Sastry, and Sanjit A Seshia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Active preference-based learning of reward functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Robotics: Science and systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [51] Seyed Kamyar Seyed Ghasemipour, Shixiang Shane Gu, and Richard Zemel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Smile: Scalable meta inverse reinforcement learning through context-conditional policies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Advances in Neural Information Processing Systems 32 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [52] Arjun Sripathy, Andreea Bobu, Zhongyu Li, Koushil Sreenath, Daniel S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Brown, and Anca D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Dragan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Teaching Robots to Span the Space of Functional Expressive Motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='48550/ARXIV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2203.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='02091 [53] Neil Stewart, Gordon DA Brown, and Nick Chater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Absolute identification by relative judgment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Psychological review 112, 4 (2005), 881.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [54] Omer Tamuz, Ce Liu, Serge Belongie, Ohad Shamir, and Adam Tauman Kalai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Adaptively learning the crowd kernel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' arXiv preprint arXiv:1105.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1033 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [55] Christian Wirth, Riad Akrour, Gerhard Neumann, Johannes Fürnkranz, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A survey of preference-based reinforcement learning methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Journal of Machine Learning Research 18, 136 (2017), 1–46.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [56] Kelvin Xu, Ellis Ratner, Anca Dragan, Sergey Levine, and Chelsea Finn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning a Prior over Intent via Meta-Inverse Reinforcement Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Pro- ceedings of the 36th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 97), Kamalika Chaudhuri and Ruslan Salakhutdi- nov (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 6952–6962.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' https://proceedings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='mlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='press/v97/xu19d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='html [57] Kelvin Xu, Ellis Ratner, Anca Dragan, Sergey Levine, and Chelsea Finn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Learning a prior over intent via meta-inverse reinforcement learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In Interna- tional Conference on Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' PMLR, 6952–6962.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [58] Lantao Yu, Tianhe Yu, Chelsea Finn, and Stefano Ermon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Meta-inverse reinforcement learning with probabilistic context variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Advances in Neural Information Processing Systems 32 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SIRL: Similarity-based Implicit Representation Learning HRI ’23, March 13–16, 2023, Stockholm, Sweden A APPENDIX A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 Trajectory generation In GridRobot the state space is discretized, so the trajectory space Ξ can be enumerated;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' however, the JacoRobot state space is continu- ous, so we need to construct Ξ by sampling the infinite-dimensional trajectory space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We randomly sample 10,000 start-goal pairs and compute the shortest path in the robot’s configuration space for each of them, 𝜉𝑆𝐺.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Each trajectory has a horizon length 𝐻 and consists of 𝑛-dimensional states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We then apply random torque de- formations 𝑢 to each trajectory to obtain a deformed trajectory 𝜉𝑆𝐺 𝐷 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In particular, we randomly select up to 3 states along the trajectory, and then deform each of the selected states with a different random torque 𝑢.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' To deform a trajectory in the direction of 𝑢 we follow: 𝜉𝑆𝐺 𝐷 = 𝜉𝑆𝐺 + 𝜇𝐴−1 ˜𝑢 , (5) where 𝐴 ∈ R𝑛(𝐻+1)×𝑛(𝐻+1) defines a norm on the Hilbert space of trajectories and dictates the deformation shape [28], 𝜇 > 0 scales the magnitude of the deformation, and ˜𝑢 ∈ R𝑛(𝐻+1) is 𝑢 at indices 𝑛𝑡 through 𝑛(𝑡 + 1) and 0 otherwise ( ˜𝑢 is 0 outside of the chosen deformation state index).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For each deformation, we randomly gen- erated 𝜇 and the index of the state the deformation is applied to.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For smooth deformations, we used a norm 𝐴 based on acceleration, but other norm choices are possible as well (see Dragan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [28] for more details).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We took inspiration for this deformation strategy from Bajcsy et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 Training details We present architecture and optimization details that can assist in reproducing our training setup.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='1 Feature networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' All embeddings have the same network size: for GridRobot we used MLPs with 2 hidden layers, 128 units each, mapping to 6 output neurons, while for JacoRobot we used 1024 units to handle the larger input space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For both environments, we used ReLU non-linearities after every linear layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We trained the VAE network with a standard variational recon- struction loss [36] also including a KL-divergence-based regular- ization term (to make the latent space regular).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' The regularization part of the loss had a weight of 𝜆 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For both environments, we optimized the loss function using Adam for 2000 epochs with an exponentially decaying learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='01 (decay rate 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='99999) and a batch-size of 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SinglePref and MultiPref with 10 and 50 humans are trained using the standard preference loss in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Christiano et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' [22] ensured that the rewards predicted by the preference network remain small by normalizing them on the fly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We instead add an 𝑙2 regulatization term on the predicted reward to the preference loss, with a weight of 10 for GridRobot and 1 for JacoRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' All three methods optimize this final loss in the same way: for GridRobot, we use Adam for 5000 epochs with a learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='01 and batch-size 32, while for JacoRobot we found a lower learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='001 to result in more stable training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Lastly, for SIRL we had the option to first pre-train with the above VAE loss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Training with the similarity objective in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (2) happens disjointly, after pre-training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For both GridRobot and JacoRobot, we optimized this loss function using Adam for 3000 epochs with an exponentially decaying learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='004 (decay rate 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='99999) and batch-size 64.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We note that our current architectures assume fixed-length tra- jectories but one could adopt an LSTM-based architecture for tra- jectories of varying length [52].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='2 Preference networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For evaluating TPA, we used preference networks on top of the embeddings for the respective methods we evaluate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For GridRobot we used MLPs with 2 hidden layers of 128 units, and for JacoRobot we used 1024 units for larger capacity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For both environments, we used ReLU non-linearities after every linear layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' We added the same 𝑙2 regularization to the loss in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' (4) as before, with weight 10 for GridRobot and 1 for JacoRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For GridRobot, we optimized our final loss function using Adam for 500 epochs with a learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='001 and a batch-size of 64.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' For JacoRobot, we increased the number of epochs to 1000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='3 Ablations Figure 5 illustrates results with frozen SIRL, and unfrozen baselines without VAE pre-training, as these were the best configurations we found for each method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' In this section, we show the complete abla- tion we performed to decide which methods benefit from frozen or unfrozen embeddings, or VAE pre-training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Figure 7 showcases the result of this ablation on both GridRobot and JacoRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Overall, we see that SIRL does better when the learned representation is frozen, while all the other methods do better when the representa- tions is unfrozen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SinglePref and the MultiPref baselines perform better without VAE pre-training, while SIRL sometimes benefits from pre-training in simple environments like GridRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' HRI ’23, March 13–16, 2023, Stockholm, Sweden TBD 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot Test Preference Accuracy (TPA) with N = 1000 SIRL (frozen) SIRL (unfrozen) SIRL+VAE (frozen) SIRL+VAE (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot Test Preference Accuracy (TPA) with N = 1000 SIRL (frozen) SIRL (unfrozen) SIRL+VAE (frozen) SIRL+VAE (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot Test Preference Accuracy (TPA) with N = 1000 SinglePref (frozen) SinglePref (unfrozen) SinglePref+VAE (frozen) SinglePref+VAE (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot Test Preference Accuracy (TPA) with N = 1000 SinglePref (frozen) SinglePref (unfrozen) SinglePref+VAE (frozen) SinglePref+VAE (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot Test Preference Accuracy (TPA) with N = 1000 MultiPref 10H (frozen) MultiPref 10H (unfrozen) MultiPref+VAE 10H (frozen) MultiPref+VAE 10H (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot Test Preference Accuracy (TPA) with N = 1000 MultiPref 10H (frozen) MultiPref 10H (unfrozen) MultiPref+VAE 10H (frozen) MultiPref+VAE 10H (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot Test Preference Accuracy (TPA) with N = 1000 MultiPref 50H (frozen) MultiPref 50H (unfrozen) MultiPref+VAE 50H (frozen) MultiPref+VAE 50H (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot Test Preference Accuracy (TPA) with N = 1000 MultiPref 50H (frozen) MultiPref 50H (unfrozen) MultiPref+VAE 50H (frozen) MultiPref+VAE 50H (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA GridRobot Test Preference Accuracy (TPA) with N = 1000 Random (frozen) Random (unfrozen) VAE (frozen) VAE (unfrozen) 10 30 50 70 90 110 130 150 170 190 Number of Preference Queries 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content='0 TPA JacoRobot Test Preference Accuracy (TPA) with N = 1000 Random (frozen) Random (unfrozen) VAE (frozen) VAE (unfrozen) Figure 7: Ablation results for GridRobot (left) and JacoRobot (right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' Overall, SIRL does better when the learned representation is frozen, while all other method do better when the representations is unfrozen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} +page_content=' SinglePref and the MultiPref baselines perform better without VAE pre-training, while SIRL sometimes benefits from pre-training in simple environments like GridRobot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/INAyT4oBgHgl3EQf5vp6/content/2301.00810v1.pdf'} diff --git a/ItAzT4oBgHgl3EQfx_5k/content/2301.01746v1.pdf b/ItAzT4oBgHgl3EQfx_5k/content/2301.01746v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8e5ab242d9bff9e7379bf6953edc959c0178e92f --- /dev/null +++ b/ItAzT4oBgHgl3EQfx_5k/content/2301.01746v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:411506b963bc5789804dba4527b02400aa15e864a8fe132de83094a5f6c5b95a +size 37257733 diff --git a/JNFOT4oBgHgl3EQfxjTe/content/tmp_files/2301.12925v1.pdf.txt b/JNFOT4oBgHgl3EQfxjTe/content/tmp_files/2301.12925v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..c97e58b32650eacd9b679af9b7cf226accebec49 --- /dev/null +++ b/JNFOT4oBgHgl3EQfxjTe/content/tmp_files/2301.12925v1.pdf.txt @@ -0,0 +1,2343 @@ +arXiv:2301.12925v1 [quant-ph] 30 Jan 2023 +Sample-Optimal Quantum Process Tomography with +Non-Adaptive Incoherent Measurements +Aadil Oufkir +Univ Lyon, Inria, ENS Lyon, UCBL, LIP, Lyon, France +Abstract +How many copies of a quantum process are necessary and sufficient to construct an +approximate classical description of it? We extend the result of Surawy-Stepney, Kahn, +Kueng, and Guta (2022) to show that ˜O(d3 +ind3 +out/ε2) copies are sufficient to learn any +quantum channel Cdin×din → Cdout×dout to within ε in diamond norm. Moreover, we show +that Ω(d3 +ind3 +out/ε2) copies are necessary for any strategy using incoherent non-adaptive +measurements. This lower bound applies even for ancilla-assisted strategies. +We consider the problem of quantum process tomography which consists of approximating +an arbitrary quantum channel–any linear map that preserves the axioms of quantum mechanics. +This task is an important tool in quantum information processing and quantum control which +has been performed in actual experiments (see e.g. [OBr+04; Bia+10; Yam+10]). Given a +quantum channel N : Cdin×din → Cdout×dout as a black box, a learner could choose the input +state and send it through the unknown quantum channel. Then, it can only extract classical +information by performing a measurement on the output state. It repeats this procedure at +different steps. After collecting a sufficient amount of classical data, the goal is to return a +quantum channel ˜ +N satisfying: +∀ρ ∈ Cdin×din ⊗ Cdin×din : ∥id ⊗ (N − ˜ +N)(ρ)∥1 ≤ ε∥ρ∥1 +(1) +with high probability. +In this work, we investigate the optimal complexity of non-adaptive +strategies using incoherent measurements. +These strategies can only use one copy of the +unknown channel at each step and must specify the input states and measurement devices +before starting the learning procedure. +Contribution The main contribution of this paper is to show that the optimal complexity of +the quantum process tomography with non-adaptive incoherent measurements is ˜Θ(d3 +ind3 +out/ε2). +First, we prove a general lower bound of Ω(d3 +ind3 +out/ε2) on the number of incoherent measure- +ments for every non-adaptive process learning algorithm. +To do so, we construct an Ω(ε)- +separated family of quantum channels close to the completely depolarizing channel of cardinal +M = exp(Ω(d2 +ind2 +out)) by choosing random Choi states of a specific form. This family is used +to encode a message from {1, . . ., M}. A process tomography algorithm can be used to decode +this message with the same error probability. Hence, the encoder and decoder should share at +least Ω(d2 +ind2 +out) nats of information. On the other hand, we show that the correlation between +the encoder and decoder can only increase by at most O(ε2/dindout) nats after each measure- +ment. Note that the naive upper bound on this correlation is O(ε2), we obtain an improvement +by a factor dindout by exploiting the randomness in the construction of the quantum channel. +This result is stated in Theorem 2.1. Next, we show that the process tomography algorithm of +1 + +[Sur+22] can be generalized to approximate an unknown quantum channel to within ε in the +diamond norm (1) using a number of incoherent measurements ˜O(d3 +ind3 +out/ε2) (Theorem 3.3). +For this, we relate the diamond norm between two quantum channels and the operator norm be- +tween their corresponding Choi states which improves on the usual inequality with the 1-norm: +∥M∥⋄ ≤ din∥JM∥1 (see e.g. [JP16]). +Related work The first works on process tomography including [CN97; PCZ97] follow +the strategy of learning the quantum states images of a complete set of basis states then +obtaining the quantum channel by an inversion. +The problem of state tomography using +incoherent measurements is fully understood even for adaptive strategies [Haa+16; Gut¸+20; +LN22; Che+22c]: +the optimal complexity is Θ(d3/ε2). +So, learning a quantum channel +can be done using O(d2 +ind3 +out) measurements, but this complexity doesn’t take into account +the accumulation of errors. +The same drawback can be seen in the resource analysis of +different strategies by [MRL08]. Another reductive approach is to use the Choi–Jamio�lkowski +isomorphism [Cho75; Jam72] to reduce the process tomography to state tomography with a +higher dimension [Leu00; DP01]. +However, this requires an ancilla and only implies a sub- +optimal upper bound O((dindout)3/(ε/din)2) = O(d5 +ind3 +out/ε2) for learning in the diamond norm. +[Sur+22] propose an algorithm for estimating the Choi state in the 2 norm that requires only +˜O(d4/ε2) ancilla-free incoherent measurements (when din = dout = d). This article generalizes +this result to the diamond norm and general input/output dimensions and shows that this +algorithm is optimal up to a logarithmic factor. +A special case of quantum process tomography is learning Pauli channels. +These channels +have weighted Pauli matrices as Kraus operators and can be learned in diamond norm using +˜O(d3/ε2) measurements [FW20] (here din = dout = d). Furthermore, it is shown that Ω(d3/ε2) +are necessary for any non-adaptive strategy [FOF23]. While the techniques of the lower bound +of this article are similar to the one in [FOF23], we obtain here a larger lower bound because, +in general, we are not restricted to weighted Pauli matrices in the Kraus operators and these +latter are implicitly chosen at random. +1 +Preliminaries +We consider quantum channels of input dimension din and output dimension dout. We use the +notation [d] := {1, . . ., d}. We adopt the bra-ket notation: a column vector is denoted |φ⟩ and +its adjoint is denoted ⟨φ| = |φ⟩†. With this notation, ⟨φ|ψ⟩ is the dot product of the vectors φ +and ψ and, for a unit vector |φ⟩ ∈ Sd, |φ⟩⟨φ| is the rank-1 projector on the space spanned by the +vector φ. The canonical basis {ei}i∈[d] is denoted {|i⟩}i∈[d] := {|ei⟩}i∈[d]. A quantum state is a +positive semi-definite Hermitian matrix of trace 1. A (din, dout)-dimensional quantum channel +is a map N : Cdin×din → Cdout×dout of the form N(ρ) = � +k AkρA† +k where the Kraus operators +{Ak}k∈K ∈ +� +Cdout×din�K satisfy � +k∈K A† +kAk = Idin. For instance, the identity map id(ρ) = ρ +admits the Kraus operator {I} and the completely depolarizing channel D(ρ) = Tr(ρ) +I +dout admits +the Kraus operators +� +1 +√dout |i⟩ ⟨j| +� +j∈[din],i∈[dout]. A map N is a quantum channel if, and only if, +it is: +• completely positive: for all ρ ≽ 0, id ⊗ N(ρ) ≽ 0 and +• trace preserving: for all ρ, Tr(N(ρ)) = Tr(ρ). +We define the diamond distance between two quantum channels N and M as the diamond +2 + +norm of their difference: +d⋄(N, M) := max +ρ +∥id ⊗ (N − M)(ρ)∥1 +where the maximization is over quantum states and the Schatten p-norm of a matrix M is +defined as ∥M∥p +p = Tr +�√ +M †M +p� +. +The diamond distance can be thought of as a worst- +case distance, while the average case distance is given by the Hilbert-Schmidt or Schatten +2-norm between the corresponding Choi states. We define the Choi state of the channel N as +JN = id ⊗ N(|Ψ⟩⟨Ψ|) ∈ Cdin×din ⊗ Cdout×dout where |Ψ⟩ = +1 +√din +�din +i=1 |i⟩ ⊗ |i⟩ is the maximally +entangled state. However, to have comparable distances, we will normalize the 2-norm which +is equivalent to unnormalizing the maximally entangled state and we define the 2-distance as +follows: +d2(N, M) := din∥JN − JM∥2 = ∥id ⊗ (N − M)(din |Ψ⟩⟨Ψ|)∥2. +This is a valid distance since the map J : N �→ id ⊗ N(|Ψ⟩⟨Ψ|) is an isomorphism called the +Choi–Jamio�lkowski isomorphism [Cho75; Jam72]. Note that J should be positive semi-definite +and satisfy Tr2(J ) = +I +din to be a valid Choi state (corresponding to a quantum channel). +We consider the channel tomography problem which consists of learning a quantum channel +N in the diamond distance. Given a precision parameter ε > 0, the goal is to construct a +quantum channel ˜ +N satisfying with at least a probability 2/3: +d⋄(N, ˜ +N ) ≤ ε. +An algorithm A is 1/3-correct for this problem if it outputs a quantum channel ε-close to N +with a probability of error at most 1/3. We choose to learn in the diamond distance because it +characterizes the minimal error probability to distinguish between two quantum channels when +auxiliary systems are allowed [Wat18]. +The learner can only extract classical information from the unknown quantum channel N +by performing a measurement on the output state. Throughout the paper, we only consider +unentangled or incoherent measurements. That is, the learner can only measure with a d (or +d × d)-dimensional measurement device. Precisely, a d-dimensional measurement is defined by +a POVM (positive operator-valued measure) with a finite number of elements: this is a set of +positive semi-definite matrices M = {Mx}x∈X acting on the Hilbert space Cd and satisfying +� +x∈X Mx = I. Each element Mx in the POVM M is associated with the outcome x ∈ X. The +tuple {Tr(ρMx)}x∈X is non-negative and sums to 1: it thus defines a probability. Born’s rule +[Bor26] says that the probability that the measurement on a quantum state ρ using the POVM +M will output x is exactly Tr(ρMx). Depending on whether an auxiliary system is allowed to +be used, we distinguish two types of strategies. +M1 +ρ1 +N +x1 +M2 +ρ2 +N +x2 +... +MN +ρN +N +xN +Figure 1: Illustration of an ancilla-free independent strategy for quantum process tomography. +Ancilla-free strategies +At each step t, the learner would choose an input din-dimensional +state ρt +∈ +Cdin×din and a dout-dimensional measurement device Mt += +{M t +x}x∈Xt +∈ +(Cdout×dout)Xt. It thus sees the outcome xt ∈ Xt with a probability Tr(N(ρt)M t +xt) (see Fig. 1). +3 + +Ancilla-assisted strategies +At each step t, the learner would choose an input d × din- +dimensional state ρt ∈ Cd×d ⊗ Cdin×din and a d × dout-dimensional measurement device +Mt = {M t +x}x∈Xt ∈ (Cd×d ⊗ Cdout×dout)Xt. It thus sees the outcome xt ∈ Xt with a probability +Tr(id ⊗ N(ρt)M t +xt) (see Fig. 2). Note that ancilla-assisted strategies were proven to provide an +ρ1 +M1 +x1 +N +ρ2 +M2 +x2 +N +... +ρN +MN +xN +N +Figure 2: Illustration of an ancilla-assisted independent strategy for quantum process tomogra- +phy. +exponential (in the number of qubits n = log2(d)) advantage over ancilla-free strategies for some +problems [Che+22a; Che+22b]. However, in this work, we show that ancilla-assisted strategies +cannot overcome ancilla-free strategies for process tomography. Finally, we only consider non- +adaptive strategies: the input states and measurement devices should be chosen before starting +the learning procedure and thus cannot depend on the observations. +Given two random variables X and Y taking values in the sets [d] and [d′] respectively, +the mutual information between X and Y is the Kullback Leibler divergence between the joint +distribution P(X,Y ) and the product distribution PX × PY : +I(X : Y ) = +d +� +i=1 +d′ +� +j=1 +P (X = i, Y = j) log +� +P (X = i, Y = j) +P (X = i) P (Y = j) +� +. +All the logs of this paper are taken in base e and the information is measured in “nats”. +2 +Lower bound +In this section, we would like to investigate the intrinsic limitations of learning quantum channels +using incoherent measurements. To avoid repetition, we consider only ancilla-assisted strategies +since they contain ancilla-free strategies as a special case: one can map every din-dimensional +input state ρ to the d×din-dimensional input state ˜ρ = I +d ⊗ρ and every dout-dimensional POVM +M = {Mx}x∈X to the d × dout-dimensional POVM +˜ +M = {Id ⊗ Mx}x∈X. Mainly, we prove the +following theorem: +Theorem 2.1. Let ε ≤ 1/16 and dout ≥ 4. Any non-adaptive ancilla-assisted algorithm for +process tomography in diamond distance requires +N = Ω +�d3 +ind3 +out +ε2 +� +incoherent measurements. +4 + +Proof. For the proof, we use the construction of the Choi state: +JU = +I +dindout ++ +ε +dindout +(U + U †) − +ε +dindout +Tr2(U + U †) ⊗ +I +dout +where U ∼ Haar(dindout). JU is Hermitian and satisfies Tr2(J ) = +I +din . Moreover, JU ≽ 0 for +ε ≤ 1/4. Indeed, U is a unitary so it has an operator norm 1 thus ∥U + U †∥∞ ≤ 2. Besides, +∥Tr2(U + U †)⊗ +I +dout ∥∞ = +1 +dout ∥Tr2(U + U †)∥∞ ≤ maxi ∥I⊗ ⟨i| (U + U †)I⊗ |i⟩ ∥∞ ≤ 2. We claim +that: +Lemma 2.2. We can construct an ε/2-separated (according to the diamond distance) family +{Nx}x∈[M] of cardinal M = exp(Ω(d2 +ind2 +out)). +Proof. It is sufficient to show that for U, V ∼ Haar(dindout): +P (∥JU − JV ∥1 ≤ ε/2) ≤ exp +� +−Ω(d2 +ind2 +out) +� +. +because, once this concentration inequality holds, we can choose our family randomly, and by the +union bound, it will be ε/2-separated with an overwhelming probability (1−exp +� +−Ω(d2 +ind2 +out) +� +) +using the inequality d⋄(NU, NV ) ≥ ∥JU − JV ∥1. First, let us lower bound the expected value. +E (∥JU − JV ∥1) ≥ +ε +dindout +E +� +∥U + U † − V − V †∥1 +� +− +ε +dind2 +out +E +� +∥Tr2(U + U † − V − V †) ⊗ I∥1 +� +. +On one hand, we can upper bound the second expectation using the triangle and the Cauchy- +Schwartz inequalities: +E +� +∥Tr2(U + U † − V − V †) ⊗ I∥1 +� +≤ 4E (∥Tr2(U) ⊗ I∥1) +≤ 4 +� +dindoutE (∥Tr2(U) ⊗ I∥2) ≤ 4 +� +dindout +� +E (Tr(Tr2(U)Tr2(U †) ⊗ I)) += 4 +� +dindout +� +dout +� +� +� +� +�E + +� +i +� +k,l +⟨i| ⊗ ⟨k| UI ⊗ |k⟩ ⟨l| U † |i⟩ ⊗ |l⟩ + + += 4 +� +dindout +� +dout +� +� +� +� +�E + + +din +� +i=1 +dout +� +k,l=1 +dinδk,l +dindout + + = 4dindout. +On the other hand, we can lower bound the first expectation using H¨older’s inequality. +E +� +∥U + U † − V − V †∥1 +� +≥ +� +(E (Tr(U + U † − V − V †)2))3 +E (Tr(U + U † − V − V †)4) +≥ +� +(4dindout)3 +16dindout += 2dindout. +Therefore: +E (∥JU − JV ∥1) ≥ +ε +dindout +E +� +∥U + U † − V − V †∥1 +� +− +4ε +dind2 +out +E (∥Tr2U ⊗ I∥1) +≥ 2ε − 4ε +dout +≥ ε +for +dout ≥ 4. +5 + +Now, we claim that the function (U, V ) �→ ∥JU − JV ∥1 is +8ε +√dindout -Lipschitz. Indeed, we have +∥Tr2(X) ⊗ I∥1 ≤ √dindout∥Tr2(X) ⊗ I∥2 = √dindout∥Tr2(X)∥2 ≤ √dindoutdout∥X∥2 where the +last inequality can be found in [LZK08]. Therefore, by letting X = U − U ′ and Y = V − V ′ and +using the triangle inequality we obtain: +|∥JU − JV ∥1 − ∥JU′ − JV ′∥1| +≤ +2ε +dindout +� +∥X∥1 + ∥Y ∥1 + +����Tr2(X) ⊗ +I +dout +���� +1 ++ +����Tr2(Y ) ⊗ +I +dout +���� +1 +� +≤ 2√dindoutε +dindout +(∥U − U ′∥2 + ∥V − V ′∥2) + 2√dindoutdoutε +dind2 +out +(∥U − U ′∥2 + ∥V − V ′∥2) +≤ +8ε +√dindout +∥(U, V ) − (U ′, V ′)∥2 +(Cauchy-Schwartz) +so by the concentration inequality for Lipschitz functions of Haar measure [MM+13]: +P (∥JU − JV ∥1 ≤ ε/2) +≤ P (∥JU − JV ∥1 − E (∥JU − JV ∥1) ≤ −ε/2) +≤ exp +� +− +dindoutε2 +48 × 64ε2/dindout +� += exp +� +−Ω(d2 +ind2 +out) +� +. +Now, we use this ε/2-separated family of quantum channels {Nx}x∈[M] (corresponding +to the Choi states {Jx}x∈[M] found in Lemma 2.2) to encode a uniformly random message +X ∼ Uniform([M]) by the map X �→ NX. Using a learning algorithm for process tomography +with precision ε/4 and an error probability at most 1/3, a decoder Y can find X with the +same error probability because the family {Nx}x∈[M] is ε/2-separated. By Fano’s inequality, +the encoder and decoder should share at least Ω(log(M)) nats of information. +Lemma 2.3. [Fan61] We have +I(X : Y ) ≥ 2/3 log(M) − log(2) ≥ Ω(d4). +The remaining part of the proof is to upper bound this mutual information in terms of the +number of measurements N, the dimensions din, dout, and the precision parameter ε. Intuitively, +the mutual information, after a few measurements, is very small and then it increases when +the number of measurements increases. To make this intuition formal, let N be a number of +measurements sufficient for process tomography and let (I1, . . . , IN) be the observations of the +learning algorithm, we apply first the data processing inequality to relate the mutual information +between the encoder and the decoder with the mutual information between the uniform random +variable X and the observations (I1, . . . , IN): +I(X : Y ) ≤ I(X : I1, . . . , IN). +Then we apply the chain rule for the mutual information: +I(X : I1, . . . , IN) = +N +� +t=1 +I(X : It|I≤t−1) +where we use the notation I≤t = (I1, . . . , It) and I(X : It|I≤t−1) is the conditional mutual +information between X and It given I≤t−1. A learning algorithm A would choose the input +6 + +states {ρt}t∈[N] and measurement devices {Mt}t∈[N] which can be chosen to have the form +Mt = {µt +i |φt +i⟩⟨φt +i|}i∈It where µt +i ≥ 0 and ⟨φt +i|φt +i⟩ = 1 for all t, i. Using Jensen’s inequality, we +can prove the following upper bound on the conditional mutual information: +Lemma 2.4. For x ∈ [M], let Mx = Nx − D where D(ρ) = Tr(ρ) +I +dout is the completely +depolarizing channel. We have for all t ∈ {1, . . . , N}: +I(X : It|I≤t−1) ≤ 3 +M +� +i∈It,x∈[M] +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� �⟨φt +i| id ⊗ Mx(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 +Proof. Let t ∈ {1, . . . , N} and x ∈ [M]. Let i = (i1, . . . , it) ∈ (I1, . . . , It), we can express the +joint probability p of (X, I1, . . . , It) as follows: +p(x, i1, . . . , it) = 1 +M +t� +k=1 +µk +ik +� +φk +ik +�� id ⊗ Nx(ρk) +��φk +ik +� +We can remark that, for all 1 ≤ k ≤ t: +p(x, i≤k) = µk +ik +� +φk +ik +�� id ⊗ Nx(ρk) +��φk +ik +� +p(x, i≤k−1) += µk +ik +� +φk +ik +�� id ⊗ D(ρk) +��φk +ik +� +(1 + Φk +x,ik)p(x, i≤k−1) +where Φk +x,ik = ⟨φk +ik|id⊗Mx(ρk)|φk +ik⟩ +� +φk +ik +���id⊗D(ρk) +���φk +ik +� +because D + Mx = Nx. +So, the ratio of conditional +probabilities can be written as: +p(x, it|i≤t−1) +p(x|i≤t−1)p(it|i≤t−1) = p(x, i≤t)p(i≤t−1) +p(x, i≤t−1)p(i≤t) += µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(1 + Φt +x,it)p(x, i≤t−1)p(i≤t−1) +p(x, i≤t−1) � +y p(y, i≤t) += µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(1 + Φt +x,it)p(i≤t−1) +� +y p(y, i≤t) += +µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(1 + Φt +x,it)p(i≤t−1) +� +y µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(1 + Φt +y,it)p(y, i≤t−1) += +(1 + Φt +x,it)p(i≤t−1) +� +y(1 + Φt +y,it)p(y, i≤t−1) = +(1 + Φt +x,it) +� +y(1 + Φt +y,it)p(y|i≤t−1) +Therefore by Jensen’s inequality: +I(X : It|I≤t−1) = E +� +log +� +p(x, it|i≤t−1) +p(x|i≤t−1)p(it|i≤t−1) +�� += E +� +log +� +(1 + Φt +x,it) +� +y p(y|i≤t−1)(1 + Φt +y,it) +�� +≤ E +� +log(1 + Φt +x,it) − +� +y +p(y|i≤t−1) log(1 + Φt +y,it) +� += E +� +log(1 + Φt +x,it) +� +− +� +y +E +� +p(y|i≤t−1) log(1 + Φt +y,it) +� +. +7 + +The first term can be upper bounded using the inequality log(1 + x) ≤ x verified for all +x ∈ (−1, ∞): +E +� +log(1 + Φt +x,it) +� += Ex,i∼p log(1 + Φt +x,it) +≤ Ex,i∼pΦt +x,it = Ex,i∼p≤tΦt +x,it += Ex,i∼p≤t−1 +� +it +µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(1 + Φt +x,it)Φt +x,it += Ex,i∼p≤t−1 +� +it +µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(Φt +x,it)2 += 1 +M +M +� +x=1 +� +it +µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(Φt +x,it)2 +because � +it µt +it +� +φt +it +�� id⊗D(ρt) +��φt +it +� +Φt +x,it = Tr(id⊗Mx(ρt)) = Tr(id⊗Nx(ρt))−Tr(id⊗D(ρt)) = +Tr(ρt) − Tr(ρt) = 0 and we use the condition that the algorithm is non-adaptive in the last line. +On the other hand, the second term can be upper bounded using the inequality − log(1 + x) ≤ +−x + x2/2 verified for all x ∈ (−1/2, ∞). Let λt +it = µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +, we have : +E +� +− +� +y +p(y|i≤t−1) log(1 + Φt +y,it) +� += +� +y +Ex,i∼pp(y|i≤t−1)(− log)(1 + Φt +y,it) += +� +y +Ex,i∼p≤tp(y|i≤t−1)(− log)(1 + Φt +y,it) +≤ +� +y +Ex,i∼p≤tp(y|i≤t−1)(−Φt +y,it + (Φt +y,it)2/2) +≤ +� +y +Ex,i∼p≤t−1p(y|i≤t−1) +� +it +λt +it((Φt +x,it)2 + (Φt +y,it)2) += 2 +� +y +Ex,i∼p≤t−1p(y|i≤t−1) +� +it +λt +it(Φt +x,it)2 += 2Ex,i∼p≤t−1 +� +it +µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(Φt +x,it)2 += 2Ex,i∼p≤t−1 +� +it +λt +it(Φt +x,it)2 += 2 1 +M +M +� +x=1 +� +it +µt +it +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +(Φt +x,it)2 +where we use the condition that the algorithm is non-adaptive in the last line. +Since the +conditional mutual information is upper bounded by the sum of these two terms, the upper +bound on the conditional mutual information follows. +It remains to approximate every mean +1 +M +�M +x=1 by the expectation EU. +8 + +Lemma 2.5. We have with at least a probability 9/10: +1 +M +� +t,i,x +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� �⟨φt +i| id ⊗ Mx(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 +≤ +� +t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� +EU +�⟨φt +i| id ⊗ MU(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 ++ 16Nε2 +� +log(10) +M +. +Proof. Denote by f t +x the function |φ⟩ �→ +⟨φ|id⊗Mx(ρt)|φ⟩2 +⟨φ|id⊗D(ρt)|φ⟩2 . +We claim that the functions f t +x +are bounded. +Indeed, we can write ρt = � +i λi |ψi⟩⟨ψi| and every |ψi⟩ can be written as +|ψi⟩ = Ai ⊗ I |Ψ⟩ so for a unit vector |φ⟩, we have: +f t +x(|φ⟩) = ⟨φ| id ⊗ Mx(ρt) |φ⟩2 +⟨φ| id ⊗ D(ρt) |φ⟩2 += +4ε2 � +⟨φ| � +i λi(Ai ⊗ I) +� +Ux − Tr2(Ux) ⊗ +I +dout +� +(A† +i ⊗ I) |φ⟩ +�2 +⟨φ| � +i λiAiA† +i ⊗ I |φ⟩2 +≤ +4ε2 ⟨φ| � +i λi(Ai ⊗ I)(A† +i ⊗ I) |φ⟩2 ���Ux − Tr2(Ux) ⊗ +I +dout +��� +2 +∞ +⟨φ| � +i λiAiA† +i ⊗ I |φ⟩2 +≤ 16ε2 ⟨φ| � +i λi(Ai ⊗ I)(A† +i ⊗ I) |φ⟩2 +⟨φ| � +i λiAiA† +i ⊗ I |φ⟩2 += 16ε2 +where we used that ∥Ux∥∞ = 1 and ∥Tr2(Ux)∥∞ ≤ dout for a unitary Ux. +But we have +� +i µt +i ⟨φt +i| id ⊗ D(ρt) |φt +i⟩ = Tr(id ⊗ D(ρt)) = 1 so for all x ∈ [M]: +� +t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� �⟨φt +i| id ⊗ Mx(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 +≤ 16Nε2. +Therefore, by Hoeffding’s inequality [Hoe63] and the union bound, we have with a probability +at least 9/10: +1 +M +� +x,t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� �⟨φt +i| id ⊗ Mx(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 +≤ +� +t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� +EU +�⟨φt +i| id ⊗ MU(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 ++ 16Nε2 +� +log(10) +M +. +9 + +These two Lemmas 2.4, 2.5 imply: +I(X : I1, . . . , IN) = +N +� +t=1 +I(X : It|I≤t−1) +≤ 3 +M +� +x,t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� +E +�⟨φt +i| id ⊗ Mx(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 +≤ 3 +� +t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� +E +�⟨φt +i| id ⊗ MU(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 ++ 48Nε2 +� +log(10) +M +≤ 3N sup +t,i +E +��⟨φt +i| id ⊗ MU(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2� ++ 48Nε2 +� +log(10) +M +(2) +where we used that fact that for all t ∈ [N]: � +i µt +i ⟨φt +i| id⊗D(ρt) |φt +i⟩ = Tr(id⊗D(ρt)) = Tr(ρt) = +1. The error probability 1/10 of this approximation can be absorbed in the construction above +by asking the unitaries {Ux}x∈[M] not only to satisfy the separability condition, but also to +satisfy the inequalities in Lemma 2.5: +1 +M +� +t,i,x +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� �⟨φt +i| id ⊗ Mx(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 +≤ +� +t,i +µt +i +� +φt +i +�� id ⊗ D(ρt) +��φt +i +� +EU +�⟨φt +i| id ⊗ MU(ρt) |φt +i⟩ +⟨φt +i| id ⊗ D(ρt) |φt +i⟩ +�2 ++ 48Nε2 +� +log(10) +M +. +Now fix t ∈ [N], it ∈ It and |φ⟩ = +��φt +it +� +. Recall that we can write ρt = � +i λi |ψi⟩⟨ψi|, the +maximally entangled state is denoted |Ψ⟩ = +1 +√din +�din +i=1 |ii⟩ and every |ψi⟩ can be written as +|ψi⟩ = Ai ⊗ I |Ψ⟩ so: +id ⊗ D(ρt) = +� +i +λi(id ⊗ D)(Ai ⊗ I |Ψ⟩⟨Ψ| A† +i ⊗ I) += +� +i +λi(Ai ⊗ I)id ⊗ D(|Ψ⟩⟨Ψ|)(A† +i ⊗ I) += +� +i +λi(Ai ⊗ I) +I +dindout +(A† +i ⊗ I) += +� +i λiAiA† +i +din +⊗ +I +dout +. +(3) +10 + +On the other hand, using the notation V = U − Tr2(U) ⊗ +I +dout , we can write: +id ⊗ M(ρt) = +� +i +λiid ⊗ M(Ai ⊗ I |Ψ⟩⟨Ψ| A† +i ⊗ I) += +� +i +λi(Ai ⊗ I)id ⊗ (N − D)(|Ψ⟩⟨Ψ|)(A† +i ⊗ I) += +� +i +λi(Ai ⊗ I) +� +JN − +I +dindout +� +(A† +i ⊗ I) += +ε +dindout +� +i +λiAi ⊗ I +� +U + U † − Tr2(U + U †) ⊗ +I +dout +� +A† +i ⊗ I) += +ε +dindout +� +i +λi +� +(Ai ⊗ I)V (A† +i ⊗ I) + (Ai ⊗ I)V †(A† +i ⊗ I) +� +. +By Ineq. 2, we need to control the expectation EU ⟨φ| id ⊗ MU(ρt) |φ⟩2. +First, we replace +id ⊗ M(ρt) with the latter expression, then we apply the inequality (x + y)2 ≤ 2x2 + 2y2 to +separate the terms involving U and the terms involving Tr2(U). The first term can be computed +and bounded as follows. +4ε2 +d2 +ind2 +out +E + + +�����⟨φ| +�� +i +λi(Ai ⊗ I)U(A† +i ⊗ I) +� +|φ⟩ +����� +2 + += +4ε2 +d2 +ind2 +out +� +i,j +λiλj +dindout +���Tr +� +A† +i ⊗ I |φ⟩ ⟨φ| Aj ⊗ I +���� +2 +(CS) +≤ +4ε2 +d2 +ind2 +out +� +i,j +λiλj +dindout +⟨φ| AiA† +i ⊗ I |φ⟩ ⟨φ| AjA† +j ⊗ I |φ⟩ += +4ε2 +d3 +ind3 +out +� +⟨φ| +� +i +λiAiA† +i ⊗ I |φ⟩ +�2 +. +(4) +Let’s move to the second term which involves the partial trace. Let Mij = (A† +i ⊗I) |φ⟩⟨φ| (Aj ⊗I). +4ε2 +d2 +ind2 +out +E + + +�����⟨φ| +� +i +λi(Ai ⊗ I) +� +Tr2(U) ⊗ +I +dout +� +(A† +i ⊗ I) |φ⟩ +����� +2 + += +4ε2 +d2 +ind4 +out +� +i,j +λiλjE +� +Tr +� +(Tr2(U) ⊗ I) Mi,j +� +Tr2(U †) ⊗ I +� +M † +ij) +�� += +4ε2 +d2 +ind4 +out +� +i,j +λiλj +din +� +x,y,z,t=1 +dout +� +k,l=1 +E +� +⟨xk| U |yk⟩ ⟨zl|U † |tl⟩ +� +Tr +� � +|y⟩ ⟨x| ⊗ IMij |t⟩ ⟨z| ⊗ IM † +ij +� � += +4ε2 +d3 +ind5 +out +� +i,j +λiλj +din +� +x=t,y=z=1 +dout +� +k=l=1 +Tr +� � +|y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † +ij +� � += +4ε2 +d3 +ind4 +out +� +i,j +λiλj +din +� +x,y=1 +Tr +� � +|y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † +ij +� � +. +11 + +To control the latter expression, we write |φ⟩ = B†⊗I |Ψ⟩ so that Mi,j = (A† +i ⊗I) |φ⟩⟨φ| (Aj⊗I) = +(A† +iB†⊗I) |Ψ⟩⟨Ψ| (BAj⊗I). Using the property of the maximally entangled state ⟨Ψ| M⊗I |Ψ⟩ = +1 +din Tr(M) we obtain: +din +� +x,y=1 +Tr +� � +|y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † +ij +� += +din +� +x,y=1 +Tr +� +|y⟩ ⟨x| ⊗ I(A† +iB† ⊗ I) |Ψ⟩⟨Ψ| (BAj ⊗ I) |x⟩ ⟨y| ⊗ I(A† +jB† ⊗ I) |Ψ⟩⟨Ψ| (BAi ⊗ I) +� += +din +� +x,y=1 +⟨Ψ| (BAj ⊗ I) |x⟩ ⟨y| ⊗ I(A† +jB† ⊗ I) |Ψ⟩ ⟨Ψ| (BAi ⊗ I)† |y⟩ ⟨x| ⊗ I(A† +iB† ⊗ I) |Ψ⟩ += 1 +d2 +in +din +� +x,y=1 +Tr(BAj |x⟩ ⟨y| A† +jB†)Tr(BAi |y⟩ ⟨x| A† +iB†) += 1 +d2 +in +din +� +x,y=1 +⟨y| A† +jB†BAj |x⟩ ⟨x| A† +iB†BAi |y⟩ = 1 +d2 +in +Tr +� +A† +jB†BAjA† +iB†BAi +� +. +On the other hand, we can write +⟨φ| +� +i +λiAiA† +i ⊗ I |φ⟩ = ⟨Ψ| +� +i +λiBAiA† +iB† ⊗ I |Ψ⟩ = 1 +din +Tr +�� +i +λiA† +iB†BAi +� +. +Note that the matrix � +i λiA† +iB†BAi is positive semi-definite so: +� +i,j +λiλj +1 +d2 +in +Tr +� +A† +jB†BAjA† +iB†BAi +� += 1 +d2 +in +Tr +�� +i +λiA† +iB†BAi +�2 +≤ +� +1 +din +Tr +�� +i +λiA† +iB†BAi +� �2 += ⟨φ| +� +i +λiAiA† +i ⊗ I |φ⟩2 . +Hence +4ε2 +d2 +ind2 +out +E + + +�����⟨φ| +� +i +λi(Ai ⊗ I) +� +Tr2(U) ⊗ +I +dout +� +(A† +i ⊗ I) |φ⟩ +����� +2 + += +4ε2 +d3 +ind4 +out +� +i,j +λiλj +din +� +x,y=1 +Tr +� � +|y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † +ij +� � += +4ε2 +d3 +ind4 +out +� +i,j +λiλj +1 +d2 +in +Tr +� +A† +jB†BAjA† +iB†BAi +� +≤ +4ε2 +d3 +ind4 +out +⟨φ| +� +i +λiAiA† +i ⊗ I |φ⟩2 +(5) +Using the equality (3) and the two inequalities (4) and (5), we deduce: +E +��⟨φ| id ⊗ MU(ρt) |φ⟩ +⟨φ| id ⊗ D(ρt) |φ⟩ +�2� +≤ +8ε2 +d3 +ind3 +out +� +⟨φ| � +i λiAiA† +i ⊗ I |φ⟩ +�2 +⟨φ| +� +i λiAiA† +i +din +⊗ +I +dout |φ⟩2 += +8ε2 +dindout +. +12 + +Therefore using the inequality (2): +I(X : I1, . . . , IN) = +N +� +t=1 +I(X : It|I≤t−1) +≤ 3N sup +t,it +E + + +�� +φt +it +�� id ⊗ MU(ρt) +��φt +it +� +� +φt +it +�� id ⊗ D(ρt) +��φt +it +� +�2 + + 48Nε2 +� +log(10) +M +≤ 24N +ε2 +dindout ++ 48Nε2 +� +log(10) +M +≤ O +� +N +ε2 +dindout +� +because M = exp(Ω(d2 +ind2 +out)). +But from the data processing inequality and Lemma 2.3, +I(X : I1, . . . , IN) ≥ I(X : Y ) ≥ Ω(d2 +ind2 +out), we deduce that: +O +� +N +ε2 +dindout +� +≥ I(X : I1, . . . , IN) ≥ Ω(d2 +ind2 +out). +Finally, the lower bound follows: +N ≥ Ω +�d3 +ind3 +out +ε2 +� +. +To assess this lower bound, it is necessary to design an algorithm for quantum process +tomography. This will be the object of the following section. +3 +Upper bound +In this section, we propose an upper bound on the complexity of the quantum process +tomography problem. We generalize the algorithm proposed by [Sur+22] which is ancilla-free. +Theorem 3.1. [Sur+22] There is an ancilla-free process tomography algorithm that learns a +quantum channel (of din = dout = d) in the distance d2 using only a number of measurements: +N = O +�d6 log(d) +ε2 +� +. +This algorithm proceeds by providing an unbiased estimator for the Choi state JN , then +projecting this matrix to the space of Choi states (PSD and partial trace I/d) and finally by +invoking the Choi–Jamio�lkowski isomorphism we obtain an approximation of the channel. This +reduction from learning the Choi state in the operator norm to learning the quantum channel +in the d2 distance uses mainly the inequality d2(N, M) = d∥JN − JM∥2 ≤ d2∥JN − JM∥∞ +when din = dout = d. We generalize this result to the diamond norm and any input/output +dimensions. For this we show the following inequality: +Lemma 3.2. Let N1 and N2 be two quantum channels. We have: +d⋄(N1, N2) ≤ dindout∥JN1 − JN2∥∞. +This inequality can also be obtained by applying the inequality (3) of [Nec+18] and the +triangle inequality. We provide a simpler proof for completeness. +13 + +Proof. Denote by M = N1 − N2. Let |φ⟩ be a maximizing unit vector of the diamond norm, +i.e., ∥id ⊗ M(|φ⟩⟨φ|)∥1 = d⋄(N1, N2). We can write |φ⟩ = A ⊗ I |Ψ⟩ where |Ψ⟩ = +1 +√din +�din +i=1 |ii⟩ +is the maximally entangled state. |φ⟩ has norm 1 so +1 +din Tr(A†A) = ⟨Ψ| A†A ⊗ I |Ψ⟩ = ⟨φ|φ⟩ = 1. +On the other hand we can write +d⋄(N1, N2) = ∥id ⊗ M(|φ⟩⟨φ|)∥1 += ∥I ⊗ M(A ⊗ Idin |Ψ⟩⟨Ψ| A† ⊗ Idin)∥1 += ∥(A ⊗ Idout)id ⊗ M(|Ψ⟩⟨Ψ|)(A† ⊗ Idout)∥1 += ∥(A ⊗ Idout)JM(A† ⊗ Idout)∥1. +JM is Hermitian so it can be written as : JM = � +i λi |ψi⟩⟨ψi|. Using the triangle inequality, +we obtain: +∥(A ⊗ Idout)JM(A† ⊗ Idout)∥1 += +�����(A ⊗ Idout) +� +i +λi |ψi⟩⟨ψi| (A† ⊗ Idout) +����� +1 +≤ +� +i +|λi|∥(A ⊗ Idout) |ψi⟩⟨ψi| (A† ⊗ Idout)∥1 +≤ max +i +|λi| +� +i +∥(A ⊗ Idout) |ψi⟩⟨ψi| (A† ⊗ Idout)∥1 += ∥J ∥∞ +� +i +Tr((A ⊗ Idout) |ψi⟩⟨ψi| (A† ⊗ Idout)) += ∥J ∥∞Tr(AA† ⊗ Idout) = dindout∥J ∥∞. +This Lemma shows that the diamond and 2 distances satisfy the same inequality with respect +to the infinity norm between the Choi states when din = dout = d. Since the algorithm of +[Sur+22] approximates first the Choi state in the infinity norm, we obtain the same upper +bound for the diamond distance. For general dimensions, we obtain the following complexity: +Theorem 3.3. There is a non-adaptive ancilla-free process tomography algorithm that learns a +quantum channel in the distance d⋄ using only a number of measurements: +N = O +�d3 +ind3 +out log(dindout) +ε2 +� +. +This complexity was expected for process tomography with incoherent measurements since +the complexity of state tomography with incoherent measurements is Θ +� +d3 +ε2 +� +[Haa+16] and +learning (din, dout)-dimensional channels can be thought of as learning states of dimension +din × dout. We believe that the log(dindout)-factor can be removed from the upper bound in +Theorem 3.3 using the techniques of [Gut¸+20]. The algorithm is formally described in Alg. 1 +and is similar to the one in [Sur+22]. By Theorem 2.1, Alg. 1 is almost optimal. +Its analysis is also similar to the one in [Sur+22]. +Correctness +Let us prove that Alg. 1 is 1/3-correct. First we show that ˆ +J = +1 +N +�N +t=1 Jt is +an unbiased estimator of JN . For this, we prove the following lemma relating the Choi state to +the average of the tensor product of a random rank-1 projector and its image by the quantum +channel. +14 + +Algorithm 1 Learning a quantum channel in the diamond distance using ancilla-free +independent measurements. +N = O(d3 +ind3 +out log(dindout)/ε2). +for t = 1 : N do +Sample two independent copies of Haar distributed unitaries V +∼ Haar(din) and +U ∼ Haar(dout) . +Let |v⟩ = V |0⟩ be a haar distributed vector. +Take the input states ρt = |v⟩⟨v| and σt = +I +din , the output states are respectively +N(|v⟩⟨v|) and N +� +I +din +� +. +Perform a measurement on N(|v⟩⟨v|) and N +� +I +din +� +using the POVM MU +:= +{U |i⟩⟨i| U †}i∈[dout] and observe it ∼ pU,V := {⟨i| U †N(|v⟩⟨v|)U |i⟩}i∈[dout] and jt ∼ qU := +{⟨i| U †N +� I +d +� +U |i⟩}i∈[dout]. +Define Jt := (din+1) |v⟩⟨v|T ⊗((dout+1)(U |it⟩⟨it| U †)−I)−I⊗((dout+1)(U |jt⟩⟨jt| U †)−I) +end for +Define the estimator ˆ +J = 1 +N +�N +t=1 Jt. +Find a valid Choi state JM such that ∥JM − ˆ +J ∥∞ ≤ +ε +2dindout . +return the quantum channel M corresponding to the Choi state JM. +Lemma 3.4. Let |φ⟩ be a Haar-distributed random vector. We have the following equality: +JN = (din + 1)E +� +|φ⟩⟨φ|T ⊗ N(|φ⟩⟨φ|) +� +− I ⊗ N +� I +din +� +. +Proof. We use the Kraus decomposition of the quantum channel N(ρ) = � +k AkρA† +K. We start +by writing the following expectation: +E (|φ⟩⟨φ| ⊗ N(|φ⟩⟨φ|)) = +� +k +E +� +|φ⟩⟨φ| ⊗ Ak |φ⟩⟨φ| A† +k +� += +� +k +I ⊗ AkE (|φ⟩⟨φ| ⊗ |φ⟩⟨φ|) I ⊗ A† +k +Let F be the flip operator F = �din +i,j=1 |ij⟩ ⟨ji|, if we take the transpose on the fist tensor we +obtain the unnormalized maximally entangled state: +F T1 = +din +� +i,j=1 +|i⟩ ⟨j|T ⊗ |j⟩ ⟨i| = +din +� +i,j=1 +|j⟩ ⟨i| ⊗ |j⟩ ⟨i| = din |Ψ⟩⟨Ψ| +where |Ψ⟩⟨Ψ| = +1 +din +�din +i,j=1 |ii⟩ ⟨jj| = +1 +din +�din +i,j=1 |i⟩ ⟨j| ⊗ |i⟩ ⟨j| is the maximally entangled state. +It is known that there is constants α and β such that: +E (|φ⟩⟨φ| ⊗ |φ⟩⟨φ|) = αI + βF. +Taking the trace we have the first relation 1 = αd2 +in+βdin, then taking the trace after multiplying +with F we obtain the second relation Tr(|φ⟩⟨φ|⊗|φ⟩⟨φ| F) = Tr(|φ⟩⟨φ| |φ⟩⟨φ|) = 1 = αdin +βd2 +in. +These relations imply α = β = +1 +din(din+1). Hence: +E (|φ⟩⟨φ| ⊗ |φ⟩⟨φ|) = +I + F +din(din + 1). +15 + +Replacing this expectation on the first expectation yields: +E +� +|φ⟩⟨φ|T ⊗ N(|φ⟩⟨φ|) +� += +� +k +E +� +|φ⟩⟨φ|T ⊗ Ak |φ⟩⟨φ| A† +k +� += +� +k +I ⊗ AkE +� +|φ⟩⟨φ|T ⊗ |φ⟩⟨φ| +� +I ⊗ A† +k += +1 +din(din + 1) +� +k +I ⊗ AkA† +k + +1 +din(din + 1) +� +k +I ⊗ Ak(din |Ψ⟩⟨Ψ|)I ⊗ A† +k += +1 +din(din + 1)I ⊗ N(I) + +1 +din + 1I ⊗ N(|Ψ⟩⟨Ψ|) += +1 +din(din + 1)I ⊗ N(I) + +1 +din + 1JN . +Then we compute another expectation: +Lemma 3.5. Let U ∼ Haar(d) and x ∼ pU,ρ := {⟨i| U †ρU |i⟩}i∈[d], we have +E +� +(d + 1)U |x⟩⟨x| U † − I +� += ρ +Proof. Since the equality is linear in ρ we can without loss of generality restrict ourselves to +a pure state ρ = |φ⟩⟨φ|. Now x ∼ {⟨i| U † |φ⟩⟨φ| U |i⟩}i∈[d] hence for k, l ∈ [d], by Weingarten +calculus: +EU,x∼pU,φ +� +⟨k| U |x⟩⟨x| U † |l⟩ +� += EU +� d +� +x=1 +⟨x| U † |φ⟩⟨φ| U |x⟩ ⟨k| U |x⟩⟨x| U † |l⟩ +� += EU +� d +� +x=1 +⟨x| U † |φ⟩⟨φ| U |x⟩ ⟨x| U † |l⟩ ⟨k| U |x⟩ +� += +d +� +x=1 +1 +d(d + 1) (δl,k + ⟨φ|l⟩⟨k|φ⟩) += +1 +(d + 1) (⟨k| I + |φ⟩⟨φ| |l⟩)) +Therefore +E +� +(d + 1)U |x⟩⟨x| U † − I +� += |φ⟩⟨φ| = ρ. +16 + +Using Lemma 3.4 and Lemma 3.5 we deduce: +E +� +ˆ +J +� += E (J1) += EV,U,it,jt +� +(din + 1) |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I) +� +− EV,U,it,jt +� +I ⊗ ((dout + 1)(U |jt⟩⟨jt| U †) − I) +� += EV +� +(din + 1) |v⟩⟨v|T ⊗ EU,it((dout + 1)(U |it⟩⟨it| U †) − I) +� +− +� +I ⊗ EU,jt((dout + 1)(U |jt⟩⟨jt| U †) − I) +� += EV +� +(din + 1) |v⟩⟨v|T ⊗ N(|v⟩⟨v|)) − I ⊗ N +� I +din +�� += JN . +So the estimator ˆ +J = 1 +N +�N +t=1 Jt is unbiased. It remains to show a concentration inequality for +the random variable ˆ +J so that we can estimate how much steps we need in order to achieve the +precision and confidence we aim to. For this, we use the matrix Bernstein inequality [Tro12]: +Theorem 3.6. [Tro12] Consider a sequence of n independent Hermitian random matrices +A1, . . . , An ∈ Cd×d. Assume that each Ai satisfies +E (Ai) = 0 +and +∥Ai∥∞ ≤ R as. +Let σ2 = ∥ �n +i=1 E +� +A2 +i +� +∥∞. Then for any t ≥ σ2 +R : +P +������ +n +� +i=1 +(Ai − E (Ai)) +����� +∞ +≥ t +� +≤ d exp +� +− 3t +8R +� +. +Moreover for any t ≤ σ2 +R : +P +������ +n +� +i=1 +(Ai − E (Ai)) +����� +∞ +≥ t +� +≤ d exp +� +− 3t2 +8σ2 +� +. +Let J = JN = E (Jt). We apply this theorem to the estimator ˆ +J − J = 1 +N +�N +t=1(Jt − J ). +Recall that +Jt = (din + 1) |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I) − I ⊗ ((dout + 1)(U |jt⟩⟨jt| U †) − I). +Let At = Jt−J +N +, we have proven that E (At) = +1 +N E (Jt − J ) = 0. Moreover +∥At∥∞ = 1 +N ∥Jt − J ∥∞ ≤ 1 +N (∥Jt∥∞ + ∥J ∥∞) ≤ 8dindout +N +:= R. +Besides +σ2 = +����� +N +� +t=1 +E +� +A2 +t +� +����� +∞ += 1 +N +��E +� +(J1 − J )2��� +∞ = 1 +N +��E +� +(J1)2��� +∞ + Θ +� 1 +N +� +. +Using the identity (a |φ⟩⟨φ| − I)2 = (a2 − 2a) |φ⟩⟨φ| + I, we have: +E +�� +I ⊗ ((dout + 1)(U |jt⟩⟨jt| U †) − I) +�2� += E +� +(I ⊗ ((d2 +out − 1)(U |jt⟩⟨jt| U †) + I) +� += E +� +(I ⊗ ((d2 +out − 1)(U |jt⟩⟨jt| U † − I/(dout + 1)) + dout I) +� += (dout − 1)I ⊗ N(I/din) + dout I ⊗ I +17 + +has an operator norm at most O(dout) so we can focus on the first term in the definition +of J1 which has the main contribution. +We have using again the identity (a |φ⟩⟨φ| − I)2 = +(a2 − 2a) |φ⟩⟨φ| + I: +E +� +(din + 1) |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I) +�2 += (din + 1)2E +� +|v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I)2� += (din + 1)2E +� +|v⟩⟨v|T ⊗ ((d2 +out − 1)(U |it⟩⟨it| U †) + I) +� += (dout − 1)(din + 1)(J + I ⊗ N(I/din)) + +�dout(din + 1)2 +din +� +I +which has an operator norm Θ(dindout). Therefore +σ2 = 1 +N +��E +� +J 2 +1 +��� +∞ + Θ +� 1 +N +� += Θ +�dindout +N +� +. +Since we have σ2 +R ≥ Ω(1) we can use the matrix-Bernstein inequality in the regime t = +ε +2dindout ≤ +O(1): +P +������ +N +� +t=1 +(At − E (At)) +����� +∞ +≥ +ε +2dindout +� +≤ dindout exp +� +− +3ε2 +8d2 +ind2 +outσ2 +� +≤ dindout exp +� +− CNε2 +d3 +ind3 +out +� +where C +> +0 is a universal constant. +Hence if N += +d3 +ind3 +out log(3dindout)/(Cε2) += +O +� +d3 +ind3 +out log(dindout)/ε2� +then with a probability at least 2/3 we have +∥ ˆ +J − JN ∥∞ = +����� +N +� +t=1 +(At − E (At)) +����� +∞ +≤ +ε +2dindout +. +This implies that ∥JM − JN ∥∞ ≤ +ε +dindout and finally ∥M − N∥⋄ ≤ ε by Lemma 3.2. This +finishes the proof of the correctness of Alg 1. +4 +Conclusion and open questions +In this work, we find the optimal complexity of quantum process tomography using non- +adaptive incoherent measurements. Furthermore, we show that ancilla-assisted strategies cannot +outperform their ancilla-free counterparts contrary to Pauli channel tomography [Che+22a]. +Still, many questions remain open. First, it is known that adaptive strategies have the same +complexity as non-adaptive ones for state tomography [Che+22c], could adaptive strategies +overcome non-adaptive ones for quantum process tomography? +Secondly, can entangled +strategies exploit the symmetry and show a polynomial (in din, dout) speedup as they do for +state tomography [Haa+16]? Lastly, what would be the potential improvements for simpler +problems such as testing identity to a fixed quantum channel or learning the expectations of +some given input states and observables? +18 + +References +[Bor26] +Max Born. “Zur Quantenmechanik der Stoßvorg¨ange”. In: Zeitschrift fur Physik +37.12 (Dec. 1926), pp. 863–867. doi: 10.1007/BF01397477. +[Fan61] +Robert M Fano. “Transmission of information: A statistical theory of communica- +tions”. In: American Journal of Physics 29.11 (1961), pp. 793–794. +[Hoe63] +Wassily Hoeffding. “Probability inequalities for sums of bounded random vari- +ables”. In: J. Amer. Statist. Assoc. 58 (1963), pp. 13–30. issn: 0162-1459. url: +http://links.jstor.org/sici?sici=0162-1459(196303)58:301%3C13:PIFSOB%3E2.0.CO;2-D&origin=MSN. +[Jam72] +Andrzej Jamio�lkowski. “Linear transformations which preserve trace and positive +semidefiniteness of operators”. In: Reports on Mathematical Physics 3.4 (1972), +pp. 275–278. +[Cho75] +Man-Duen Choi. “Completely positive linear maps on complex matrices”. In: Linear +algebra and its applications 10.3 (1975), pp. 285–290. +[CN97] +Isaac L Chuang and Michael A Nielsen. “Prescription for experimental determi- +nation of the dynamics of a quantum black box”. In: Journal of Modern Optics +44.11-12 (1997), pp. 2455–2467. +[PCZ97] +JF Poyatos, J Ignacio Cirac, and Peter Zoller. “Complete characterization of a +quantum process: the two-bit quantum gate”. In: Physical Review Letters 78.2 +(1997), p. 390. +[Leu00] +Debbie Wun Chi Leung. Towards robust quantum computation. stanford university, +2000. +[DP01] +GM D’Ariano and P Lo Presti. “Quantum tomography for measuring experimen- +tally the matrix elements of an arbitrary quantum operation”. In: Physical review +letters 86.19 (2001), p. 4195. +[OBr+04] +Jeremy L O’Brien et al. “Quantum process tomography of a controlled-NOT gate”. +In: Physical review letters 93.8 (2004), p. 080502. +[LZK08] +Daniel A Lidar, Paolo Zanardi, and Kaveh Khodjasteh. “Distance bounds on +quantum dynamics”. In: Physical Review A 78.1 (2008), p. 012308. +[MRL08] +Masoud Mohseni, Ali T Rezakhani, and Daniel A Lidar. “Quantum-process +tomography: Resource analysis of different strategies”. In: Physical Review A 77.3 +(2008), p. 032322. +[Bia+10] +Radoslaw C Bialczak et al. “Quantum process tomography of a universal entangling +gate implemented with Josephson phase qubits”. In: Nature Physics 6.6 (2010), +pp. 409–413. +[Yam+10] +T Yamamoto et al. “Quantum process tomography of two-qubit controlled-Z and +controlled-NOT gates using superconducting phase qubits”. In: Physical Review B +82.18 (2010), p. 184515. +[Tro12] +Joel A Tropp. “User-friendly tail bounds for sums of random matrices”. In: +Foundations of computational mathematics 12.4 (2012), pp. 389–434. +[MM+13] +Elizabeth Meckes, Mark Meckes, et al. “Spectral measures of powers of random +matrices”. In: Electronic communications in probability 18 (2013). +[Haa+16] +Jeongwan Haah et al. “Sample-optimal tomography of quantum states”. In: +Proceedings of the forty-eighth annual ACM symposium on Theory of Computing. +2016, pp. 913–925. +19 + +[JP16] +Anna Jenˇcov´a and Martin Pl´avala. “Conditions for optimal input states for +discrimination of quantum channels”. In: Journal of Mathematical Physics 57.12 +(2016), p. 122203. +[Nec+18] +Ion Nechita et al. “Almost all quantum channels are equidistant”. In: Journal of +Mathematical Physics 59.5 (2018), p. 052201. +[Wat18] +John Watrous. The theory of quantum information. Cambridge university press, +2018. +[FW20] +Steven T Flammia and Joel J Wallman. “Efficient estimation of Pauli channels”. +In: ACM Transactions on Quantum Computing 1.1 (2020), pp. 1–32. +[Gut¸+20] +Madalin Gut¸˘a et al. “Fast state tomography with optimal error bounds”. In: +Journal of Physics A: Mathematical and Theoretical 53.20 (2020), p. 204001. +[Che+22a] +Senrui Chen et al. “Quantum advantages for pauli channel estimation”. In: Physical +Review A 105.3 (2022), p. 032435. +[Che+22b] +Sitan Chen et al. “Exponential separations between learning with and without +quantum memory”. In: 2021 IEEE 62nd Annual Symposium on Foundations of +Computer Science (FOCS). IEEE. 2022, pp. 574–585. +[Che+22c] +Sitan Chen et al. “Tight bounds for state tomography with incoherent measure- +ments”. In: arXiv preprint arXiv:2206.05265 (2022). +[LN22] +Angus Lowe and Ashwin Nayak. “Lower bounds for learning quantum states with +single-copy measurements”. In: arXiv preprint arXiv:2207.14438 (2022). +[Sur+22] +Trystan Surawy-Stepney et al. “Projected least-squares quantum process tomogra- +phy”. In: Quantum 6 (2022), p. 844. +[FOF23] +Omar Fawzi, Aadil Oufkir, and Daniel Stilck Fran¸ca. “Lower Bounds on Learning +Pauli Channels”. In: arXiv preprint arXiv:2301.09192 (2023). +20 + diff --git a/JNFOT4oBgHgl3EQfxjTe/content/tmp_files/load_file.txt b/JNFOT4oBgHgl3EQfxjTe/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..5ecb6e7020902c344e014d6e67f84c2e706bb1c4 --- /dev/null +++ b/JNFOT4oBgHgl3EQfxjTe/content/tmp_files/load_file.txt @@ -0,0 +1,579 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf,len=578 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='12925v1 [quant-ph] 30 Jan 2023 Sample-Optimal Quantum Process Tomography with Non-Adaptive Incoherent Measurements Aadil Oufkir Univ Lyon, Inria, ENS Lyon, UCBL, LIP, Lyon, France Abstract How many copies of a quantum process are necessary and sufficient to construct an approximate classical description of it?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We extend the result of Surawy-Stepney, Kahn, Kueng, and Guta (2022) to show that ˜O(d3 ind3 out/ε2) copies are sufficient to learn any quantum channel Cdin×din → Cdout×dout to within ε in diamond norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Moreover, we show that Ω(d3 ind3 out/ε2) copies are necessary for any strategy using incoherent non-adaptive measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This lower bound applies even for ancilla-assisted strategies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We consider the problem of quantum process tomography which consists of approximating an arbitrary quantum channel–any linear map that preserves the axioms of quantum mechanics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This task is an important tool in quantum information processing and quantum control which has been performed in actual experiments (see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [OBr+04;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Bia+10;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Yam+10]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Given a quantum channel N : Cdin×din → Cdout×dout as a black box, a learner could choose the input state and send it through the unknown quantum channel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Then, it can only extract classical information by performing a measurement on the output state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It repeats this procedure at different steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' After collecting a sufficient amount of classical data, the goal is to return a quantum channel ˜ N satisfying: ∀ρ ∈ Cdin×din ⊗ Cdin×din : ∥id ⊗ (N − ˜ N)(ρ)∥1 ≤ ε∥ρ∥1 (1) with high probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In this work, we investigate the optimal complexity of non-adaptive strategies using incoherent measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' These strategies can only use one copy of the unknown channel at each step and must specify the input states and measurement devices before starting the learning procedure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Contribution The main contribution of this paper is to show that the optimal complexity of the quantum process tomography with non-adaptive incoherent measurements is ˜Θ(d3 ind3 out/ε2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' First, we prove a general lower bound of Ω(d3 ind3 out/ε2) on the number of incoherent measure- ments for every non-adaptive process learning algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' To do so, we construct an Ω(ε)- separated family of quantum channels close to the completely depolarizing channel of cardinal M = exp(Ω(d2 ind2 out)) by choosing random Choi states of a specific form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This family is used to encode a message from {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=', M}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' A process tomography algorithm can be used to decode this message with the same error probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Hence, the encoder and decoder should share at least Ω(d2 ind2 out) nats of information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' On the other hand, we show that the correlation between the encoder and decoder can only increase by at most O(ε2/dindout) nats after each measure- ment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Note that the naive upper bound on this correlation is O(ε2), we obtain an improvement by a factor dindout by exploiting the randomness in the construction of the quantum channel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This result is stated in Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Next, we show that the process tomography algorithm of 1 [Sur+22] can be generalized to approximate an unknown quantum channel to within ε in the diamond norm (1) using a number of incoherent measurements ˜O(d3 ind3 out/ε2) (Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For this, we relate the diamond norm between two quantum channels and the operator norm be- tween their corresponding Choi states which improves on the usual inequality with the 1-norm: ∥M∥⋄ ≤ din∥JM∥1 (see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [JP16]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Related work The first works on process tomography including [CN97;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' PCZ97] follow the strategy of learning the quantum states images of a complete set of basis states then obtaining the quantum channel by an inversion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The problem of state tomography using incoherent measurements is fully understood even for adaptive strategies [Haa+16;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Gut¸+20;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' LN22;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Che+22c]: the optimal complexity is Θ(d3/ε2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' So, learning a quantum channel can be done using O(d2 ind3 out) measurements, but this complexity doesn’t take into account the accumulation of errors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The same drawback can be seen in the resource analysis of different strategies by [MRL08].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Another reductive approach is to use the Choi–Jamio�lkowski isomorphism [Cho75;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Jam72] to reduce the process tomography to state tomography with a higher dimension [Leu00;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' DP01].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' However, this requires an ancilla and only implies a sub- optimal upper bound O((dindout)3/(ε/din)2) = O(d5 ind3 out/ε2) for learning in the diamond norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Sur+22] propose an algorithm for estimating the Choi state in the 2 norm that requires only ˜O(d4/ε2) ancilla-free incoherent measurements (when din = dout = d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This article generalizes this result to the diamond norm and general input/output dimensions and shows that this algorithm is optimal up to a logarithmic factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' A special case of quantum process tomography is learning Pauli channels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' These channels have weighted Pauli matrices as Kraus operators and can be learned in diamond norm using ˜O(d3/ε2) measurements [FW20] (here din = dout = d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Furthermore, it is shown that Ω(d3/ε2) are necessary for any non-adaptive strategy [FOF23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' While the techniques of the lower bound of this article are similar to the one in [FOF23], we obtain here a larger lower bound because, in general, we are not restricted to weighted Pauli matrices in the Kraus operators and these latter are implicitly chosen at random.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1 Preliminaries We consider quantum channels of input dimension din and output dimension dout.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We use the notation [d] := {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=', d}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We adopt the bra-ket notation: a column vector is denoted |φ⟩ and its adjoint is denoted ⟨φ| = |φ⟩†.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' With this notation, ⟨φ|ψ⟩ is the dot product of the vectors φ and ψ and, for a unit vector |φ⟩ ∈ Sd, |φ⟩⟨φ| is the rank-1 projector on the space spanned by the vector φ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The canonical basis {ei}i∈[d] is denoted {|i⟩}i∈[d] := {|ei⟩}i∈[d].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' A quantum state is a positive semi-definite Hermitian matrix of trace 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' A (din, dout)-dimensional quantum channel is a map N : Cdin×din → Cdout×dout of the form N(ρ) = � k AkρA† k where the Kraus operators {Ak}k∈K ∈ � Cdout×din�K satisfy � k∈K A† kAk = Idin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For instance, the identity map id(ρ) = ρ admits the Kraus operator {I} and the completely depolarizing channel D(ρ) = Tr(ρ) I dout admits the Kraus operators � 1 √dout |i⟩ ⟨j| � j∈[din],i∈[dout].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' A map N is a quantum channel if, and only if, it is: completely positive: for all ρ ≽ 0, id ⊗ N(ρ) ≽ 0 and trace preserving: for all ρ, Tr(N(ρ)) = Tr(ρ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We define the diamond distance between two quantum channels N and M as the diamond 2 norm of their difference: d⋄(N, M) := max ρ ∥id ⊗ (N − M)(ρ)∥1 where the maximization is over quantum states and the Schatten p-norm of a matrix M is defined as ∥M∥p p = Tr �√ M †M p� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The diamond distance can be thought of as a worst- case distance, while the average case distance is given by the Hilbert-Schmidt or Schatten 2-norm between the corresponding Choi states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We define the Choi state of the channel N as JN = id ⊗ N(|Ψ⟩⟨Ψ|) ∈ Cdin×din ⊗ Cdout×dout where |Ψ⟩ = 1 √din �din i=1 |i⟩ ⊗ |i⟩ is the maximally entangled state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' However, to have comparable distances, we will normalize the 2-norm which is equivalent to unnormalizing the maximally entangled state and we define the 2-distance as follows: d2(N, M) := din∥JN − JM∥2 = ∥id ⊗ (N − M)(din |Ψ⟩⟨Ψ|)∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This is a valid distance since the map J : N �→ id ⊗ N(|Ψ⟩⟨Ψ|) is an isomorphism called the Choi–Jamio�lkowski isomorphism [Cho75;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Jam72].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Note that J should be positive semi-definite and satisfy Tr2(J ) = I din to be a valid Choi state (corresponding to a quantum channel).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We consider the channel tomography problem which consists of learning a quantum channel N in the diamond distance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Given a precision parameter ε > 0, the goal is to construct a quantum channel ˜ N satisfying with at least a probability 2/3: d⋄(N, ˜ N ) ≤ ε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' An algorithm A is 1/3-correct for this problem if it outputs a quantum channel ε-close to N with a probability of error at most 1/3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We choose to learn in the diamond distance because it characterizes the minimal error probability to distinguish between two quantum channels when auxiliary systems are allowed [Wat18].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The learner can only extract classical information from the unknown quantum channel N by performing a measurement on the output state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Throughout the paper, we only consider unentangled or incoherent measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' That is, the learner can only measure with a d (or d × d)-dimensional measurement device.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Precisely, a d-dimensional measurement is defined by a POVM (positive operator-valued measure) with a finite number of elements: this is a set of positive semi-definite matrices M = {Mx}x∈X acting on the Hilbert space Cd and satisfying � x∈X Mx = I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Each element Mx in the POVM M is associated with the outcome x ∈ X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The tuple {Tr(ρMx)}x∈X is non-negative and sums to 1: it thus defines a probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Born’s rule [Bor26] says that the probability that the measurement on a quantum state ρ using the POVM M will output x is exactly Tr(ρMx).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Depending on whether an auxiliary system is allowed to be used, we distinguish two types of strategies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' M1 ρ1 N x1 M2 ρ2 N x2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' MN ρN N xN Figure 1: Illustration of an ancilla-free independent strategy for quantum process tomography.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Ancilla-free strategies At each step t, the learner would choose an input din-dimensional state ρt ∈ Cdin×din and a dout-dimensional measurement device Mt = {M t x}x∈Xt ∈ (Cdout×dout)Xt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It thus sees the outcome xt ∈ Xt with a probability Tr(N(ρt)M t xt) (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 3 Ancilla-assisted strategies At each step t, the learner would choose an input d × din- dimensional state ρt ∈ Cd×d ⊗ Cdin×din and a d × dout-dimensional measurement device Mt = {M t x}x∈Xt ∈ (Cd×d ⊗ Cdout×dout)Xt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It thus sees the outcome xt ∈ Xt with a probability Tr(id ⊗ N(ρt)M t xt) (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Note that ancilla-assisted strategies were proven to provide an ρ1 M1 x1 N ρ2 M2 x2 N .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' ρN MN xN N Figure 2: Illustration of an ancilla-assisted independent strategy for quantum process tomogra- phy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' exponential (in the number of qubits n = log2(d)) advantage over ancilla-free strategies for some problems [Che+22a;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Che+22b].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' However, in this work, we show that ancilla-assisted strategies cannot overcome ancilla-free strategies for process tomography.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Finally, we only consider non- adaptive strategies: the input states and measurement devices should be chosen before starting the learning procedure and thus cannot depend on the observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Given two random variables X and Y taking values in the sets [d] and [d′] respectively, the mutual information between X and Y is the Kullback Leibler divergence between the joint distribution P(X,Y ) and the product distribution PX × PY : I(X : Y ) = d � i=1 d′ � j=1 P (X = i, Y = j) log � P (X = i, Y = j) P (X = i) P (Y = j) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' All the logs of this paper are taken in base e and the information is measured in “nats”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 2 Lower bound In this section, we would like to investigate the intrinsic limitations of learning quantum channels using incoherent measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' To avoid repetition, we consider only ancilla-assisted strategies since they contain ancilla-free strategies as a special case: one can map every din-dimensional input state ρ to the d×din-dimensional input state ˜ρ = I d ⊗ρ and every dout-dimensional POVM M = {Mx}x∈X to the d × dout-dimensional POVM ˜ M = {Id ⊗ Mx}x∈X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Mainly, we prove the following theorem: Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let ε ≤ 1/16 and dout ≥ 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Any non-adaptive ancilla-assisted algorithm for process tomography in diamond distance requires N = Ω �d3 ind3 out ε2 � incoherent measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 4 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For the proof, we use the construction of the Choi state: JU = I dindout + ε dindout (U + U †) − ε dindout Tr2(U + U †) ⊗ I dout where U ∼ Haar(dindout).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' JU is Hermitian and satisfies Tr2(J ) = I din .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Moreover, JU ≽ 0 for ε ≤ 1/4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Indeed, U is a unitary so it has an operator norm 1 thus ∥U + U †∥∞ ≤ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Besides, ∥Tr2(U + U †)⊗ I dout ∥∞ = 1 dout ∥Tr2(U + U †)∥∞ ≤ maxi ∥I⊗ ⟨i| (U + U †)I⊗ |i⟩ ∥∞ ≤ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We claim that: Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We can construct an ε/2-separated (according to the diamond distance) family {Nx}x∈[M] of cardinal M = exp(Ω(d2 ind2 out)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It is sufficient to show that for U, V ∼ Haar(dindout): P (∥JU − JV ∥1 ≤ ε/2) ≤ exp � −Ω(d2 ind2 out) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' because, once this concentration inequality holds, we can choose our family randomly, and by the union bound, it will be ε/2-separated with an overwhelming probability (1−exp � −Ω(d2 ind2 out) � ) using the inequality d⋄(NU, NV ) ≥ ∥JU − JV ∥1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' First, let us lower bound the expected value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' E (∥JU − JV ∥1) ≥ ε dindout E � ∥U + U † − V − V †∥1 � − ε dind2 out E � ∥Tr2(U + U † − V − V †) ⊗ I∥1 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' On one hand, we can upper bound the second expectation using the triangle and the Cauchy- Schwartz inequalities: E � ∥Tr2(U + U † − V − V †) ⊗ I∥1 � ≤ 4E (∥Tr2(U) ⊗ I∥1) ≤ 4 � dindoutE (∥Tr2(U) ⊗ I∥2) ≤ 4 � dindout � E (Tr(Tr2(U)Tr2(U †) ⊗ I)) = 4 � dindout � dout � � � � �E \uf8eb \uf8ed� i � k,l ⟨i| ⊗ ⟨k| UI ⊗ |k⟩ ⟨l| U † |i⟩ ⊗ |l⟩ \uf8f6 \uf8f8 = 4 � dindout � dout � � � � �E \uf8eb \uf8ed din � i=1 dout � k,l=1 dinδk,l dindout \uf8f6 \uf8f8 = 4dindout.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' On the other hand, we can lower bound the first expectation using H¨older’s inequality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' E � ∥U + U † − V − V †∥1 � ≥ � (E (Tr(U + U † − V − V †)2))3 E (Tr(U + U † − V − V †)4) ≥ � (4dindout)3 16dindout = 2dindout.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Therefore: E (∥JU − JV ∥1) ≥ ε dindout E � ∥U + U † − V − V †∥1 � − 4ε dind2 out E (∥Tr2U ⊗ I∥1) ≥ 2ε − 4ε dout ≥ ε for dout ≥ 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 5 Now, we claim that the function (U, V ) �→ ∥JU − JV ∥1 is 8ε √dindout -Lipschitz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Indeed, we have ∥Tr2(X) ⊗ I∥1 ≤ √dindout∥Tr2(X) ⊗ I∥2 = √dindout∥Tr2(X)∥2 ≤ √dindoutdout∥X∥2 where the last inequality can be found in [LZK08].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Therefore,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' by letting X = U − U ′ and Y = V − V ′ and using the triangle inequality we obtain: |∥JU − JV ∥1 − ∥JU′ − JV ′∥1| ≤ 2ε dindout � ∥X∥1 + ∥Y ∥1 + ����Tr2(X) ⊗ I dout ���� 1 + ����Tr2(Y ) ⊗ I dout ���� 1 � ≤ 2√dindoutε dindout (∥U − U ′∥2 + ∥V − V ′∥2) + 2√dindoutdoutε dind2 out (∥U − U ′∥2 + ∥V − V ′∥2) ≤ 8ε √dindout ∥(U,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' V ) − (U ′,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' V ′)∥2 (Cauchy-Schwartz) so by the concentration inequality for Lipschitz functions of Haar measure [MM+13]: P (∥JU − JV ∥1 ≤ ε/2) ≤ P (∥JU − JV ∥1 − E (∥JU − JV ∥1) ≤ −ε/2) ≤ exp � − dindoutε2 48 × 64ε2/dindout � = exp � −Ω(d2 ind2 out) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Now, we use this ε/2-separated family of quantum channels {Nx}x∈[M] (corresponding to the Choi states {Jx}x∈[M] found in Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='2) to encode a uniformly random message X ∼ Uniform([M]) by the map X �→ NX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Using a learning algorithm for process tomography with precision ε/4 and an error probability at most 1/3, a decoder Y can find X with the same error probability because the family {Nx}x∈[M] is ε/2-separated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' By Fano’s inequality, the encoder and decoder should share at least Ω(log(M)) nats of information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Fan61] We have I(X : Y ) ≥ 2/3 log(M) − log(2) ≥ Ω(d4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The remaining part of the proof is to upper bound this mutual information in terms of the number of measurements N, the dimensions din, dout, and the precision parameter ε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Intuitively, the mutual information, after a few measurements, is very small and then it increases when the number of measurements increases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' To make this intuition formal, let N be a number of measurements sufficient for process tomography and let (I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN) be the observations of the learning algorithm, we apply first the data processing inequality to relate the mutual information between the encoder and the decoder with the mutual information between the uniform random variable X and the observations (I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN): I(X : Y ) ≤ I(X : I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Then we apply the chain rule for the mutual information: I(X : I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN) = N � t=1 I(X : It|I≤t−1) where we use the notation I≤t = (I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , It) and I(X : It|I≤t−1) is the conditional mutual information between X and It given I≤t−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' A learning algorithm A would choose the input 6 states {ρt}t∈[N] and measurement devices {Mt}t∈[N] which can be chosen to have the form Mt = {µt i |φt i⟩⟨φt i|}i∈It where µt i ≥ 0 and ⟨φt i|φt i⟩ = 1 for all t, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Using Jensen’s inequality, we can prove the following upper bound on the conditional mutual information: Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For x ∈ [M], let Mx = Nx − D where D(ρ) = Tr(ρ) I dout is the completely depolarizing channel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We have for all t ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , N}: I(X : It|I≤t−1) ≤ 3 M � i∈It,x∈[M] µt i � φt i �� id ⊗ D(ρt) ��φt i � �⟨φt i| id ⊗ Mx(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let t ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , N} and x ∈ [M].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let i = (i1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , it) ∈ (I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , It), we can express the joint probability p of (X, I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , It) as follows: p(x, i1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , it) = 1 M t� k=1 µk ik � φk ik �� id ⊗ Nx(ρk) ��φk ik � We can remark that, for all 1 ≤ k ≤ t: p(x, i≤k) = µk ik � φk ik �� id ⊗ Nx(ρk) ��φk ik � p(x, i≤k−1) = µk ik � φk ik �� id ⊗ D(ρk) ��φk ik � (1 + Φk x,ik)p(x, i≤k−1) where Φk x,ik = ⟨φk ik|id⊗Mx(ρk)|φk ik⟩ � φk ik ���id⊗D(ρk) ���φk ik � because D + Mx = Nx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' So,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' the ratio of conditional probabilities can be written as: p(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' it|i≤t−1) p(x|i≤t−1)p(it|i≤t−1) = p(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t)p(i≤t−1) p(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t−1)p(i≤t) = µt it � φt it �� id ⊗ D(ρt) ��φt it � (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t−1)p(i≤t−1) p(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t−1) � y p(y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t) = µt it � φt it �� id ⊗ D(ρt) ��φt it � (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(i≤t−1) � y p(y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t) = µt it � φt it �� id ⊗ D(ρt) ��φt it � (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(i≤t−1) � y µt it � φt it �� id ⊗ D(ρt) ��φt it � (1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t−1) = (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(i≤t−1) � y(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' i≤t−1) = (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � y(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)p(y|i≤t−1) Therefore by Jensen’s inequality: I(X : It|I≤t−1) = E � log � p(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' it|i≤t−1) p(x|i≤t−1)p(it|i≤t−1) �� = E � log � (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � y p(y|i≤t−1)(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) �� ≤ E � log(1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) − � y p(y|i≤t−1) log(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � = E � log(1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � − � y E � p(y|i≤t−1) log(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 7 The first term can be upper bounded using the inequality log(1 + x) ≤ x verified for all x ∈ (−1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' ∞): E � log(1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � = Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p log(1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) ≤ Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼pΦt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it = Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤tΦt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it = Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤t−1 � it µt it � φt it �� id ⊗ D(ρt) ��φt it � (1 + Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it = Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤t−1 � it µt it � φt it �� id ⊗ D(ρt) ��φt it � (Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 = 1 M M � x=1 � it µt it � φt it �� id ⊗ D(ρt) ��φt it � (Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 because � it µt it � φt it �� id⊗D(ρt) ��φt it � Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it = Tr(id⊗Mx(ρt)) = Tr(id⊗Nx(ρt))−Tr(id⊗D(ρt)) = Tr(ρt) − Tr(ρt) = 0 and we use the condition that the algorithm is non-adaptive in the last line.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' On the other hand, the second term can be upper bounded using the inequality − log(1 + x) ≤ −x + x2/2 verified for all x ∈ (−1/2, ∞).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let λt it = µt it � φt it �� id ⊗ D(ρt) ��φt it � ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' we have : E � − � y p(y|i≤t−1) log(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) � = � y Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼pp(y|i≤t−1)(− log)(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) = � y Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤tp(y|i≤t−1)(− log)(1 + Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it) ≤ � y Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤tp(y|i≤t−1)(−Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it + (Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2/2) ≤ � y Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤t−1p(y|i≤t−1) � it λt it((Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 + (Φt y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2) = 2 � y Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤t−1p(y|i≤t−1) � it λt it(Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 = 2Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤t−1 � it µt it � φt it �� id ⊗ D(ρt) ��φt it � (Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 = 2Ex,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i∼p≤t−1 � it λt it(Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 = 2 1 M M � x=1 � it µt it � φt it �� id ⊗ D(ρt) ��φt it � (Φt x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='it)2 where we use the condition that the algorithm is non-adaptive in the last line.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Since the conditional mutual information is upper bounded by the sum of these two terms, the upper bound on the conditional mutual information follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It remains to approximate every mean 1 M �M x=1 by the expectation EU.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 8 Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We have with at least a probability 9/10: 1 M � t,i,x µt i � φt i �� id ⊗ D(ρt) ��φt i � �⟨φt i| id ⊗ Mx(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 ≤ � t,i µt i � φt i �� id ⊗ D(ρt) ��φt i � EU �⟨φt i| id ⊗ MU(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 + 16Nε2 � log(10) M .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Denote by f t x the function |φ⟩ �→ ⟨φ|id⊗Mx(ρt)|φ⟩2 ⟨φ|id⊗D(ρt)|φ⟩2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We claim that the functions f t x are bounded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Indeed,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' we can write ρt = � i λi |ψi⟩⟨ψi| and every |ψi⟩ can be written as |ψi⟩ = Ai ⊗ I |Ψ⟩ so for a unit vector |φ⟩,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' we have: f t x(|φ⟩) = ⟨φ| id ⊗ Mx(ρt) |φ⟩2 ⟨φ| id ⊗ D(ρt) |φ⟩2 = 4ε2 � ⟨φ| � i λi(Ai ⊗ I) � Ux − Tr2(Ux) ⊗ I dout � (A† i ⊗ I) |φ⟩ �2 ⟨φ| � i λiAiA† i ⊗ I |φ⟩2 ≤ 4ε2 ⟨φ| � i λi(Ai ⊗ I)(A† i ⊗ I) |φ⟩2 ���Ux − Tr2(Ux) ⊗ I dout ��� 2 ∞ ⟨φ| � i λiAiA† i ⊗ I |φ⟩2 ≤ 16ε2 ⟨φ| � i λi(Ai ⊗ I)(A† i ⊗ I) |φ⟩2 ⟨φ| � i λiAiA† i ⊗ I |φ⟩2 = 16ε2 where we used that ∥Ux∥∞ = 1 and ∥Tr2(Ux)∥∞ ≤ dout for a unitary Ux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' But we have � i µt i ⟨φt i| id ⊗ D(ρt) |φt i⟩ = Tr(id ⊗ D(ρt)) = 1 so for all x ∈ [M]: � t,i µt i � φt i �� id ⊗ D(ρt) ��φt i � �⟨φt i| id ⊗ Mx(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 ≤ 16Nε2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Therefore, by Hoeffding’s inequality [Hoe63] and the union bound, we have with a probability at least 9/10: 1 M � x,t,i µt i � φt i �� id ⊗ D(ρt) ��φt i � �⟨φt i| id ⊗ Mx(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 ≤ � t,i µt i � φt i �� id ⊗ D(ρt) ��φt i � EU �⟨φt i| id ⊗ MU(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 + 16Nε2 � log(10) M .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 9 These two Lemmas 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='4, 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='5 imply: I(X : I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' IN) = N � t=1 I(X : It|I≤t−1) ≤ 3 M � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i µt i � φt i �� id ⊗ D(ρt) ��φt i � E �⟨φt i| id ⊗ Mx(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 ≤ 3 � t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i µt i � φt i �� id ⊗ D(ρt) ��φt i � E �⟨φt i| id ⊗ MU(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 + 48Nε2 � log(10) M ≤ 3N sup t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='i E ��⟨φt i| id ⊗ MU(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2� + 48Nε2 � log(10) M (2) where we used that fact that for all t ∈ [N]: � i µt i ⟨φt i| id⊗D(ρt) |φt i⟩ = Tr(id⊗D(ρt)) = Tr(ρt) = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The error probability 1/10 of this approximation can be absorbed in the construction above by asking the unitaries {Ux}x∈[M] not only to satisfy the separability condition, but also to satisfy the inequalities in Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='5: 1 M � t,i,x µt i � φt i �� id ⊗ D(ρt) ��φt i � �⟨φt i| id ⊗ Mx(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 ≤ � t,i µt i � φt i �� id ⊗ D(ρt) ��φt i � EU �⟨φt i| id ⊗ MU(ρt) |φt i⟩ ⟨φt i| id ⊗ D(ρt) |φt i⟩ �2 + 48Nε2 � log(10) M .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Now fix t ∈ [N], it ∈ It and |φ⟩ = ��φt it � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Recall that we can write ρt = � i λi |ψi⟩⟨ψi|, the maximally entangled state is denoted |Ψ⟩ = 1 √din �din i=1 |ii⟩ and every |ψi⟩ can be written as |ψi⟩ = Ai ⊗ I |Ψ⟩ so: id ⊗ D(ρt) = � i λi(id ⊗ D)(Ai ⊗ I |Ψ⟩⟨Ψ| A† i ⊗ I) = � i λi(Ai ⊗ I)id ⊗ D(|Ψ⟩⟨Ψ|)(A† i ⊗ I) = � i λi(Ai ⊗ I) I dindout (A† i ⊗ I) = � i λiAiA† i din ⊗ I dout .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' (3) 10 On the other hand, using the notation V = U − Tr2(U) ⊗ I dout , we can write: id ⊗ M(ρt) = � i λiid ⊗ M(Ai ⊗ I |Ψ⟩⟨Ψ| A† i ⊗ I) = � i λi(Ai ⊗ I)id ⊗ (N − D)(|Ψ⟩⟨Ψ|)(A† i ⊗ I) = � i λi(Ai ⊗ I) � JN − I dindout � (A† i ⊗ I) = ε dindout � i λiAi ⊗ I � U + U † − Tr2(U + U †) ⊗ I dout � A† i ⊗ I) = ε dindout � i λi � (Ai ⊗ I)V (A† i ⊗ I) + (Ai ⊗ I)V †(A† i ⊗ I) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' By Ineq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 2, we need to control the expectation EU ⟨φ| id ⊗ MU(ρt) |φ⟩2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' First, we replace id ⊗ M(ρt) with the latter expression, then we apply the inequality (x + y)2 ≤ 2x2 + 2y2 to separate the terms involving U and the terms involving Tr2(U).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The first term can be computed and bounded as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 4ε2 d2 ind2 out E \uf8eb \uf8ed �����⟨φ| �� i λi(Ai ⊗ I)U(A† i ⊗ I) � |φ⟩ ����� 2\uf8f6 \uf8f8 = 4ε2 d2 ind2 out � i,j λiλj dindout ���Tr � A† i ⊗ I |φ⟩ ⟨φ| Aj ⊗ I ���� 2 (CS) ≤ 4ε2 d2 ind2 out � i,j λiλj dindout ⟨φ| AiA† i ⊗ I |φ⟩ ⟨φ| AjA† j ⊗ I |φ⟩ = 4ε2 d3 ind3 out � ⟨φ| � i λiAiA† i ⊗ I |φ⟩ �2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' (4) Let’s move to the second term which involves the partial trace.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let Mij = (A† i ⊗I) |φ⟩⟨φ| (Aj ⊗I).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 4ε2 d2 ind2 out E \uf8eb \uf8ed �����⟨φ| � i λi(Ai ⊗ I) � Tr2(U) ⊗ I dout � (A† i ⊗ I) |φ⟩ ����� 2\uf8f6 \uf8f8 = 4ε2 d2 ind4 out � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j λiλjE � Tr � (Tr2(U) ⊗ I) Mi,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j � Tr2(U †) ⊗ I � M † ij) �� = 4ε2 d2 ind4 out � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j λiλj din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='z,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='t=1 dout � k,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='l=1 E � ⟨xk| U |yk⟩ ⟨zl|U † |tl⟩ � Tr � � |y⟩ ⟨x| ⊗ IMij |t⟩ ⟨z| ⊗ IM † ij � � = 4ε2 d3 ind5 out � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j λiλj din � x=t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=z=1 dout � k=l=1 Tr � � |y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † ij � � = 4ε2 d3 ind4 out � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j λiλj din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 Tr � � |y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † ij � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 11 To control the latter expression, we write |φ⟩ = B†⊗I |Ψ⟩ so that Mi,j = (A† i ⊗I) |φ⟩⟨φ| (Aj⊗I) = (A† iB†⊗I) |Ψ⟩⟨Ψ| (BAj⊗I).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Using the property of the maximally entangled state ⟨Ψ| M⊗I |Ψ⟩ = 1 din Tr(M) we obtain: din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 Tr � � |y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † ij � = din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 Tr � |y⟩ ⟨x| ⊗ I(A† iB† ⊗ I) |Ψ⟩⟨Ψ| (BAj ⊗ I) |x⟩ ⟨y| ⊗ I(A† jB† ⊗ I) |Ψ⟩⟨Ψ| (BAi ⊗ I) � = din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 ⟨Ψ| (BAj ⊗ I) |x⟩ ⟨y| ⊗ I(A† jB† ⊗ I) |Ψ⟩ ⟨Ψ| (BAi ⊗ I)† |y⟩ ⟨x| ⊗ I(A† iB† ⊗ I) |Ψ⟩ = 1 d2 in din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 Tr(BAj |x⟩ ⟨y| A† jB†)Tr(BAi |y⟩ ⟨x| A† iB†) = 1 d2 in din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 ⟨y| A† jB†BAj |x⟩ ⟨x| A† iB†BAi |y⟩ = 1 d2 in Tr � A† jB†BAjA† iB†BAi � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' On the other hand, we can write ⟨φ| � i λiAiA† i ⊗ I |φ⟩ = ⟨Ψ| � i λiBAiA† iB† ⊗ I |Ψ⟩ = 1 din Tr �� i λiA† iB†BAi � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Note that the matrix � i λiA† iB†BAi is positive semi-definite so: � i,j λiλj 1 d2 in Tr � A† jB†BAjA† iB†BAi � = 1 d2 in Tr �� i λiA† iB†BAi �2 ≤ � 1 din Tr �� i λiA† iB†BAi � �2 = ⟨φ| � i λiAiA† i ⊗ I |φ⟩2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Hence 4ε2 d2 ind2 out E \uf8eb \uf8ed �����⟨φ| � i λi(Ai ⊗ I) � Tr2(U) ⊗ I dout � (A† i ⊗ I) |φ⟩ ����� 2\uf8f6 \uf8f8 = 4ε2 d3 ind4 out � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j λiλj din � x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='y=1 Tr � � |y⟩ ⟨x| ⊗ IMij |x⟩ ⟨y| ⊗ IM † ij � � = 4ε2 d3 ind4 out � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j λiλj 1 d2 in Tr � A† jB†BAjA† iB†BAi � ≤ 4ε2 d3 ind4 out ⟨φ| � i λiAiA† i ⊗ I |φ⟩2 (5) Using the equality (3) and the two inequalities (4) and (5),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' we deduce: E ��⟨φ| id ⊗ MU(ρt) |φ⟩ ⟨φ| id ⊗ D(ρt) |φ⟩ �2� ≤ 8ε2 d3 ind3 out � ⟨φ| � i λiAiA† i ⊗ I |φ⟩ �2 ⟨φ| � i λiAiA† i din ⊗ I dout |φ⟩2 = 8ε2 dindout .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 12 Therefore using the inequality (2): I(X : I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN) = N � t=1 I(X : It|I≤t−1) ≤ 3N sup t,it E \uf8eb \uf8ed �� φt it �� id ⊗ MU(ρt) ��φt it � � φt it �� id ⊗ D(ρt) ��φt it � �2\uf8f6 \uf8f8 + 48Nε2 � log(10) M ≤ 24N ε2 dindout + 48Nε2 � log(10) M ≤ O � N ε2 dindout � because M = exp(Ω(d2 ind2 out)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' But from the data processing inequality and Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3, I(X : I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN) ≥ I(X : Y ) ≥ Ω(d2 ind2 out), we deduce that: O � N ε2 dindout � ≥ I(X : I1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , IN) ≥ Ω(d2 ind2 out).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Finally, the lower bound follows: N ≥ Ω �d3 ind3 out ε2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' To assess this lower bound, it is necessary to design an algorithm for quantum process tomography.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This will be the object of the following section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 3 Upper bound In this section, we propose an upper bound on the complexity of the quantum process tomography problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We generalize the algorithm proposed by [Sur+22] which is ancilla-free.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Sur+22] There is an ancilla-free process tomography algorithm that learns a quantum channel (of din = dout = d) in the distance d2 using only a number of measurements: N = O �d6 log(d) ε2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This algorithm proceeds by providing an unbiased estimator for the Choi state JN , then projecting this matrix to the space of Choi states (PSD and partial trace I/d) and finally by invoking the Choi–Jamio�lkowski isomorphism we obtain an approximation of the channel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This reduction from learning the Choi state in the operator norm to learning the quantum channel in the d2 distance uses mainly the inequality d2(N, M) = d∥JN − JM∥2 ≤ d2∥JN − JM∥∞ when din = dout = d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We generalize this result to the diamond norm and any input/output dimensions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For this we show the following inequality: Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let N1 and N2 be two quantum channels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We have: d⋄(N1, N2) ≤ dindout∥JN1 − JN2∥∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This inequality can also be obtained by applying the inequality (3) of [Nec+18] and the triangle inequality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We provide a simpler proof for completeness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 13 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Denote by M = N1 − N2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let |φ⟩ be a maximizing unit vector of the diamond norm, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=', ∥id ⊗ M(|φ⟩⟨φ|)∥1 = d⋄(N1, N2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We can write |φ⟩ = A ⊗ I |Ψ⟩ where |Ψ⟩ = 1 √din �din i=1 |ii⟩ is the maximally entangled state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' |φ⟩ has norm 1 so 1 din Tr(A†A) = ⟨Ψ| A†A ⊗ I |Ψ⟩ = ⟨φ|φ⟩ = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' On the other hand we can write d⋄(N1, N2) = ∥id ⊗ M(|φ⟩⟨φ|)∥1 = ∥I ⊗ M(A ⊗ Idin |Ψ⟩⟨Ψ| A† ⊗ Idin)∥1 = ∥(A ⊗ Idout)id ⊗ M(|Ψ⟩⟨Ψ|)(A† ⊗ Idout)∥1 = ∥(A ⊗ Idout)JM(A† ⊗ Idout)∥1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' JM is Hermitian so it can be written as : JM = � i λi |ψi⟩⟨ψi|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Using the triangle inequality, we obtain: ∥(A ⊗ Idout)JM(A† ⊗ Idout)∥1 = �����(A ⊗ Idout) � i λi |ψi⟩⟨ψi| (A† ⊗ Idout) ����� 1 ≤ � i |λi|∥(A ⊗ Idout) |ψi⟩⟨ψi| (A† ⊗ Idout)∥1 ≤ max i |λi| � i ∥(A ⊗ Idout) |ψi⟩⟨ψi| (A† ⊗ Idout)∥1 = ∥J ∥∞ � i Tr((A ⊗ Idout) |ψi⟩⟨ψi| (A† ⊗ Idout)) = ∥J ∥∞Tr(AA† ⊗ Idout) = dindout∥J ∥∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This Lemma shows that the diamond and 2 distances satisfy the same inequality with respect to the infinity norm between the Choi states when din = dout = d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Since the algorithm of [Sur+22] approximates first the Choi state in the infinity norm, we obtain the same upper bound for the diamond distance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For general dimensions, we obtain the following complexity: Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' There is a non-adaptive ancilla-free process tomography algorithm that learns a quantum channel in the distance d⋄ using only a number of measurements: N = O �d3 ind3 out log(dindout) ε2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This complexity was expected for process tomography with incoherent measurements since the complexity of state tomography with incoherent measurements is Θ � d3 ε2 � [Haa+16] and learning (din, dout)-dimensional channels can be thought of as learning states of dimension din × dout.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We believe that the log(dindout)-factor can be removed from the upper bound in Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3 using the techniques of [Gut¸+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The algorithm is formally described in Alg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1 and is similar to the one in [Sur+22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' By Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1, Alg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1 is almost optimal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Its analysis is also similar to the one in [Sur+22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Correctness Let us prove that Alg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1 is 1/3-correct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' First we show that ˆ J = 1 N �N t=1 Jt is an unbiased estimator of JN .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For this, we prove the following lemma relating the Choi state to the average of the tensor product of a random rank-1 projector and its image by the quantum channel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 14 Algorithm 1 Learning a quantum channel in the diamond distance using ancilla-free independent measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' N = O(d3 ind3 out log(dindout)/ε2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' for t = 1 : N do Sample two independent copies of Haar distributed unitaries V ∼ Haar(din) and U ∼ Haar(dout) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let |v⟩ = V |0⟩ be a haar distributed vector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Take the input states ρt = |v⟩⟨v| and σt = I din , the output states are respectively N(|v⟩⟨v|) and N � I din � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Perform a measurement on N(|v⟩⟨v|) and N � I din � using the POVM MU := {U |i⟩⟨i| U †}i∈[dout] and observe it ∼ pU,V := {⟨i| U †N(|v⟩⟨v|)U |i⟩}i∈[dout] and jt ∼ qU := {⟨i| U †N � I d � U |i⟩}i∈[dout].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Define Jt := (din+1) |v⟩⟨v|T ⊗((dout+1)(U |it⟩⟨it| U †)−I)−I⊗((dout+1)(U |jt⟩⟨jt| U †)−I) end for Define the estimator ˆ J = 1 N �N t=1 Jt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Find a valid Choi state JM such that ∥JM − ˆ J ∥∞ ≤ ε 2dindout .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' return the quantum channel M corresponding to the Choi state JM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let |φ⟩ be a Haar-distributed random vector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We have the following equality: JN = (din + 1)E � |φ⟩⟨φ|T ⊗ N(|φ⟩⟨φ|) � − I ⊗ N � I din � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We use the Kraus decomposition of the quantum channel N(ρ) = � k AkρA† K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We start by writing the following expectation: E (|φ⟩⟨φ| ⊗ N(|φ⟩⟨φ|)) = � k E � |φ⟩⟨φ| ⊗ Ak |φ⟩⟨φ| A† k � = � k I ⊗ AkE (|φ⟩⟨φ| ⊗ |φ⟩⟨φ|) I ⊗ A† k Let F be the flip operator F = �din i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j=1 |ij⟩ ⟨ji|,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' if we take the transpose on the fist tensor we obtain the unnormalized maximally entangled state: F T1 = din � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j=1 |i⟩ ⟨j|T ⊗ |j⟩ ⟨i| = din � i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j=1 |j⟩ ⟨i| ⊗ |j⟩ ⟨i| = din |Ψ⟩⟨Ψ| where |Ψ⟩⟨Ψ| = 1 din �din i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j=1 |ii⟩ ⟨jj| = 1 din �din i,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='j=1 |i⟩ ⟨j| ⊗ |i⟩ ⟨j| is the maximally entangled state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It is known that there is constants α and β such that: E (|φ⟩⟨φ| ⊗ |φ⟩⟨φ|) = αI + βF.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Taking the trace we have the first relation 1 = αd2 in+βdin, then taking the trace after multiplying with F we obtain the second relation Tr(|φ⟩⟨φ|⊗|φ⟩⟨φ| F) = Tr(|φ⟩⟨φ| |φ⟩⟨φ|) = 1 = αdin +βd2 in.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' These relations imply α = β = 1 din(din+1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Hence: E (|φ⟩⟨φ| ⊗ |φ⟩⟨φ|) = I + F din(din + 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 15 Replacing this expectation on the first expectation yields: E � |φ⟩⟨φ|T ⊗ N(|φ⟩⟨φ|) � = � k E � |φ⟩⟨φ|T ⊗ Ak |φ⟩⟨φ| A† k � = � k I ⊗ AkE � |φ⟩⟨φ|T ⊗ |φ⟩⟨φ| � I ⊗ A† k = 1 din(din + 1) � k I ⊗ AkA† k + 1 din(din + 1) � k I ⊗ Ak(din |Ψ⟩⟨Ψ|)I ⊗ A† k = 1 din(din + 1)I ⊗ N(I) + 1 din + 1I ⊗ N(|Ψ⟩⟨Ψ|) = 1 din(din + 1)I ⊗ N(I) + 1 din + 1JN .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Then we compute another expectation: Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let U ∼ Haar(d) and x ∼ pU,ρ := {⟨i| U †ρU |i⟩}i∈[d], we have E � (d + 1)U |x⟩⟨x| U † − I � = ρ Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Since the equality is linear in ρ we can without loss of generality restrict ourselves to a pure state ρ = |φ⟩⟨φ|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Now x ∼ {⟨i| U † |φ⟩⟨φ| U |i⟩}i∈[d] hence for k, l ∈ [d], by Weingarten calculus: EU,x∼pU,φ � ⟨k| U |x⟩⟨x| U † |l⟩ � = EU � d � x=1 ⟨x| U † |φ⟩⟨φ| U |x⟩ ⟨k| U |x⟩⟨x| U † |l⟩ � = EU � d � x=1 ⟨x| U † |φ⟩⟨φ| U |x⟩ ⟨x| U † |l⟩ ⟨k| U |x⟩ � = d � x=1 1 d(d + 1) (δl,k + ⟨φ|l⟩⟨k|φ⟩) = 1 (d + 1) (⟨k| I + |φ⟩⟨φ| |l⟩)) Therefore E � (d + 1)U |x⟩⟨x| U † − I � = |φ⟩⟨φ| = ρ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 16 Using Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='4 and Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='5 we deduce: E � ˆ J � = E (J1) = EV,U,it,jt � (din + 1) |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I) � − EV,U,it,jt � I ⊗ ((dout + 1)(U |jt⟩⟨jt| U †) − I) � = EV � (din + 1) |v⟩⟨v|T ⊗ EU,it((dout + 1)(U |it⟩⟨it| U †) − I) � − � I ⊗ EU,jt((dout + 1)(U |jt⟩⟨jt| U †) − I) � = EV � (din + 1) |v⟩⟨v|T ⊗ N(|v⟩⟨v|)) − I ⊗ N � I din �� = JN .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' So the estimator ˆ J = 1 N �N t=1 Jt is unbiased.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' It remains to show a concentration inequality for the random variable ˆ J so that we can estimate how much steps we need in order to achieve the precision and confidence we aim to.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' For this, we use the matrix Bernstein inequality [Tro12]: Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Tro12] Consider a sequence of n independent Hermitian random matrices A1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' , An ∈ Cd×d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Assume that each Ai satisfies E (Ai) = 0 and ∥Ai∥∞ ≤ R as.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let σ2 = ∥ �n i=1 E � A2 i � ∥∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Then for any t ≥ σ2 R : P ������ n � i=1 (Ai − E (Ai)) ����� ∞ ≥ t � ≤ d exp � − 3t 8R � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Moreover for any t ≤ σ2 R : P ������ n � i=1 (Ai − E (Ai)) ����� ∞ ≥ t � ≤ d exp � − 3t2 8σ2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let J = JN = E (Jt).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We apply this theorem to the estimator ˆ J − J = 1 N �N t=1(Jt − J ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Recall that Jt = (din + 1) |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I) − I ⊗ ((dout + 1)(U |jt⟩⟨jt| U †) − I).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Let At = Jt−J N , we have proven that E (At) = 1 N E (Jt − J ) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Moreover ∥At∥∞ = 1 N ∥Jt − J ∥∞ ≤ 1 N (∥Jt∥∞ + ∥J ∥∞) ≤ 8dindout N := R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Besides σ2 = ����� N � t=1 E � A2 t � ����� ∞ = 1 N ��E � (J1 − J )2��� ∞ = 1 N ��E � (J1)2��� ∞ + Θ � 1 N � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Using the identity (a |φ⟩⟨φ| − I)2 = (a2 − 2a) |φ⟩⟨φ| + I, we have: E �� I ⊗ ((dout + 1)(U |jt⟩⟨jt| U †) − I) �2� = E � (I ⊗ ((d2 out − 1)(U |jt⟩⟨jt| U †) + I) � = E � (I ⊗ ((d2 out − 1)(U |jt⟩⟨jt| U † − I/(dout + 1)) + dout I) � = (dout − 1)I ⊗ N(I/din) + dout I ⊗ I 17 has an operator norm at most O(dout) so we can focus on the first term in the definition of J1 which has the main contribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' We have using again the identity (a |φ⟩⟨φ| − I)2 = (a2 − 2a) |φ⟩⟨φ| + I: E � (din + 1) |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I) �2 = (din + 1)2E � |v⟩⟨v|T ⊗ ((dout + 1)(U |it⟩⟨it| U †) − I)2� = (din + 1)2E � |v⟩⟨v|T ⊗ ((d2 out − 1)(U |it⟩⟨it| U †) + I) � = (dout − 1)(din + 1)(J + I ⊗ N(I/din)) + �dout(din + 1)2 din � I which has an operator norm Θ(dindout).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Therefore σ2 = 1 N ��E � J 2 1 ��� ∞ + Θ � 1 N � = Θ �dindout N � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Since we have σ2 R ≥ Ω(1) we can use the matrix-Bernstein inequality in the regime t = ε 2dindout ≤ O(1): P ������ N � t=1 (At − E (At)) ����� ∞ ≥ ε 2dindout � ≤ dindout exp � − 3ε2 8d2 ind2 outσ2 � ≤ dindout exp � − CNε2 d3 ind3 out � where C > 0 is a universal constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Hence if N = d3 ind3 out log(3dindout)/(Cε2) = O � d3 ind3 out log(dindout)/ε2� then with a probability at least 2/3 we have ∥ ˆ J − JN ∥∞ = ����� N � t=1 (At − E (At)) ����� ∞ ≤ ε 2dindout .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This implies that ∥JM − JN ∥∞ ≤ ε dindout and finally ∥M − N∥⋄ ≤ ε by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' This finishes the proof of the correctness of Alg 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 4 Conclusion and open questions In this work, we find the optimal complexity of quantum process tomography using non- adaptive incoherent measurements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Furthermore, we show that ancilla-assisted strategies cannot outperform their ancilla-free counterparts contrary to Pauli channel tomography [Che+22a].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Still, many questions remain open.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' First, it is known that adaptive strategies have the same complexity as non-adaptive ones for state tomography [Che+22c], could adaptive strategies overcome non-adaptive ones for quantum process tomography?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Secondly, can entangled strategies exploit the symmetry and show a polynomial (in din, dout) speedup as they do for state tomography [Haa+16]?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Lastly, what would be the potential improvements for simpler problems such as testing identity to a fixed quantum channel or learning the expectations of some given input states and observables?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 18 References [Bor26] Max Born.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Zur Quantenmechanik der Stoßvorg¨ange”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Zeitschrift fur Physik 37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='12 (Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1926), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 863–867.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1007/BF01397477.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Fan61] Robert M Fano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Transmission of information: A statistical theory of communica- tions”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: American Journal of Physics 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='11 (1961), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 793–794.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Hoe63] Wassily Hoeffding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Probability inequalities for sums of bounded random vari- ables”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Statist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Assoc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 58 (1963), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 13–30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' issn: 0162-1459.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' url: http://links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='jstor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='org/sici?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='sici=0162-1459(196303)58:301%3C13:PIFSOB%3E2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='CO;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='2-D&origin=MSN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Jam72] Andrzej Jamio�lkowski.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Linear transformations which preserve trace and positive semidefiniteness of operators”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Reports on Mathematical Physics 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='4 (1972), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 275–278.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Cho75] Man-Duen Choi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Completely positive linear maps on complex matrices”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Linear algebra and its applications 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3 (1975), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 285–290.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [CN97] Isaac L Chuang and Michael A Nielsen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Prescription for experimental determi- nation of the dynamics of a quantum black box”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Journal of Modern Optics 44.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='11-12 (1997), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 2455–2467.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [PCZ97] JF Poyatos, J Ignacio Cirac, and Peter Zoller.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Complete characterization of a quantum process: the two-bit quantum gate”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical Review Letters 78.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='2 (1997), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 390.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Leu00] Debbie Wun Chi Leung.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Towards robust quantum computation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' stanford university, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [DP01] GM D’Ariano and P Lo Presti.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Quantum tomography for measuring experimen- tally the matrix elements of an arbitrary quantum operation”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical review letters 86.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='19 (2001), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 4195.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [OBr+04] Jeremy L O’Brien et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Quantum process tomography of a controlled-NOT gate”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical review letters 93.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='8 (2004), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 080502.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [LZK08] Daniel A Lidar, Paolo Zanardi, and Kaveh Khodjasteh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Distance bounds on quantum dynamics”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical Review A 78.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1 (2008), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 012308.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [MRL08] Masoud Mohseni, Ali T Rezakhani, and Daniel A Lidar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Quantum-process tomography: Resource analysis of different strategies”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical Review A 77.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3 (2008), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 032322.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Bia+10] Radoslaw C Bialczak et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Quantum process tomography of a universal entangling gate implemented with Josephson phase qubits”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Nature Physics 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='6 (2010), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 409–413.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Yam+10] T Yamamoto et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Quantum process tomography of two-qubit controlled-Z and controlled-NOT gates using superconducting phase qubits”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical Review B 82.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='18 (2010), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 184515.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Tro12] Joel A Tropp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “User-friendly tail bounds for sums of random matrices”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Foundations of computational mathematics 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='4 (2012), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 389–434.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [MM+13] Elizabeth Meckes, Mark Meckes, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Spectral measures of powers of random matrices”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Electronic communications in probability 18 (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Haa+16] Jeongwan Haah et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Sample-optimal tomography of quantum states”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Proceedings of the forty-eighth annual ACM symposium on Theory of Computing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 2016, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 913–925.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 19 [JP16] Anna Jenˇcov´a and Martin Pl´avala.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Conditions for optimal input states for discrimination of quantum channels”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Journal of Mathematical Physics 57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='12 (2016), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 122203.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Nec+18] Ion Nechita et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Almost all quantum channels are equidistant”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Journal of Mathematical Physics 59.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='5 (2018), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 052201.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Wat18] John Watrous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' The theory of quantum information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' Cambridge university press, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [FW20] Steven T Flammia and Joel J Wallman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Efficient estimation of Pauli channels”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: ACM Transactions on Quantum Computing 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='1 (2020), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 1–32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Gut¸+20] Madalin Gut¸˘a et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Fast state tomography with optimal error bounds”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Journal of Physics A: Mathematical and Theoretical 53.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='20 (2020), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 204001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Che+22a] Senrui Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Quantum advantages for pauli channel estimation”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Physical Review A 105.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='3 (2022), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 032435.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Che+22b] Sitan Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Exponential separations between learning with and without quantum memory”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: 2021 IEEE 62nd Annual Symposium on Foundations of Computer Science (FOCS).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' IEEE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 2022, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 574–585.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Che+22c] Sitan Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Tight bounds for state tomography with incoherent measure- ments”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: arXiv preprint arXiv:2206.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='05265 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [LN22] Angus Lowe and Ashwin Nayak.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Lower bounds for learning quantum states with single-copy measurements”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: arXiv preprint arXiv:2207.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='14438 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [Sur+22] Trystan Surawy-Stepney et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Projected least-squares quantum process tomogra- phy”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: Quantum 6 (2022), p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 844.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' [FOF23] Omar Fawzi, Aadil Oufkir, and Daniel Stilck Fran¸ca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' “Lower Bounds on Learning Pauli Channels”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' In: arXiv preprint arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content='09192 (2023).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} +page_content=' 20' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/JNFOT4oBgHgl3EQfxjTe/content/2301.12925v1.pdf'} diff --git a/PtAyT4oBgHgl3EQfhPgp/content/tmp_files/2301.00372v1.pdf.txt b/PtAyT4oBgHgl3EQfhPgp/content/tmp_files/2301.00372v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..af09b75324a3a48cffcffb0719c88382feccd8e9 --- /dev/null +++ b/PtAyT4oBgHgl3EQfhPgp/content/tmp_files/2301.00372v1.pdf.txt @@ -0,0 +1,1643 @@ +Lying Aversion and Vague Communication: +An Experimental Study +Keh-Kuan Sun +Stella Papadokonstantaki†‡ +December 31, 2022 +The latest version of the draft is available here. +Abstract +An agent may strategically employ a vague message to mislead an audience’s belief +about the state of the world, but this may cause the agent to feel guilt or negatively impact +how the audience perceives the agent. Using a novel experimental design that allows +participants to be vague while at the same time isolating the internal cost of lying from +the social identity cost of appearing dishonest, we explore the extent to which these two +types of lying costs affect communication. We find that participants exploit vagueness to be +consistent with the truth, while at the same time leveraging the imprecision to their own +benefit. More participants use vague messages in treatments where concern with social +identity is relevant. In addition, we find that social identity concerns substantially affect +the length and patterns of vague messages used across the treatments. +Keywords: Lying; Vagueness; Communication; Experiments; Behavioral Economics +JEL codes: C91, D90 +Smith Institute for Political Economy and Philosophy, Chapman University. Email: sun.k@wustl.edu +†Department of Economics, Washington University in St. Louis. Email: p.stella@wustl.edu +‡We thank Guangying Chen for her contribution to the early version of this paper. +We also thank Brian +Rogers, Marta Serra-Garcia, Agne Kajackaite, Cynthia Cryder, John Nachbar, Jonathan Weinstein, Marcus Berliant, +Sangmok Lee, Paulo Natenzon, John Rehbeck, Erik Kimbrough, and seminar participants at Washington University +in St. Louis, the ESA Job-Market Candidates Seminar Series, the Berlin School of Economics, the Korea Informa- +tion Society Development Institute, the Tokyo Institute of Technology, the ECONtribute Summer Workshop 2021 +on Social Image and Moral Behavior, the 2021 North American Summer Meeting of the Econometric Society, +the 2021 Asian Summer Meeting of the Econometric Society, and the 2021 ESA Global Online Around-the-Clock +Meetings. +1 +arXiv:2301.00372v1 [econ.GN] 1 Jan 2023 + +1 +Introduction +Communication has been studied extensively in economics in recent decades. In particular, +the types of messages sent by an agent in the communication process and their impact on an +audience’s beliefs have been important subjects of game theory. Standard economic models +predict that a sender will choose a message that yields the greatest benefit from the receiver’s +action, even when such a message is a lie. It has however been observed that senders often +choose to be evasive or use vague messages instead of telling maximal lies despite the possibility +that doing so will yield a lower benefit than lying would. This raises two questions regarding +why and when a sender prefers using a vague message over telling a maximal lie. This issue +is crucial to achieving a unified understanding of misleading behavior in many applications, +including problems associated with public good provision (Serra-Garcia, Damme, and Potters +2011), sender-receiver disclosure games (Hagenbach and Perez-Richet 2018), and persuasion +games (Deversi, Ispano, and Schwardmann 2021). +To answer these questions we must carefully examine the behavioral aspects of lying and +vague communication. First, as demonstrated by recent developments in the literature on lying +behavior (Serra-Garcia 2018; Fischbacher and Föllmi-Heusi 2013; Abeler, Nosenzo, and Ray- +mond 2019), most people exhibit a non-trivial degree of lying aversion. If an agent intrinsically +prefers being honest, then the internal cost of lying should inhibit the use of lies and encour- +age the use of vague yet relatively truthful messages. Second, a message affects not only an +audience’s belief about the state of the world but also their belief about how honest an agent +is. When the agent cares about their social identity as an honest person, this external concern +should impact the message they choose to communicate. After all, “a vague belief has a much +better chance of being true than a precise one, because there are more possible facts that would +verify it” (Russell 1923). The nontrivial cost of lying suggests that communication is no longer +cheap talk and the misleading messages may bear strategic significance as a credible signal to +the audience about the agent’s behavioral types. On the other hand, such a strategic incentive +may also influence behavior. Thus, a rational agent must balance the degree of truthfulness and +vagueness of the message communicated. +In this paper, we present a model of a cheating game in which an agent may report a vague +(set-valued) message or a precise (single-valued) message to an audience after privately ob- +serving the state of the world. The agent’s utility depends on the monetary payoff and their +2 + +truth-telling preferences. The monetary payoff is determined solely by the reported message. +We study two separable motivations for honesty: the internal motivation for being honest and +an external concern related to social identity for being seen as honest. We isolate the two costs +by employing an anonymous environment where the audience cannot identify an agent with a +message. Based on the model, we hypothesize that people use vague messages to reduce their +internal guilt and increase their monetary payoffs. The absence of the social identity concern in +the anonymous environment should lead to more straightforward profit-maximizing message +choices, while the prediction is more opaque in the non-anonymous environment because of the +multiplicity of equilibria introduced in a richer message space. +To test hypotheses pertaining to vague communication with lying aversion, we compare +treatments from an online experiment in which subjects face a variant of the Fischbacher and +Föllmi-Heusi (2013) type of reporting task (hereafter “FFH"). In this experiment, subjects pri- +vately observe an integer drawn randomly from a uniform distribution ranging from 1 to 10. +The subjects are asked to report the number to the experimenter, and their monetary payment +increases with the reported number. The basic idea of the experiment is that the discrepancy be- +tween the maximum numbers they could have reported and the actual numbers reported should +capture subjects’ aversion to lying or, by the same token, their preferences for truth-telling. We +generalize the FFH model by allowing subjects to transmit set-valued messages to understand +the effect of vagueness in communication. In our setup, we allow subjects to be vague by report- +ing multiple numbers. The experimenter then chooses one number randomly from the reported +numbers and pays the subject accordingly. +We use a 2x2 experimental design to test the predictions and, in particular, to separate the +intrinsic cost of lying from the social identity cost of appearing dishonest. First, there are two +types of experimental sessions that represent the variation in the anonymity of agents. In the +anonymous session, the responses are recorded under screen names so that the experimenter +cannot map a subject’s identity to their response. In the non-anonymous session, on the other +hand, the experimenter knows each subject’s response. The social identity cost should not af- +fect their responses in the anonymous treatment. Therefore, the difference in behavior between +anonymous and non-anonymous treatments enables us to capture the impact of the social iden- +tity cost. Second, a subject confronts two reporting tasks within each session. The observation +process is identical and independent for the two tasks. When the subject reports, however, +the set of available messages differs between the two tasks. In the task with restricted com- +3 + +munication, the set is restricted to single-valued messages only. In the task with unrestricted +communication, the subjects are allowed to use both single-valued and set-valued messages. +The experimental data indicate that most subjects use vague messages and that the means +of the numbers they report are higher on average when vague messages are allowed but the +way they use vague messages differs across the treatments. First, in the anonymous treatments, +we observe less frequent lying behavior when vague messages are allowed compared with when +they are not allowed. As anonymity should suppress subjects’ concern with social identity, the +reduction in lying behavior suggests that a vague yet truthful message reduces the internal cost +of lying. Yet, at the same time, they report higher numbers on average. That is, subjects exploit +vagueness to avoid uttering falsehoods while leveraging the imprecision to their own benefit. +Second, subjects use vague messages in the non-anonymous treatment more often than in the +anonymous counterpart. The increased frequency of vague messages and the relevance of the +concern with social identity suggest that a message’s vagueness affects message choices through +both the internal and external costs of lying. Furthermore, we find that the pattern of reported +messages differs substantially between the two types of sessions. When subjects use vague +messages in the anonymous treatment, most do not refrain from the most obvious forms: that +is, they report a combination of their true observations and the maximum number (10). We +find, however, that only a much smaller fraction of subjects using vague messages employ such +obvious messages in the non-anonymous treatment. It is possible that, in the non-anonymous +treatments, subjects may expect those obvious messages to be interpreted by the audience as +lies, which will discredit their social identities. +Our paper bridges the literature on lying behavior with a broader set of studies that involve +vague communication. Existing literature on vague communication focuses on the strategic use +of vague messages without considering the behavioral aspect of lying aversion. For instance, +Serra-Garcia, Damme, and Potters (2011), analyzing a public-goods provision game between +two agents with information asymmetry, assume that a set-valued vague message would incur +a much lower lying cost than a precise, single-valued outright lie. Our experiment tests this +assumption and provides further insight into the cost of lying with respect to the social identity +concern as well. Agranov and Schotter (2012) show that, in coordination games with multiple +equilibria, it might be beneficial for a benevolent sender to use vague communication. Zhang +and Bayer (2022) show that, in a delegation game, social welfare is higher when messages are +intervals. Wood (2022) concludes that transmitted information is more accurate when senders +4 + +have the option to send either precise or vague messages. Deversi, Ispano, and Schwardmann +(2021) study the strategic use of vagueness in a voluntary disclosure game. They allow subjects +to send an interval that contains their type, and this formal message structure is similar to that in +our design. Yet, there are two distinctive features that separate our design from theirs. First, we +provide subjects with greater flexibility in choosing messages by allowing them to choose any +subset of the state space. We also add a between-subject variation on anonymity. This flexibility +and the variation of anonymity allow us to explore the extent to which attitudes toward lying +and misleading behavior affect the use of vague messages in conjunction with strategic motives. +Conceptually, we view lying behavior as an optimization process in which a rational agent +chooses the optimal degree of dishonesty, by balancing the marginal monetary benefit of lying +with the non-monetary costs that depend on the likelihood that an individual is found to be +lying. In a more specific context involving the nature of lying behavior, Sobel (2020) provides +a comprehensive framework in which to understand lying and deception in games. He distin- +guishes between three properties —the form, the interpretation, and the consequence —of a +message, and our model closely follows the framework suggested in that paper. We would like +to particularly mention the notions of malleable lies in Turmunkh, Assem, and Dolder (2019) +and deniable lies in Tergiman and Villeval (2022) which are closely related to our empirical +findings pertaining to the use of vague messages. However, while the definitions of malleable +and deniable lies hinge on their interpretations, we define lying strictly by reference to the for- +mal property of a message: a message is a lie only when it does not represent the true state of +the world. The agents in our model face both monetary and, potentially, psychological conse- +quences for their message choices. Furthermore, through an analysis of equilibrium strategies, +we model how the interpretation of a message affects an agent’s decision. This structure allows +us to decompose the internal cost of lying from the concern for social identity. +Structurally, we study lying behavior within the framework of signaling games. Since Craw- +ford and Sobel (1982), a stream of research has tested the assumption that lying is costless, +which implies that individuals will lie whenever there is a material incentive to do so. The +empirical evidence continues, however, to make the case against this notion (e.g. Dickhaut, +McCabe, and Mukherji (1995), Blume et al. (1998), Gneezy (2005), Sánchez-Pagés and Vor- +satz (2007), Shalvi et al. (2011) and many more). Many studies argue that individuals clearly +express aversion to lying and accept significantly lower payoffs than the theories have predicted. +An empirical consensus regarding the FFH experiments is that people do not lie to the ex- +5 + +tent that they could, preferring to tell only minor lies, potentially because lying is costly. As the +monetary payoff in such an experiment is independent of the drawn number, subjects should +report non-maximal numbers only if aversion to lying is present, for otherwise this becomes a +simple case of a cheap-talk game. Abeler, Nosenzo, and Raymond (2019) combine data from +90 such experiments that describe the average reporting behavior. They find that the behavior +is indeed bounded away from the maximal report but also departs from a complete truth-telling +scenario. In addition, by ruling out other explanations for truth-telling which are popular in the +literature (e.g. inequality aversion or seeking a reputation for not being greedy), they conclude +that a preference for being seen as honest and a preference for being honest are the main moti- +vations. As Abeler, Nosenzo, and Raymond (2019) note in their conclusion, however, the FFH +paradigm has focused on subjects reporting a single number and excludes lies by omission or +vagueness. Our paper extends the FFH paradigm by allowing vague communication using set- +valued messages and contributes to our knowledge of how a message space in communication +plays a role in an agent’s reporting decision when lying entails a cost. +The studies that are closest to our work are Gneezy, Kajackaite, and Sobel (2018) and Khal- +metski and Sliwka (2019). In these studies, an agent cares not only about obtaining a monetary +payoff but also about whether lying takes place as well as how others interpret their report, +a finding that is consistent with the empirical findings of the aforementioned study by Abeler, +Nosenzo, and Raymond. These papers adopt the FFH paradigm and simplify the sender/receiver +structure of a signaling game into a cheating game between reporting agents and an observing +audience. This idea, in conjunction with the assumption of a non-atomic game, minimizes the +role of the receiver and allows a sharper focus on that of the sender. Moreover, this simplification +makes it possible to develop a unique characterization of off-path beliefs in the context of mild +conditions. While we inherit their assumptions of these assumed motivations for truth-telling +—internal guilt and the external concern with one’s social identity —extending the message +space to include set-valued messages yields more opaque predictions with multiple equilibria +and off-path beliefs. We overcome this difficulty by introducing an anonymous environment to +hold the impact of social identity constant. +We observe that, as long as a message can remain even remotely truthful, most agents exhibit +no hesitation in increasing their monetary payoff. This observation is analogous to behavior +reflecting “moral wiggle room” (Dana et al. 2007), the notion that people engage more resolutely +in self-interested behavior when excuses for selfish actions are available. On the other hand, +6 + +there exists a group of truth-tellers in our experiment whose reports are not only truthful but also +precise, thus possibly relinquishing opportunities to obtain monetary gains. The non-trivial size +of this group suggests another motivation for truth-telling in addition to internal and external +concerns regarding honesty. Candidates might include a concern for the consequences of choices +made or concern with good intentions. +The remainder of the paper is organized as follows. In Section 2 we define our terms and +present the model setting. Section 3 provides theoretical analysis, while in Section 4 we present +the experiment hypotheses based on the theoretical predictions. In Section 5 we describe our +experiment design, in Section 6 we summarize the experimental outcomes, and Section 7 con- +cludes. We list all the proofs and additional details regarding the experiment in the appendices. +2 +Model +2.1 +A model of lying aversion with vague communication +We study lying aversion with vague communication by considering a variant of the Fischbacher +and Föllmi-Heusi cheating game with a population of agents and one audience. An agent pri- +vately observes the state of the world i ∈ Ω, where Ω = {1,2,..., N} is finite. We assume i is +drawn i.i.d. from a uniform distribution over Ω across agents. Each agent has a private type t +that represents their intrinsic aversion to lying. We also assume t is i.i.d. across agents. Its CDF +F(t) is strictly increasing, continuous, and has support [0, T]. Denote an agent who observes +the state i and has the intrinsic aversion type t as a type (i, t) agent. +An agent reports a message J to the audience after observing the true state of the world +i. This reporting takes place only once and there are no repeated interactions. Like Gneezy, +Kajackaite, and Sobel (2018)(GKS) and Khalmetski and Sliwka (2019)(KS), we assume that +an agent’s utility consists of three components: a monetary payoff, an internal concern with +being honest, and a social identity for being seen as honest. We use the term ‘social identity’ +to distinguish it from the typical understanding of reputation in a dynamic game, as we are +modeling a one-shot game. Formally, we write a type (i, t) agent’s utility for reporting a message +J as: +U(i, J, t) = +¯π(J) +���� +monetary payoff +−1(i ̸∈ J)[t + c(π({i}), ¯π(J))] +� +�� +� +internal guilt ++ +γρ(J) +� �� � +external social identity +. +(1) +7 + +The message J is a nonempty1 subset of the state space: J ∈ MΩ ≡ 2Ω \ �. An important +distinction between our model and GKS and KS is that in our model the message space admits +set-value messages instead of being isomorphic with the state space. This generalization allows +messages to be categorized in multiple ways. We define the relevant terms below. +Definition 1. A message J is truthful if i ∈ J. A message is a lie if it is not truthful. +Definition 2. A message is called precise if it is a singleton set; otherwise it is vague. +Let MΩ +P ≡ {{1},{2},...,{N}} and MΩ +V ≡ MΩ \ MΩ +P denote the sets of precise messages and +vague messages, respectively. For example, if Ω = {1,2,3}, the set of possible precise messages is +{{1},{2},{3}}, and the set of possible vague messages is {{1,2},{2,3},{1,3},{1,2,3}}.2 Note +that our definition of lying depends purely on the form of a message, as in the framework +proposed in Sobel (2020). The consequences of a message choice are reflected through effects +on the components of the agent’s utility. +The monetary payoff for an agent maps a message to the agent’s utility: π : MΩ → �. For +simplicity, we assume that π(J) is a uniform draw over J and that the agent is a risk-neutral, +expected-utility maximizer. Let us denote ¯π(J) = E[π(J)]. Note that, when a message is precise, +i.e. J = {x} with x ∈ Ω, the monetary payoff π(J) is simply x. +In addition to receiving a monetary payoff, the agent also has two distinct motivations for +being honest. First, the agent has an internal motivation for being honest. When the agent’s re- +port J is not truthful, that is, i ̸∈ J, the dishonesty incurs an internal cost, t+c(π({i}), ¯π(J)). The +agent’s private type t captures their sensitivity to the intrinsic (fixed) cost of lying. The function +c(π({i}), ¯π(J)) : �×� → � represents the variable cost of lying depending on the size of the lie. +The size of the lie is measured as the ex-ante difference between the monetary payoff for a report +J and that of the true and precise report {i}. We assume that i) c(·) ≥ 0; ii) c(π({i}),π({i})) = 0; +iii) c(π({i}), ¯π(J)) is weakly increasing in |π({i}) − ¯π(J)|; iv) c(π({i}),π({i}) + 1) < 1;3 and v) +c(π({i}), ¯π(J)) + c( ¯π(J), ¯π(K)) ≥ c(π({i}), ¯π(K)). +1. An empty set is often interpreted as silence and plays an interesting role in the literature on vague +communication. We abstract away from silent messages to compare cases where lying aversion is accom- +panied by vague language with cases where lying aversion is accompanied by precise language. +2. The fully vague message Ω ∈ MΩ is analogous to the notion of evasive lying in Khalmetski, Rock- +enbach, and Werner (2017) by which the sender falsely states to have not observed the state of the +world. +3. This assumption excludes the trivial case where the variable cost of lying is so high that no agent +chooses to lie. +8 + +The agent also has an external motivation to be ascribed a social identity for being seen as +honest; that is, the agent’s utility depends on the audience’s belief about how honest the agent +is. We assume that the audience is a rational Bayesian who forms a posterior belief based on +the agent’s report J. This belief, in turn, depends on the agent’s mixed strategy in equilibrium. +Definition 3. A type (i, t) agent’s mixed strategy is a mapping σ : Ω × [0, T] → [0,1]2N−1 such +that +σ(i, t) = +� +σ{1} +it ,σ{2} +it ,...,σ{1,2,...,N} +it +� +(2) +where σJ +it is the probability which the intrinsic aversion type t agent with true observation i +assigns to the report J ∈ MΩ. +In equilibrium, the audience’s posterior belief about whether an agent’s report J is truthful +is computed using Bayes’ rule: +ρ(J) = P(agent is honest ∧ agent reports J) +P(agent reports J) += +� +k∈J( +� T +0 σJ +ktd f (t)) +�N +k=0( +� T +0 σJ +ktd f (t)) +(3) +We normalize the posterior belief ρ in terms of the agent’s utility by parameter γ. The pa- +rameter measures the agent’s sensitivity to their social identity reflected in reporting message +J. We further assume that i) γ is homogeneous across agents and is common knowledge; more- +over, ii) N + γ < T. The latter condition ensures that F(N + γ) < 1, or there always exists a +positive mass of agents who will be truth-telling for any observation i. +Because the agent’s payoff depends on the audience’s belief, we adopt the notion of se- +quential equilibria in an induced psychological game in the sense of Battigalli and Dufwenberg +(2009) and Geanakoplos, Pearce, and Stacchetti (1989). +An equilibrium is a set of mixed strategies and beliefs satisfying the following conditions: +∀(i, J, t) : σJ +it > 0 only if J ∈ argmax +J′ +U(i, J′, t), +(4) +∀(i, t) : +� +J∈MΩ +σJ +it = 1, +(5) +∀J : ρ(J) = +� +k∈J( +� T +0 σJ +tkd f (t)) +�N +k=0( +� T +0 σJ +tkd f (t)) +. +(6) +9 + +As noted by GKS, the existence of an equilibrium follows Schmeidler (1973) in treating each +type (i, t) as a player. +2.2 +The communication environment and the anonymity of agents +The baseline model assumes no restriction on message space MΩ. If we restrict message space +to MΩ +P , on the other hand, we obtain a model with restricted communication. We refer to +the baseline model as the model with unrestricted communication. At the risk of abusing the +notation, let us denote a message as j in the model with restricted communication. +Furthermore, for a more comprehensive understanding of the relationship between the use +of vague messages and the costs of lying, it is helpful to isolate their effects on internal guilt +from their effects on external social identity. We can achieve this by employing an anonymous +environment where the audience cannot identify an agent with a message. This implies that the +agent’s report does not alter the audience’s belief, and hence the agent’s social identity remains +constant independent of their reporting choice in the anonymity-preserving environment. On +the other hand, we call an environment non-anonymous when the audience can associate a +message with its sender. Formally, we write an agent’s utility in the anonymous environment +as: +UA(i, J, t) = ¯π(J) − 1(i ̸∈ J)[t + c(π({i}), ¯π(J))]. +(7) +We thus create four distinct environments by varying the restriction on the message space +and the anonymity of the agents. In these environments, the agent is non-anonymous with re- +stricted (precise only) communication (NA-R), non-anonymous with unrestricted (potentially +vague) communication (NA-UR), anonymous with restricted communication (A-R), and anony- +mous with unrestricted communication (A-UR). Note that the NA-R environment corresponds +to the standard FFH model where an agent can send only precise messages and the audience +can identify a message’s sender. +Non-anonymous +Anonymous +Restricted +NA-R +A-R +Unrestricted +NA-UR +A-UR +Table 1: Four environments +10 + +3 +Analysis +In this section we describe the equilibria and agents’ behavior in each of the four environments +as defined above. We begin by arguing that we should expect to observe vague messages used +at positive probabilities in any equilibrium. +Proposition 1. In any equilibrium under unrestricted communication, there exist types of agents +who use at least one vague message with positive probability in their mixed strategies. +Proof. See Appendix A. +The intuition behind Proposition 1 is that there exist agents with moderately high intrinsic +lying aversion type t who would prefer to be truthful while increasing the monetary payoff. +Therefore, in their messages, they include higher numbers apart from their true observations. +For these agents, such messages strictly dominate any precise lie. +In comparison with the unrestricted communication environment in which we are inter- +ested, a particularly useful element when analyzing a restricted communication environment is +the assumption that there always exists a positive mass of truth-tellers for each observation i. +While this assumption is sufficiently plausible in many contexts, it aids us especially in elimi- +nating less interesting equilibria by uniquely characterizing off-path beliefs. Thus, we are able +to make sharp equilibrium predictions even in the case of the non-anonymous environment. +This, however, is unfortunately not the case when we relax the message space to allow vague +messages: the richer message space induces multiplicity in agents’ reporting strategies and the +audience’s off-path beliefs. We illustrate this point with the following examples. +Example 1. Consider the case where γ > N−1 +2 . Then this case in which all agents report the +vaguest messages {1,2,..., N}, such that ρ(J) = 1 only when J = {1,2,..., N} and 0 otherwise, +is an equilibrium. +This example describes the scenario in which agents care greatly about social identity and are +forced to stick to the (exogeneously) given norm of full vagueness. While this specific example +is similar to situations that might occur in the real world,4 an important implication is that there +4. For instance, we may imagine a group of politicians all replying with the same vague message to a +politically sensitive question because they know that any message other than a fully vague message will +be interpreted as a lie and hurt their social identity. Many similar examples can arise in situations where +agents value their social identity highly. We appreciate Brian Rogers for suggesting this interpretation of +the equilibrium. +11 + +can exist a multiplicity of equilibria depending on the combination of γ and the off-path beliefs. +Consider the following example. +Example 2. All agents report a message in the form of an interval: [i, N] = {i, i+1,..., N−1, N}, +and the audience assigns a positive posterior belief only to those interval-messages and zero to +all other messages. We need the following conditions to hold to constitute such equilibria: +• truth-tellers: i+N +2 +γρ(J) > U(i,{i, x∗, x∗+1,..., N}, t)5 for all i and an interval message +J +• liars: k+N +2 +− 1(i ̸= k)(t + ¯π(J)) + γρ(J) > U(i, J′, t) for any non-interval message J′ +The conditions hold trivially when γ → ∞, and the signaling game becomes isomorphic with +the NA-R scenario, where each message i is simply replaced with the interval [i, N]. The ending +sequence conveys no information to the audience, and the equilibrium follows that of the NA-R. +The multiplicity of equilibria obscures our understanding of the agents’ behavior regarding +vague communication. We thus first turn our attention to anonymous environments to delineate +the relationship between one’s internal cost of lying and vague messages. +Because we are now free of the social identity component, we can utilize the fact there is +a clear mapping between an agent’s type (i, t) and the resulting behavior. That is, fixing the +observation i, one’s lying behavior simplifies to a monotone function of t in the A-R environ- +ment: the agent reports j > i when (j −i)+c(i, j) > t. Thus, there exists a threshold t∗ +i for each +observation i below which the agent reports a lie and above which the agent tells the truth. This +threshold, however, becomes much higher in the A-UR environment because now the agent is +equipped with vague messages that allow them to remain truthful while increasing the expected +payoff. This fact leads to the following observations. +Observation. If a type (i, t) agent reports truthfully after observing state i in the A-R environment, +then the agent also reports truthfully when observing i in the A-UR environment. +Observation. For each observation i < N, there exists a positive mass of agents who lie in the A-R +environment but report truthfully in the A-UR environment. +5. This message maximizes the expected payoff conditional on including the true state i. We later +define this kind of message as an optimally vague message. +12 + +The immediate corollary of these two observations is the following proposition, which states +that the mass of liars is greater when communication is restricted in the anonymous environ- +ments. +Proposition 2 (A-R/A-UR). The set of types (i,t) of agents who lie in equilibrium in the A-UR +environment is a subset of liar types in any equilibrium in the A-R environment. The expected +monetary earnings are greater on average when the communication is not restricted. +Another implication of the anonymous environment worth noting is that the absence of the +social identity concern simplifies the problem into a straightforward comparison between the +solutions to truth-telling-constrained and unconstrained optimizations. That is, for each type +(i, t) agent, there exists both maxJ:i∈J U(i, J, t) and maxJ∈MΩ U(i, J, t), and the agent reports +truthfully only if the two maximums coincide. +We can add structure to the constrained optimization. Intuitively, this maximum has the +form of a union of the true observation i and some ending sequence x, x + 1,..., N. We can +solve the optimization problem to find the threshold x∗ = ⌈(N + 2) − +� +2N − 2i + 3⌉. 6 +Definition 4. The optimal vague message (OVM) for each true observation i is defined as +OV Mi ≡ {i, x∗, x∗ + 1,..., N}, +(8) +where x∗ the threshold above which including the values maximizes. +Then OV Mi ∈ argmaxJ:i∈J U(i, J, t) and weakly dominates all truthful messages. This allows +us to make the following inferences in the A-UR environment: +Proposition 3. In the A-UR environment, all truth-tellers use the optimal vague messages. +Corollary. In the A-UR environment, +i. no agent’s message contains a number below the true observation. +ii. no precise message except {N} is truthful. +6. We can simplify the optimization problem by approximating it with a continuous uniform distribu- +tion: argmaxx +� i+1 +i +u +N−x+2 du + +� N +x +u +N−x+2 du +13 + +Lastly, we conclude this section by comparing the behavior of an agent in the non-anonymous +environment with that of an agent in the anonymous environment. A thought experiment that +involves choosing an agent and comparing their behavior in the two environments easily leads +to the conjecture that the absence of the social identity concern should only encourage more +lies. The following lemma shows that this is indeed the case under restricted communication. +Lemma 1. If a type (i, t) agent lies in an equilibrium in the NA-R environment, then the agent lies +in the A-R environment. +Proof. See Appendix A. +Using the above lemma, we now generalize the argument to compare the probability that +an arbitrarily chosen agent lies in the two environments. +Proposition 4 (NA-R/A-R). The set of types (i, t) of agents who lie in an equilibrium in the NA-R +environment is a subset of the set of liar types in any equilibrium in the A-R environment. The +expected monetary earnings are higher on average in the A-R environment. +Proof. See Appendix A. +4 +Hypotheses +In this section, we have transformed our theoretical analysis into specific hypotheses that can be +tested in an experiment. The results of the experiment, which will be discussed in the following +section, will be based on these hypotheses. +Hypothesis 1. In both the NA-UR and A-UR environments, agents use vague messages. +Hypothesis 1 is a consequence of Proposition 1. +Hypothesis 2 (A-R/A-UR). In the anonymous environment, +i. more agents lie when communication is restricted (precise): lieA−R ≥ lieA−UR; +ii. an agent who is truthful in A-R is also truthful in A-UR conditional on the same observation; +iii. some agents who lie in A-R report truthfully in A-UR conditional on the same observation; +14 + +iv. agents earn higher monetary payoffs on average when communication is not restricted (vague): +earningA−R ≤ earningA−UR. +Hypothesis 2 is a consequence of Proposition 2 and the above observations about behavior +in the A-UR environment. +Hypothesis 3. In the A-UR environment, +i. all truth-tellers in A-UR use OVM; +ii. no message contains a number below the true observation; +iii. no precise message except {N} is truthful. +Hypothesis 3 is a consequence of Proposition 3 and its corollary. +Hypothesis 4 (NA-R/A-R). Under restricted communication, agents earn higher monetary payoffs +on average in the anonymous environment: earningNA−R ≤ earningA−R. +Hypothesis 4 is a consequence of Proposition 4. +5 +Experimental Design +We vary the experimental treatments along two dimensions: 1) we consider precise or vague +messages, and 2) we vary the experimenter’s ability to identify responses to an individual sub- +ject. We represent variation in the anonymity of subjects in two types of experiments. Within +each session, a subject confronts two stages of reporting tasks that represent the availability of +vague messages. In each stage, subjects are incentivized to observe a random number and later +report the number to the experimenter. Their earnings depend on the number or numbers they +report. +In the anonymous session, responses are recorded under screen names so that the exper- +imenter cannot map a subject’s identity to their response. As the subjects are instructed that +the experiment prohibits such mapping by design, this treatment should establish the effect of +suppressing the social identity concern and emulate the environment where γ → 0. In the non- +anonymous (identifiable) session, on the other hand, the experimenter knows each subject’s +response. +15 + +The ‘stage’ is our basic unit of observation. In each stage, subjects first observe the realiza- +tion of a random integer uniformly distributed between 1 and 10 and later are asked to report +the number. In the anonymous treatment, the number is generated within the experimental +software so that the experimenter knows both the true observation what each subject reports. +In the non-anonymous treatment, subjects use an external website to generate the random num- +ber so that the experimenter cannot know the true observation.7 The experimenter is unable +to identify whether a particular individual has lied or not in any of the treatments. This design +choice derives from the idea that an environment in which the experimenter can detect cheating +at the individual level is overly artificial; as result, subjects’ discomfort with lying behavior may +be exaggerated in such an environment. +The observation process for the two tasks is identical and independent but the set of available +messages differs between the two tasks. In the restricted stage, the set is restricted to single- +valued messages only. In the unrestricted stage, subjects are allowed to use both single-valued +and set-valued messages. In the restricted stage, the subjects are paid the equivalent in dollars +of the number they report divided by 2. In the unrestricted stage, when they report a single +number, they are paid the equivalent in dollars to the number they report divided by 2, while +when they report multiple numbers, the computer will randomly choose one of the numbers and +they are paid the equivalent in dollars of this number divided by 2. To emulate the one-shot +game structure of our model, we asked subjects to participate only once in each of the stages +and we randomized the order of precise and vague stages within a session.8 +The combination of these between-subject and within-subject variations yields four treat- +ments in our experiment: NA-R,9 NA-UR, A-R, and A-UR. +We recruited 176 student subjects from the subject pool of the Missouri Social Science Ex- +perimental Laboratory (MISSEL) at Washington University in St. Louis. For recruitment and +experiment management, we used the ORSEE system (Greiner 2015). Students were invited +to participate in a virtual Zoom meeting where the experimenter read the instructions and pro- +vided a link to the main experiment web page. The main experiment was implemented using +the Qualtrics online survey platform. +We conducted 18 sessions and each session lasted for approximately 30 minutes, including +7. We provided a link to a Google search result for the phrase “random number between 1 and 10.” +8. Using OLS regressions, we find that there is no order effect on the average report. +9. The NA-R treatment is equivalent to the original experiment conducted by Fischbacher and Föllmi- +Heusi (2013). +16 + +the instructions read during Zoom meetings, a screening quiz to make sure that subjects under- +stood the experimental procedures, and two stages of the main experiment task via Qualtrics. +10 In all cases, subjects received a $2 show-up fee, so the total amount they could earn ranged +from $2 to $12, with an average total fee, including the show-up fee, of $9.84. No subject +participated in more than one session. See Appendix D for our experimental instructions and +procedural details. +6 +Results +We present basic summary statistics in Table 2. +There were 8 anonymous sessions and 10 +non-anonymous sessions, with an average of 12.3 and 8.2 participants, respectively, in each +session.11. In the unrestricted communication treatments, we used the mean of the numbers +included in each subject’s report when calculating the average report. We computed the average +length of vague messages using the average number of numbers included in the messages. +Restricted +Unrestricted +Observations +Average +Average +Vague (%) +Length +Anonymous +7.619 +8.132 +48% +2.103 +97 +Non-anonymous +7.000 +8.282 +75% +2.810 +79 +Table 2: Data summary +We report first that the majority of subjects used vague messages in both treatments. +Result 1. In the NA-UR environment, 59 of 79 participants (74.7%) used vague messages. In the +A-UR environment, 47 of 97 participants (48.4%) used vague messages. +This result supports Hypothesis 1. +In the anonymous treatment, the numbers reported for the restricted and unrestricted treat- +ments average 7.619 and 8.132, respectively, while in the non-anonymous treatment, the num- +bers reported average 7 and 8.282, respectively. These reported numbers are significantly lower +10. The first 11 sessions were conducted from August 2020 to October 2020, while the remainder were +conducted in September 2022. The results of a logit regression show no significant differences in the +probability of lying between the two groups of sessions. +11. Participants chose time blocks voluntarily, and we provided the anonymous version of the software +in larger sessions. +17 + +than those in the profit-maximizing reports and are in line with findings reported in the liter- +ature that have highlighted agents’ preference for truth-telling. Overall, we find that allowing +vague messages increased the mean of numbers reported. We also find that the probability of +lying is independent of gender in both the restricted and unrestricted treatments. Additional +details regarding the logit regression outcomes are available in Appendix B. In addition, we +find that the impact of anonymity on average reports is statistically not significant. Finally, we +show that the introduction of non-anonymity significantly affects the choice to send vague mes- +sages. In the treatment where social identity concern is relevant, subjects use vague and longer +messages more often. +Result 2. In the anonymous environment, +1. more participants lied when communication was restricted (43.3% in A-R and 30.9% in +A-UR); +2. almost all participants who were truthful in A-R remained truthful in A-UR;12; +3. 52.58 % of subjects reported truthfully under both A-R and A-UR, 26.80% always lied, while +16% switched from lying to truth-telling when they were allowed to be vague; +4. participants reported higher numbers on average when communication was not restricted +(7.619 in A-R and 8.282 in A-UR). +A comparison between the results obtained with restricted and unrestricted communication +in anonymous sessions confirms Hypothesis 2. More subjects lied in the restricted than in the +unrestricted treatment and their behavior can be classified into three types: truth-tellers (re- +ported truthfully under both treatments), conditional liars (switched from lying to truth-telling +when they were allowed to be vague) and liars (lied under both treatments). This is in line with +previous work on lying aversion (Gneezy, Kajackaite, and Sobel (2018), Khalmetski and Sliwka +(2019), among others). An interesting point is that 15 of the 16 participants who switched from +lying to truth-telling reported picking the number 10 in A-R, yet they chose vague and truthful +messages when the expected earnings were lower than 10. This finding supports our conjecture +that the internal cost of lying depends on the inclusion of true states in the reported messages. +12. Four subjects were truthful in the A-R and lied in the A-UR, but their true observations were much +larger in the A-UR than in the A-R treatments. It is likely that the counterfactual would be consistent +with our prediction conditional on observing similar numbers in both treatments. +18 + +Table 3 presents the average numbers reported in the A-R and A-UR treatments for truth- +tellers, conditional liars, and liars. We find that truth-tellers report on average 1.282 higher +numbers when they are allowed to be vague. This result is significant at the 1% significance +level and is consistent with the model’s prediction that truth-tellers seek payoff maximization +conditional on including true observations in their reports. Conditional liars reported on av- +erage 1.156 lower numbers when they were allowed to be vague. This result is significant at +the 1% significance level and supports Hypothesis 2. We interpret this result as implying that +subjects with moderate t prefer to use vague messages and include true states to reduce the +internal cost of lying. Consistent with our expectation regarding agents with higher t, we did +not find a significant impact of unrestricted communication on liars as reflected in their aver- +age reports. In addition, t-tests of the difference between subjects’ true observations and their +reports show over-reporting across all three types in the A-UR treatment. This result is also +in line with Hypothesis 2, which states that agents earn higher monetary pay-offs under un- +restricted communication. Likewise, we also find that a true observation has a negative effect +on the probability of lying in the restricted treatment, while it has no effect in the unrestricted +one. Table 4 presents the estimated coefficients of the logit model of the effect of true obser- +vations on the probability of lying. This result is intuitive, as a subject who randomly drew a +low observation can increase their payoff only by lying when the communication is restricted to +precise messages. These subjects can, however, always employ a vague yet truthful message in +the unrestricted environment; hence the true observation has no effect. +19 + +Restricted +Unrestricted +Differences +in means +Observations +Truth-tellers +5.608 +6.890 +1.282 +51 +(0.416) +(0.293) +(0.464) +Conditional liars +9.875 +8.718 +-1.156 +16 +(0.125) +(0.316) +(0.370) +Liars +10 +9.962 +0.038 +26 +(0.027) +(0.027) +Test (Report = True) +Truth-tellers +0 +1.953 +51 +(0.267) +Conditional liars +5.000 +1.531 +16 +(0.791) +(0.432) +Liars +5.462 +5.076 +26 +(0.521) +(0.543) +Table 3: Anonymous Treatment +Observation (Restricted) +Observation (Unrestricted) +Lying dummy +-0.138 +-0.081 +(0.072) +(0.081) +Table 4: Logit model of the probability of lying +Result 3-1. In the A-UR environment, +1. +(a) 28.4% (19 of 67) of truth-tellers reported the optimal messages (either the OVM or +the honest 10), 26.9% (18 of 67) reported a pair of true observations and 10; +(b) 23.9% (16 of 67) of truth-tellers used a precise message below {10}; +2. only 2 of 97 subjects included a number below the true observation in the report; +3. all precise messages reporting numbers below {10} were truthful. +This result partially supports Hypothesis 3. Figure 1 summarizes the types of messages used +in A-UR treatments. Overall, 51.5% (50 of 97) of participants used precise messages, while +48.5% (47 of 97) used vague messages. Among the 50 precise messages, 54% (27 of 50) were +20 + +Figure 1: Message types used in A-UR treatments +lies and 46% (23 of 50) were not lies. All the liars reported drawing the maximum of 10. Among +the 23 precise truth-tellers, 7 observed 10 and reported so. +The model predicts that all truth-tellers seek payoff maximization conditional on including +true observations in their reports. If we combine both optimal vague messages (including honest +10 as the optimal message) and the pair-type messages into a broader set of payoff-increasing +truthful messages, we find that the majority (55.2%) of truth-tellers maximized their monetary +payoffs conditional on being honest.13 Yet there remains a noticeable number of precise truth- +tellers who reported drawing a number below 10, which contradicts our hypothesis. This may +suggest the possibility of other motivation for truth-telling that is not captured by our model. +Comparing behaviors across the A-UR and the NA-UR treatments, it is interesting to note +that a higher percentage of participants used vague messages in the non-anonymous environ- +ment. Overall, 25.3% (20 of 79) of participants used precise messages while 74.7% (59 of +79) used vague messages. This difference might reflect the fact that some maximal liars in the +13. When a message is both optimal and precise, namely {10}, we count it as an optimal message. +Likewise, when a message is both optimal and contains two numbers, like {8,10}, we label it ‘Optimal.’ +We classify sub-optimal messages using two numbers as ‘Pair.’ +21 + +75 +Percent (%) +Category +50 +27.84 +lie +not lie +25 +23.71 +1.03 +2.06 +19.59 +14.43 +11.34 +0 +Precise +Optimal (vague) +Pair +Others +Form of reportFigure 2: Message types used in NA-UR treatments +anonymous case lie by reporting 10 in the absence of any concern with social identity. Also, it +is worth noting that vague messages were longer in the non-anonymous treatments (2.81 num- +bers reported on average) compared with what occurred in the anonymous environment (2.10 +numbers reported on average). The difference is significant at the 1% significance level and +could reflect the impact of vague messages on the external cost of lying. To further compare +the effect of social identity on vague communication, we categorized vague messages into the +following three categories: ‘Pseudo-optimal,’ ‘Pair,’ and ‘Others.’ While we do not know the true +observation of each subject in the non-anonymous treatments, we classified optimal-looking +messages as ‘Pseudo-optimal’ in a similar manner to how we defined the OVM in Definition 8, +using the minimum reported numbers in the messages as pseudo-true-observations. For exam- +ple, a report consisting of {6,9,10} is considered pseudo-optimal because this message would +maximize the expected payoff if the true observation was 6. A report {6,7,8,9,10}, on the other +hand, is not classified as pseudo-optimal.14 Figure 2 summarizes the types of messages used in +14. When a message is both pseudo-optimal and pair, like {8,10}, we count it as ‘Pseudo-optimal,’ +similar to how we classified vague messages in the A-UR treatments. Only sub-optimal messages using +two numbers are labeled ‘Pair.’ +22 + +75 +Percent (%) +Category +50 +Interval +27.85 +Non-interval +25 +25.32 +20.25 +8.86 +10.13 +7.59 +0 +Precise +Pseudo-optimal (vague) +Pair +Others +Form of reportNA-UR treatments. +Result 3-2. In the NA-UR environment, +1. 22.0% (13 of 59) and 13.6% (8 of 59) of vague messages are pseudo-optimal and pairs, +respectively; and +2. 49.2% (29 of 59) of vague messages take the form of an interval. +Note that, compared with the results observed in the anonymous treatment, as displayed in +Figure 1, participants used fewer precise and pair-type messages. Conditional on using vague +messages, only 36.6% are either pseudo-optimal or pairs in the NA-UR treatment. This ratio +differs starkly from the 53.2% observed in the A-UR treatment. In addition, we see that the vast +majority of the messages now fall into the ‘Others’ category. Among the ‘Others’ messages, half +take the form of intervals, which was not the case in the anonymous treatments. In particular, +49.1% of vague messages reported in the NA-UR treatment used intervals, while only 37.2% did +in A-UR. We can infer from the forms of the reported messages that subjects understood that +pseudo-optimal and pair messages are obvious and can negatively affect the audience’s belief +in an equilibrium. +The distribution of the lengths of messages also indicates that there is a significant differ- +ence in reporting patterns between the A-UR and the NA-UR treatments. In Figure 3 we report +the cumulative distribution of message lengths for the A-UR and NA-UR treatments. The fig- +ure shows that subjects use vague and longer messages more often in the NA-UR treatment. +Together with the frequent use of interval messages in the non-anonymous treatment, we inter- +pret this pattern as indirect evidence of the impact of vagueness on the external cost of lying. +Result 4. Comparison between the A and NA environments +1. The average numbers reported in A-R and NA-R are 7.619 and 7, respectively; +2. the average numbers reported in A-UR and NA-UR are 8.132 and 8.232, respectively. +We do not find statistically significant differences between average numbers reported in +the anonymous and non-anonymous treatments. Our results do not reject the null hypothe- +ses under either restricted or unrestricted communication, with p-values 0.185 and 0.562, re- +spectively. While these results run against our Hypothesis 4 prediction, they are in line with +23 + +Figure 3: The cumulative distribution of lengths of vague messages in the A-UR and +NA-UR treatments +Fries et al. (2021), who find a similarly perplexing outcome according to which increasing the +anonymity of a report has no effect on average numbers reported or lying behavior. One possi- +ble explanation is that the impact of anonymity on behavior creates only a minuscule difference +when collapsed into an average and projected into a unidimensional space, and our sample is +far from the infinitely large sample needed to provide such statistical power. The difference we +documented across the restricted and unrestricted treatments makes this explanation particu- +larly plausible. Despite the similar magnitudes of the average numbers reported, the types of +messages used in the two environments differ substantially in form and length. +24 + +100 +8 +4 +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +length of messages +Non-Anonymous +Anonymous7 +Concluding Remarks +We contribute to the literature by bridging lying behavior with studies that involve vague com- +munication. The extensive use of vague messages among our subjects, in both anonymous and +non-anonymous sessions, provides us with insights that can be applied to various communica- +tion contexts. We find that subjects exploit vagueness to be consistent with the truth, while at +the same time leveraging such imprecision to their own benefit. In addition, we find that 38% +of liars in the anonymous sessions switch to truth-telling when they are allowed to be vague, +which supports the key conjecture that a vague and truthful message reduces the internal cost +of lying. We show that subjects use vague and longer messages more often when they are con- +cerned with social identity. The difference in the reporting patterns across the NA-UR and the +A-UR sessions suggests that vagueness still plays an important behavioral role. +Overall, subjects included higher numbers in their messages more often when vague com- +munication was allowed. This finding sheds new light on our understanding of lying aversion, +suggesting that a restricted message space could be a source of the observed abstention from +monetary-payoff maximization in previous experiments reported in the literature. The aver- +age numbers reported in the treatments with restricted communication in our experiment are +comparable to results reported in the previous experiments, capturing approximately 70% of +the maximum numbers available. Insofar, as we allow the use of vague messages, however, +the average number reported increases to 8.1, reducing the degree of lying aversion from 30% +of the maximum monetary payoff amount to less than 20%. In addition, our findings in the +anonymous treatments indicate that our subjects lie less often as they increase their expected +monetary payoffs. That is, the option to use vague messages enables the subjects to maintain +their integrity by including true states in their messages at a lower monetary cost. Similarly, sub- +jects in the non-anonymous treatments also reported higher numbers on average when vague +messages were allowed. Yet, the difference in the reporting patterns —vague messages with +longer and more sophisticated forms —suggests that the use of vague messages provides sub- +jects an alternative means to retain their social identity. In other words, when subjects are given +a new option to use vague messages to convey their honesty to an audience, they no longer need +to sacrifice their monetary payoffs significantly to provide a credible signal. +Our finding is analogous to the “moral wiggle room” mentioned in Dana et al. 2007, where +dictators care only whether they maintain an image of fairness, but without having a signifi- +25 + +cant effect on their partners’ payoffs. In our experiment, most subjects exhibit no hesitation in +increasing their monetary payoffs as long as their messages can remain even remotely truth- +ful. The presence of precise truth-tellers in the anonymous environment, while small, suggests +however that the moral-wiggle-room argument alone may not paint the whole picture. The de- +composition of the aversion to monetary-payoff maximization in our experimental design calls +for a new perspective on misleading behavior by showing that the observed aversion in many +individuals is independent of the consequences of their message choices. It is possible that +another motivation for truth-telling, such as a concern with projecting a well-intentioned self- +image. That is, although subjects might have understood that there was no external observer +to judge their behavior, their moral standards may combine the honesty achieved by reporting +true observations with the uprightness of reporting the most accurate messages. +26 + +References +Abeler, Johannes, Daniele Nosenzo, and Collin Raymond. 2019. “Preferences for Truth- +Telling.” Econometrica 87 (4): 1115–1153. +Agranov, Marina, and Andrew Schotter. 2012. “Ignorance is bliss: an experimental study +of the use of ambiguity and vagueness in the coordination games with asymmetric +payoffs.” American Economic Journal: Microeconomics 4 (2): 77–103. +Battigalli, Pierpaolo, and Martin Dufwenberg. 2009. “Dynamic psychological games.” +Journal of Economic Theory 144 (1): 1–35. +Blume, Andreas, Douglas V DeJong, Yong-Gwan Kim, and Geoffrey B Sprinkle. 1998. +“Experimental Evidence on the Evolution of Meaning of Messages in Sender- +Receiver Games.” The American Economic Review 88 (5): 1323–1340. +Crawford, Vincent P, and Joel Sobel. 1982. “Strategic Information Transmission.” +Econometrica 50 (6): 1431–1451. +Dana, Jason, Roberto A Weber Jason Xi Kuang, Carnegie Mel-lon, J Dana, R A Weber, +and J X Kuang. 2007. “Exploiting moral wiggle room: experiments demonstrating +an illusory preference for fairness.” Economic Theory 33:67–80. +Deversi, Marvin, Alessandro Ispano, and Peter Schwardmann. 2021. “Spin doctors: +An experiment on vague disclosure.” European Economic Review 139 (October): +103872. +Dickhaut, John W, Kevin A McCabe, and Arijit Mukherji. 1995. “An experimental study +of strategic information transmission.” Economic Theory 6 (3): 389–403. +Fischbacher, Urs, and Franziska Föllmi-Heusi. 2013. “Lies in Disguise - An Experimental +Study on Cheating.” Journal of the European Economic Association 11 (3): 525– +547. +Fries, Tilman, Uri Gneezy, Agne Kajackaite, and Daniel Parra. 2021. “Observability and +lying.” Journal of Economic Behavior & Organization 189:132–149. +27 + +Geanakoplos, John, David Pearce, and Ennio Stacchetti. 1989. “Psychological games +and sequential rationality.” Games and Economic Behavior 1 (1): 60–79. +Gneezy, Uri. 2005. “Deception: The Role of Consequences.” 95 (1): 384–394. +Gneezy, Uri, Agne Kajackaite, and Joel Sobel. 2018. “Lying Aversion and the Size of the +Lie.” American Economic Review 108 (2): 419–453. +Greiner, Ben. 2015. “Subject pool recruitment procedures: organizing experiments with +ORSEE.” Journal of the Economic Science Association 1 (1): 114–125. +Hagenbach, Jeanne, and Eduardo Perez-Richet. 2018. “Communication with evidence +in the lab.” Games and Economic Behavior 112 (November): 139–165. +Khalmetski, Kiryl, Bettina Rockenbach, and Peter Werner. 2017. “Evasive lying in strate- +gic communication.” Journal of Public Economics 156 (December): 59–72. +Khalmetski, Kiryl, and Dirk Sliwka. 2019. “Disguising Lies—Image Concerns and Partial +Lying in Cheating Games.” American Economic Journal: Microeconomics 11 (4): +79–110. +Russell, Bertrand. 1923. “Vagueness.” Australasian Journal of Psychology and Philosophy +1 (2): 84–92. +Sánchez-Pagés, Santiago, and Marc Vorsatz. 2007. “An experimental study of truth- +telling in a sender–receiver game.” Games and Economic Behavior 61 (1): 86–112. +Schmeidler, +David. +1973. +“Equilibrium +points +of +nonatomic +games.” +Journal of Statistical Physics 7 (4): 295–300. +Serra-Garcia, Marta. 2018. “Lying in Economics.” Chap. 38 in The Oxford Handbook of +Lying, edited by Jörg Meibauer. Oxford University Press. +Serra-Garcia, Marta, Eric van Damme, and Jan Potters. 2011. “Hiding an inconvenient +truth: Lies and vagueness.” Games and Economic Behavior 73 (1): 244–261. +28 + +Shalvi, Shaul, Jason Dana, Michel JJ Handgraaf, and Carsten KW De Dreu. 2011. “Justi- +fied ethicality: Observing desired counterfactuals modifies ethical perceptions and +behavior.” Organizational Behavior and Human Decision Processes 115 (2): 181– +190. +Sobel, Joel. 2020. “Lying and Deception in Games.” Journal of Political Economy 128 +(3): 907–947. +Tergiman, Chloe, and Marie Claire Villeval. 2022. “The Way People Lie in Markets: +Detectable Vs. Deniable Lies.” Management Science (Forthcoming). +Turmunkh, Uyanga, Martijn J. van den Assem, and Dennie van Dolder. 2019. “Mal- +leable Lies: Communication and Cooperation in a High Stakes TV Game Show.” +Management Science 65 (10): 4795–4812. +Wood, Daniel H. 2022. “Communication-Enhancing Vagueness.” Games 13 (4): 49. +Zhang, Sookie Xue, and Ralph-Christopher Bayer. 2022. “Delegation based on cheap +talk.” Theory and Decision, 1–29. +29 + +Appendices +A +Proofs +Let us first begin with restating the previously known results about the NA-R environment. The +main intuition is that some agents lie when they see a small number, while some others always +report the truthful message. Thus, when a reported number is small, it is safe to believe that +the agent is being honest. The following equilibrium results establish a baseline for comparison +with the other environments. +Lemma 2 (KS LM 1). In any equilibrium in the NA-R environment, there exists a strict positive +probability that agents lie. +Proof. Suppose there exists an equilibrium in the NA-R environment where no agent lies and +the audience believes P(j is truthful) = 1 for any report j. The utility of an arbitrary agent with +the true observation of i < N and the type t is +U(i, j, t) = j − 1(i ̸= j)[t + c(i, j)] + γ · 1. +when the agent reports some j. +As the social identity is constant under the audience’s belief that everyone is truth-telling, +the agent is better off by reporting i + 1 instead of i when +U(i, i + 1, t) − U(i, i, t) = ((i + 1) − i) − (t + c(i, i + 1)) > 0. +Because we assumed c(i, i + 1) < 1, there exists some 0 < t < 1 − c(i, i + 1). +Lemma 3 (KS PR 4; GKS PR 2). In any equilibrium in the NA-R environment, no agent underre- +ports. +Proof. Suppose there exists an equilibrium where an agent lies by reporting a number j below +their true observation i. Then it must be the case ρ(j) > ρ(i). +30 + +It also follows that reporting j must yield a larger utility than reporting i: +(j − i) + γ(ρ(j) − ρ(i)) ≥ t + c(i, j). +We complete the proof by showing that ρ(i) = 1; that is, if any agent choose to report i, +then it must be the case that the report is truthful. +Take another agent with the intrinsic aversion type t who observes some i′ ̸= i. We claim +that U(i′, j, t′) > U(i′, i, t): +(j − i) + [c(i′, i) − c(i′, j)] + γ(ρ(j) − ρ(i)) ≥ c(i′, i) − c(i′, j) + t + c(i, j) > 0 +because of the triangular inequality assumption: c(i, i′) + c(i′, j) ≥ c(i, j). As the choice of this +agent is arbitrary, this is the case for all agents who do not observe i never lies by reporting i; +in turn, this implies ρ(i) = 1, a contradiction. +Lemma 4 (KS THM 1; GKS PR 5). In any equilibrium in the NA-R environment, +i. there exists a threshold 1 < l∗ < N such that +∀j ≥ l∗ ∃i ̸= j, t ∈ � s.t. σj +it > 0 and +∀j < l∗, i ∈ Ω, t ∈ � σj +it = 0; +ii. all agents who observe a value above the threshold report their observed value truthfully. +Proof. Let LΩ +P ⊆ MΩ +P be the set of messages that liars use to lie with positive probability. Let +l∗ = min LΩ +P . By Lemma 2, LΩ +P is nonempty and l∗ is well-defined. Also, by the no-underreporting +condition, we can deduce that ρ(1) = 1 and 1 < l∗. +We now show LΩ +P = {l∗, l∗ + 1,..., N} by contradiction: suppose there exists some elements +of Ω greater than l∗ which is not an element of LΩ +P . Let n be the minimum of such elements, so +that n − 1 ∈ LΩ +P . As ρ(n) = 1, we can see that any agent who lies by reporting n − 1 is strictly +better off by reporting n instead: +U(i, n, t) − U(i, n − 1, t) = (n − (n − 1)) − (c(i, n) − c(i, n − 1)) + γ(ρ(n) − ρ(n − 1)) +> c(n − 1, n) − (c(i, n) − c(i, n − 1)) + γ(ρ(n) − ρ(n − 1)) ≥ 0. +31 + +Now it remains to show that the agents who observe i ∈ LΩ +P reports truthfully. Suppose there +exists j, j′ ∈ LΩ +P such that some agent observing j instead chooses to report j′. That is, +U(j, j′, t) − U(j, j, t) = (j′ − j) − (t + c(j, j′)) + γ(ρ(j′) − ρ(j)) ≥ 0 +for any intrinsic aversion type t. This implies that any agent who lies by reporting j is strictly +better off by reporting j′ instead: +U(i, j′, t) − U(i, j, t) ≥ (j′ − j) + (c(i, j) − c(i, j′)) + γ(ρ(j′) − ρ(j)) +≥ t + c(j, j′) + (c(i, j) − c(i, j′)) > 0 +for any intrinsic aversion type t and any true observation i ̸= j. Therefore, all agents who +observe i ∈ LΩ +P reports truthfully. +Based on the three results, we bring a simple comparison of the behavior of an agent in the +NA-R environment to that in the A-R environment. A thought experiment of choosing an agent +and comparing their behavior in the two environments easily leads to a conjecture that the ab- +sence of the social identity concern should only facilitate more lies. Applying these observation +to the unrestricted communication leads to Proposition 1. We also obtain Lemma 1 from these +three lemmas, which we will use to prove Proposition 4. +Proof of Proposition 1. +Case 1. Anonymous environment: as shown in Proposition 3, all truth-tellers use the optimal +vague message. +Case 2. Non-anonymous environment: suppose there exists an equilibrium where no agent +uses a vague message. Let ρ(j) = 0 for all j ∈ MΩ +V . Given that all messages used with +positive probability are precise, we know all precise messages are used with positive +probability in this equilibrium. We use Lemma 2 and 3 to argue that there exists a +positive probability that agents lie upward. Let l∗ be the threshold defined in Lemma +4. +Let us first argue that there exists agents who observe l∗ − 1 and lie by reporting l∗. +32 + +Suppose not. Then there must exist some l > l∗ such that +U(l∗ − 1, l, t) − U(l∗ − 1, l∗, t) = (l − l∗) + (c(l∗ − 1, l∗) − c(l∗ − 1, l)) + γ(ρ(l) − ρ(l∗)) > 0. +Note that c(l∗ −1, l∗)− c(l∗ −1, l) ≤ 0 because l > l∗ and c is increasing in the distance +between the two arguments. Also, c(l∗ −1, l∗)− c(l∗ −1, l) ≤ c(i, l∗)− c(i, l) ≤ 0 for all +i < l∗ because of the triangular inequality assumption. This implies that for all agents +whose true observation is below l∗ is better off by reporting l instead of l∗. As no agent +would lie by reporting l∗, this is a contradiction to the definition of threshold. +Now consider the agent who observes l∗−1 and lies by reporting l∗. The agent receives +the utility of +U(l∗ − 1, l∗, t) = l∗ − (t + c(l∗ − 1, l∗)) + γρ(l∗). +However, if the agent reports {l∗ − 1, l∗ + 1} instead, the agent receives +U(l∗ − 1,{l∗ − 1, l∗ + 1}, t) = l∗. +That is, the agent is better off by reporting {l∗ − 1, l∗ + 1} when +t > γρ(l∗) − c(l∗ − 1, l∗), +which happens with a positive probability. Furthermore, the above analysis is valid for +all ρ({l∗ − 1, l∗ + 1}) ≥ 0. This is a contradiction to the assumption of an equilibrium +with no vague messages. Therefore, there exists a positive probability that agents use +a vague message in any equilibrium in the NA-UR environment. +Proof of Lemma 1. Let j > i be a part of an equilibrium strategy for an (i, t)-agent in the NA-R +environment: +U(i, j, t) = j − [t + c(i, j)] + γρ(j) ≥ U(i, j′, t) +∀j′ ∈ MΩ +P . +We can infer that +j − [t + c(i, j)] + γρ(j) ≥ i + γρ(i), +33 + +or +(j − i) − [t + c(i, j)] ≥ γ(ρ(i) − ρ(j)). +As no agent underreports in the A-R environment, it suffices to show that this agent does +not tells the truth in the A-R. Suppose not. +UA(i, i, t) = i > UA(i, j, t) = j − [t + c(i, j)]. +This implies ρ(i) < ρ(j), meaning the agent lied in the NA-R both because there were a +monetary gain and a social identity gain. +(j − i) + [c(i′, i) − c(i′, j)] + γ(ρ(j) − ρ(i)) ≥ c(i′, i) − c(i′, j) + t + c(i, j) > 0 +because of the triangular inequality assumption: c(i, i′) + c(i′, j) ≥ c(i, j). As the choice of this +agent is arbitrary, this is the case for all agents who do not observe i never lies by reporting i; +in turn, this implies ρ(i) = 1, a contradiction. Therefore, the agent must lie in the A-R if the +agent sees the same observation i. +Proof of Proposition 4. WTS P(lieNA−R) < P(lieA−R). +We argue by the law of total probability: +P(lieNA−R) = +N +� +i=1 +P(observe i)P(lieNA−R|observe i); +P(lieA−R) = +N +� +i=1 +P(observe i)P(lieA−R|observe i). +Because the probability of observing some i is uniform in both environments, it suffices to show +that the conditional probability of lying in A-R is greater than or equal to that in NA-R for all +true observation i. +We learned that there exists some threshold 1 < l∗ < N in Lemma 4. Thus, conditional on +that an agent observing i ≥ l∗, we have the conditional probability of the agent reporting a lie +as +P(lieNA−R|observe i) = 0; +34 + +for all intrinsic aversion type t, while +P(lieA−R|observe i) > 0 +because of the positive probability that the agent has the type t small enough to report a lie. +Now consider the case of i < l∗. Let Ti be a subset of � such that +P(lieNA−R|observe i, t ∈ Ti) > 0; +P(lieNA−R|observe i, t ̸∈ Ti) = 0. +By Lemma 1, P(lieA−R|observe i, t ∈ Ti) = 1; and P(lieA−R|observe i, t ̸∈ Ti) ≥ 0. Thus, +regardless of P(t ∈ Ti), we have +P(lieNA−R|observe i) ≤ P(lieA−R|observe i) +for all i < l∗. Therefore, P(lieNA−R) < P(lieA−R). +Also, because the monetary payment is a monotone mapping of the reports under the re- +stricted communication, and because any lying takes the form of reporting upward, agents +would earn more monetary payoff on average as the probability of lying is greater in the A- +R. +B +Gender Difference +B.1 +Non-anonymous Environment +In the non-anonymous environment, 28 males and 51 females attended the experiments. We do +not find statistically significant differences between males and females neither on their reports +nor on the length of vague messages. +Under the unrestricted communication, male and female subjects are almost equally likely +to send a precise message. Men are more likely to send a pseudo-optimal vague message, while +women are slightly more likely to send a message with contains a pair or an interval. +35 + +Table 5: Gender difference on Reports in the Non-anonymous Environment +Dependent variable: +Report_P +Exp_V +vague_size +(1) +(2) +(3) +Gender difference +0.0553 +−0.1587 +0.2941 +(0.7291) +(0.2963) +(0.3582) +Constant +6.9804 +8.3379 +2.7059 +(0.4341) +(0.1764) +(0.2133) +Observations +79 +79 +79 +R2 +0.0001 +0.0037 +0.0087 +Figure 4: Message types male subjects used in NA-UR treatments +36 + +60 +Percent (%) +40 +Category +17.86 +14.29 +Interval +28.57 +Not Interval +20 +17.86 +17.86 +3.57 +0 +Precise Pseudo-optimal(vague) Pair +Others +Form of reportFigure 5: Message types female subjects used in NA-UR treatments +37 + +60 +33.33 +Percent (%) +40 +Category +Interval +Not Interval +20 +23.53 +21.57 +13.73 +5.88 +0 +.96 +Precise Pseudo-optimal(vague) Pair +Others +Form of reportB.2 +Anonymous Environment +We used only a subset of anonymous sessions to analyze the gender effect as we could not +identify the participants’ gender in principle. We ran four single sexed sessions (two male-only +sessions and two female-only sessions) to create a gender dummy variable without interfering +the anonymous property of the experiment design. +Under the restricted communication, we do not see statistically significant differences be- +tween males and females neither on their reports nor on the length of vague messages. +Table 6: Gender difference on Reports in the Anonymous Environment +Dependent variable: +Report_P +Exp_V +vague_size +(1) +(2) +(3) +Gender difference +0.1034 +−0.6061 +−0.4867 +(0.7107) +(0.5810) +(0.4461) +Constant +8.000 +8.3290 +2.3143 +(0.4784) +(0.3911) +(0.3003) +Observations +64 +64 +64 +R2 +0.0003 +0.0172 +0.0030 +The results show that, 34.29 % of males and 31.03% of females lied in the non-anonymous +treatment. Therefore, we do not find substantial difference in the lying behavior between them. +In addition, we do not see significant differences in the choice of type and length of vague +messages, as shown by Figures 6 and 7. +38 + +Figure 6: Message types male subjects used in A-UR treatments +39 + +60 +(%) +31.03 +40 +Category +Percent ( +lie +not lie +20 +27.59 +3.45 +13.79 +17.24 +6.9 +0 +Precise +Optimal(vague) +Pair +Others +Form of reportFigure 7: Message types female subjects used in A-UR treatments +40 + +60 +Category +Percent ! +34.29 +lie +not lie +20 +2 86 +2 86 +20 +17.14 +14.29 +8.57 +0 - +Precise +Optimal(vague) +Pair +Others +Form of reportC +Experimental procedure +An experiment session contains the instructions, a preliminary quiz, and two stages of choice +tasks. The subjects receive invitations through the Missouri Social Science Experimental Labo- +ratory(MISSEL)’s ORSEE system. Once registered, they receive a link to a Zoom meeting. After +subjects join the Zoom meeting room, the experimenter reads the instructions, followed by a +preliminary quiz to ensure the understanding of the procedures. In an anonymous session, ev- +eryone including the experimenter has their video off, and the experimenter does not explicitly +take attendance. In order to assure the subjects about anonymity, subjects are asked to use a +screen name. The experimenter emphasizes that the screen name is used solely for data analysis +and that they cannot associate the screen name with their true identity. In an non-anonymous +session, on the other hand, the experimenter has their video on and asks the subjects to turn +their video on as well. Also, the experimenter takes attendance with the experiment roster. +The subjects are allowed to participate in the main experiment only after successfully passing +the preliminary quiz. Each subject has three chances to attempt the quiz before turned away. +The experimenter provides a link to the main experiment website via the chat window of the +Zoom meeting upon the successful completion of the quiz. +At the beginning of the main experiment part, subjects enter their identification information. +In the anonymous treatment, subjects are asked to enter their screen name; whereas, in the non- +anonymous treatment, subjects are asked their name and student ID. +After entering the identification information, the subjects observe the realization of a ran- +dom integer uniformly distributed between 1 to 10 and later asked to report the number. In the +anonymous treatment, the number is generated within the experiment software. In the non- +anonymous treatment, subjects click on a link to a Google search result for the phrase “random +number between 1 to 10.” +In the reporting screen, subjects are presented with ten boxes labeled from 1 to 10 and +report by selecting the numbered boxes. In a precise stage (Figure 8), which allows for precise +messaging only, a subject may select only one box at a time, whereas in a vague stage (Figure +9), they may select multiple boxes. Subjects are told that this selection is deemed as a statement +that the numbers represented by the selected box include the number they observed. +In the case of a precise message, the subject is paid the equivalent in dollars to the number +she selected divided by two. In the case of a vague message, the computer randomly chooses +41 + +Figure 8: A screenshot of the experiment software displaying the precise-message stage +one number from the submitted selection of numbers and pay the equivalent in dollars to the +randomly chosen number divided by two. +After subjects complete both precise and vague stages, a confirmation screen reviews their +compensation information. In the anonymous treatment, the experimenter bulk-purchases Ama- +zon gift card to match the earnings for each screen name. The experimenter provides a third +party (Department of Economics) with a list of the gift codes and matching screen names, ex- +cluding the amount associated with each code. The subjects visit an external website and re- +trieve their compensation in the form of the gift code by entering their screen name. This pro- +cess is to ensure the anonymity of the experiment session. The experiment can map between +responses and screen names, but not between screen names and true identities. The third party +can only map between true identities and screen names, but not between true identities and +their corresponding responses. In the non-anonymous treatment, they are asked to enter their +email address and taxpayer information to receive the Amazon gift card directly emailed to +them. +42 + +XM Preview - Online Survey Softwa X +→ +0 https:/wustl.az1.qualtrics.com/fe/preview/SV_2ggP5vixBroObNOF?Q_SurveyVersionID=current&Q..:。☆ +三 +Restart Survey +PlaceBookmark +Mobile view off +Toolsv +:Washington +University in St.Louis +Stage 2 +Whatnumberdidyouobserve?Pleaseselectonenumber. +2 +3 +4 +5 +6 +9 +10 +PoweredbyQualtricsFigure 9: A screenshot of the experiment software displaying the vague-message stage +43 + +XM Preview - Online Survey Softwa X +→ +0 https:/wustl.az1.qualtrics.com/fe/preview/SV_2ggP5vixBrObNOF?Q_SurveyVersionID=current&Q..。☆ +三 +Restart Survey +PlaceBookmark +Mobile view off +Tools +: Washington +University in St.Louis +Stage 1 +Whatnumberdidyouobserve?Youmayselectasetof numbersbyclickingmultiple +numbers.Afteryousubmityourselection,thecomputerwillrandomly chooseonefromthe +setof numbersyouselected. +2 +3 +4 +5 +9 +7 +8 +9 +10 +Powered by Qualtrics D +Experimental Instructions +D.1 +Instructions for Anonymous session (Zoom session; no video) +Welcome to the experiment! Thank you very much for participating today, and I will first walk +you through the instructions for the experiment. +We need your full attention during the experiment. If you have trouble with hearing the +audio or seeing the shared screen, please let me know. Do not turn the video on during the +experiment. If you have any questions during the instructions, please raise hand so that I can +unmute you. Your question will be answered out loud, so everyone can hear. +The experiment you will be participating in today is an experiment in individual decision +making. At the end of the experiment, you will be paid an Amazon gift card. You will receive +the show-up fee of $2 for completing the experiment, with the additional amount that depends +on your decisions and on chance. The details of the compensation will be described later. All +instructions and descriptions that you will be given in the experiment are accurate and true. In +accordance with the policy of this lab, at no point will we attempt to deceive you in any way. +I would like to first point out that we want to ensure this experiment is conducted anony- +mously, meaning that we cannot connect the responses recorded in this experiment to any par- +ticular individual who participated in this research. Qualtrics, the survey platform we are using, +provides an option for the researchers to not collect any personal information, such as IP address +or geographic location of the participants, for anonymous surveys. Also, your response will be +recorded with a SCREEN NAME. You will be asked to choose a screen name that is at least 8 +characters in length using letters, numbers, and underscore. This SCREEN NAME is only used +in data analysis and distributing your Amazon gift card after the experiment. We cannot and +will not attempt to associate SCREEN NAMEs to any particular individual. +I will now describe the main features of the experiment. First, there is a short quiz to ensure +your understanding of the procedures. You will be able to repeat the quiz if you make mistakes. +You will have three chances to attempt the quiz. If you fail to get all questions correct after +three attempts, you may not participate in the main experiment. Even in such a case, please +remain connected to the Zoom session until everyone finishes, and you will receive the show-up +fee for today’s experiment. The main part of the experiment after the preliminary quiz consists +of two STAGES. In each STAGE you will observe a number that we ask you to remember and +later report to us. The number you report to us determines how much money you will be paid. +44 + +At the end of the experiment, a confirmation screen will summarize today’s experiment and +provide the information to retrieve your payment. I will give you more details about each step. +At the end of this instructions, we will first provide a link for the preliminary quiz using the +chat window. This is a screenshot of the webpage. Please choose a screen name, and make +sure you keep this screen name. After you successfully complete the preliminary quiz, you will +be reminded of your screen name once again. We recommend you copy and paste the screen +name. We cannot recover this information for you, and you will not be able to receive your +compensation without the correct screen name. Please wait while everyone else finishes the +quiz. Once everyone finishes, we will provide another link for the main experiment using the +chat window. This is a screenshot of the first page of the main experiment. Make sure you use +the same screen name you used in the preliminary quiz. You may not receive your compensation +if the screen names do not match. As previously mentioned, your main task today is to observe +a number that we ask you to remember and later report to us. The observation process is +identical in both STAGES. At the beginning of both STAGES, the computer will randomly draw a +number between 1 to 10. The probabilities are equal across the numbers; that is, each number +is chosen with the same probability of one-tenth. We ask you to remember the number and +report on the next screen. However, the way you can report differs between STAGES. In one +STAGE, you are allowed to select one number after you observe the draw. This is a screenshot +of the experiment stage where you can select one number. By one number, we mean that you +may click only one number on the screen. We will interpret this selection as a statement that +the number you observed is the number you selected, and we will regard this number as your +report. The number you report determines how much money you will be paid. You will be +paid the equivalent in dollars to the number you report divided by 2. In other words, if you +report “1”, you receive 50 cents. If you report “2”, you receive $1, if you report “3”, you receive +$1.50 and so on. A confirmation screen after your report will help you review your selection +and the corresponding payment. In another STAGE, you are allowed to select a set of numbers +after you observe the draw. This is a screenshot of the experiment stage where you can select +a set of numbers. By a set of numbers, we mean that you may click multiple numbers on the +screen. For instance, you may choose to click on four numbers, one number, two numbers, or +even all ten numbers. If you select multiple numbers, we will interpret it as a statement that the +number you observed is one of the numbers you selected. If you select a single number, we will +interpret it as a statement that the number you observed is the number you selected. After you +45 + +submit your selection, the computer will randomly choose one number from the set of numbers +you selected. We will regard this randomly chosen number as your report. Again, the number +you report determines how much money you will be paid. You will be paid the equivalent in +dollars to the randomly chosen number from the set of numbers you selected divided by 2. If the +randomly chosen number is “1”, you receive 50 cents. If “2”, you receive $1, if “3”, you receive +$1.50 and so on. A confirmation screen after your report will help you review your selection +and the corresponding payment. +The order of the two STAGES is randomly determined. In other words, it is equally likely +that you first participate in the STAGE allowing only a single number and then participate in the +STAGE allowing a set of numbers, or first participate in the STAGE allowing a set of numbers +and then participate in the STAGE allowing only a single number. In any case, you will play +each STAGE only once. +After the completion of both STAGES, a final review screen will provide the information to +receive your Amazon gift card code. This is a screenshot of the review screen. This includes your +SCREEN NAME you entered, the amount you will receive, and a randomly generated passcode. +Because the experiment is anonymous, we have no means to recover this information for you. +Please make sure you either print or take the screenshot of this page for your record, because it +is very important when you retrieve your compensation. +In the final review screen, you will find a link of a google form. To receive your amazon +gift card you have to copy and paste the link in another browser and then fill your personal +information. The google form is created by a staff of the department of economics, who won’t +have access to the data of this experiment.The experimenter only has a list of SCREEN NAMES +and the amount associated with them. The experimenter will never be able to access your +personal information. This is a screenshot of the google form which you will have to fill to +receive your amazon gift card. +Washington University in St. Louis recommends student subjects to report their taxpayer +identification information for tax purposes. If you are an international student and do not have +the taxpayer identification information, please indicate so by entering ‘Foreign’ in the form. If +you do not have or do not wish to provide the identification information, please indicate that +you would like to opt out by entering ’Refuse’ in the form. +We are sorry for the inconvenience that we are not able to email you with the gift code +directly. This payment process is to ensure anonymity in this experiment, and we appreciate +46 + +your understanding that the anonymity of the reports constitutes a crucial component of our +research. +This is the end of the instructions. If you have any questions, please raise hand. Otherwise, +I will provide the link via the chat window. Please copy and paste the link to your browser to +generate the SCREEN NAME. +D.2 +Instructions for Non-Anonymous session (Zoom session; video) +Welcome to the experiment! Thank you very much for participating today. Before we start, I will +go over the roster to take attendance to make sure I have everyone registered for the session. +During the attendance please turn your video on. +I will now walk you through the instructions for the experiment. We need your full attention +during the experiment. If you have trouble with hearing the audio or seeing the shared screen, +please let me know. If you have any questions during the instructions, please use the hand- +raising feature of Zoom and your question will be answered out loud, so everyone can hear. +The experiment you will be participating in today is an experiment in individual decision +making. At the end of the experiment, you will be paid an Amazon gift card. You will receive the +show-up fee of $2 for completing the experiment, with the additional amount that depends on +your decisions and on chance. The details of the payment will be described later. All instructions +and descriptions that you will be given in the experiment are accurate and true. In accordance +with the policy of this lab, at no point will we attempt to deceive you in any way. +This is a screenshot of the first page in the main experiment. At the end of the instructions, +I will provide the link to the experiment using the chat window. Please copy and paste the link +to your browser. The first screen will ask your identification information – your first and last +name and your student ID. +After you enter your information, you will proceed to the next screen and take a short quiz +to ensure your understanding of the procedures. You will be able to repeat the quiz if you make +mistakes. You will have three chances to attempt the quiz. If you fail to get all questions correct +after three attempts, you may not participate in today’s experiment. In such case, you will only +receive the show-up fee for today’s experiment. +The main part of the experiment consists of two STAGES after the preliminary quiz. In each +STAGE you will observe a number that we ask you to remember and later report to us. The +47 + +number you report to us determines how much money you will be paid. At the end of the +experiment, a confirmation screen will summarize today’s experiment and provide the informa- +tion to retrieve your payment. I will give you more details about the observation, reporting, and +payment processes. +The observation process is identical in both STAGES. At the beginning of both STAGES, +we will provide a link to Google page that randomly generates a number between 1 and 10. +The probabilities are equal across the numbers; that is, each number is chosen with the same +probability of one tenth. We ask you to open the link, remember the number, close the Google +page, and report the number on the next screen. This is a screenshot of an example Google +page. You will see that this is a search result for ‘random number between 1 and 10’, and +the page displays a randomly generated number that matches the search phrase. Do not click +on the ‘generate’ button on the Google page, because the number you see is already randomly +generated. Any additional generation only distorts the statistical properties of the experiment. I +will now demonstrate how this Google page works. You will find this link during the experiment, +and this is equivalent of opening a google page and typing in “random number between 1 and +10.” When you open the link, a new window pops up. As you can see, this number is already +randomly generated and you should not generate the number again. +The way you can report differs between STAGES. In one STAGE, you are allowed to select +one number after you observe the randomly generated number. This is a screenshot of the +experiment stage where you can select one number. By one number, we mean that you may +click only one number on the screen. We will interpret this selection as a statement that the +number you observed is the number you selected, and we will regard this number as your +report. The number you report determines how much money you will be paid. You will be paid +the equivalent in dollars to the number you report divided by 2. In other words, if you report +“1”, you receive 50 cents. If you report “2”, you receive $1, if you report “3”, you receive $1.50 +and so on. A confirmation screen after your report will help you review your selection and the +corresponding payment. In another STAGE, you are allowed to select a set of numbers after you +observe the randomly generated number. This is a screenshot of the experiment stage where +you can select a set of numbers. By a set of numbers, we mean that you may click multiple +numbers on the screen. For instance, you may choose to click on four numbers, one number, +two numbers, or even all ten numbers. If you select multiple numbers, we will interpret it as +a statement that the number you observed is one of the numbers you selected. If you select a +48 + +single number, we will interpret it as a statement that the number you observed is the number +you selected. After you submit your selection, the computer will randomly choose one number +from the set of numbers you selected. We will regard this randomly chosen number as your +report. +Again, the number you report determines how much money you will be paid. You will be +paid the equivalent in dollars to the randomly chosen number from the set of numbers you +selected divided by 2. If the randomly chosen number is “1”, you receive 50 cents. If “2”, you +receive $1, if “3”, you receive $1.50 and so on. A confirmation screen after your report will help +you review your selection and the corresponding payment. +The order of the two STAGES is randomly determined. In other words, it is equally likely +that you either participate in the STAGE allowing a single number first and then participate in +the STAGE allowing a set of numbers or participate in the STAGE allowing a set of numbers first +and then participate in the STAGE allowing a single number. In any case, you will play each +STAGE only once. +After the completion of both STAGES, a final review screen will summarize today’s exper- +iment. This is a screenshot of the review screen. The last screen will ask your email address +to receive the Amazon Gift Code of the amount that corresponds to your responses. We will +directly send you Amazon gift code to the email address you provide. Please allow us a few +hours after the completion of the experiment to validate the data and send the email. In addi- +tion, Washington University in St. Louis recommends student subjects to report their taxpayer +identification information for tax purposes. If you are an international student and do not have +the taxpayer identification information, please indicate so by entering ‘Foreign’ in the form. If +you do not have or do not wish to provide the identification information, please indicate that +you would like to opt out by entering ’Refuse’ in the form. +This is the end of instructions. If you have any questions, please raise your hand. Otherwise, +I will provide the link via the chat window. Please copy and paste the link to your browser and +start the experiment. +49 + diff --git a/PtAyT4oBgHgl3EQfhPgp/content/tmp_files/load_file.txt b/PtAyT4oBgHgl3EQfhPgp/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..2b121f48f38ab17cba50f435c4ec033d80d1690a --- /dev/null +++ b/PtAyT4oBgHgl3EQfhPgp/content/tmp_files/load_file.txt @@ -0,0 +1,1085 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf,len=1084 +page_content='Lying Aversion and Vague Communication: An Experimental Study Keh-Kuan Sun Stella Papadokonstantaki†‡ December 31, 2022 The latest version of the draft is available here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Abstract An agent may strategically employ a vague message to mislead an audience’s belief about the state of the world, but this may cause the agent to feel guilt or negatively impact how the audience perceives the agent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Using a novel experimental design that allows participants to be vague while at the same time isolating the internal cost of lying from the social identity cost of appearing dishonest, we explore the extent to which these two types of lying costs affect communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We find that participants exploit vagueness to be consistent with the truth, while at the same time leveraging the imprecision to their own benefit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' More participants use vague messages in treatments where concern with social identity is relevant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, we find that social identity concerns substantially affect the length and patterns of vague messages used across the treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Keywords: Lying;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Vagueness;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Communication;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Experiments;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Behavioral Economics JEL codes: C91, D90 Smith Institute for Political Economy and Philosophy, Chapman University.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Email: sun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='k@wustl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='edu †Department of Economics, Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Louis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Email: p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='stella@wustl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='edu ‡We thank Guangying Chen for her contribution to the early version of this paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We also thank Brian Rogers, Marta Serra-Garcia, Agne Kajackaite, Cynthia Cryder, John Nachbar, Jonathan Weinstein, Marcus Berliant, Sangmok Lee, Paulo Natenzon, John Rehbeck, Erik Kimbrough, and seminar participants at Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Louis, the ESA Job-Market Candidates Seminar Series, the Berlin School of Economics, the Korea Informa- tion Society Development Institute, the Tokyo Institute of Technology, the ECONtribute Summer Workshop 2021 on Social Image and Moral Behavior, the 2021 North American Summer Meeting of the Econometric Society, the 2021 Asian Summer Meeting of the Econometric Society, and the 2021 ESA Global Online Around-the-Clock Meetings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='00372v1 [econ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='GN] 1 Jan 2023 1 Introduction Communication has been studied extensively in economics in recent decades.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In particular, the types of messages sent by an agent in the communication process and their impact on an audience’s beliefs have been important subjects of game theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Standard economic models predict that a sender will choose a message that yields the greatest benefit from the receiver’s action, even when such a message is a lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' It has however been observed that senders often choose to be evasive or use vague messages instead of telling maximal lies despite the possibility that doing so will yield a lower benefit than lying would.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This raises two questions regarding why and when a sender prefers using a vague message over telling a maximal lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This issue is crucial to achieving a unified understanding of misleading behavior in many applications, including problems associated with public good provision (Serra-Garcia, Damme, and Potters 2011), sender-receiver disclosure games (Hagenbach and Perez-Richet 2018), and persuasion games (Deversi, Ispano, and Schwardmann 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' To answer these questions we must carefully examine the behavioral aspects of lying and vague communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' First, as demonstrated by recent developments in the literature on lying behavior (Serra-Garcia 2018;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Fischbacher and Föllmi-Heusi 2013;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Abeler, Nosenzo, and Ray- mond 2019), most people exhibit a non-trivial degree of lying aversion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If an agent intrinsically prefers being honest, then the internal cost of lying should inhibit the use of lies and encour- age the use of vague yet relatively truthful messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Second, a message affects not only an audience’s belief about the state of the world but also their belief about how honest an agent is.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' When the agent cares about their social identity as an honest person, this external concern should impact the message they choose to communicate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After all, “a vague belief has a much better chance of being true than a precise one, because there are more possible facts that would verify it” (Russell 1923).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The nontrivial cost of lying suggests that communication is no longer cheap talk and the misleading messages may bear strategic significance as a credible signal to the audience about the agent’s behavioral types.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' On the other hand, such a strategic incentive may also influence behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thus, a rational agent must balance the degree of truthfulness and vagueness of the message communicated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In this paper, we present a model of a cheating game in which an agent may report a vague (set-valued) message or a precise (single-valued) message to an audience after privately ob- serving the state of the world.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The agent’s utility depends on the monetary payoff and their 2 truth-telling preferences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The monetary payoff is determined solely by the reported message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We study two separable motivations for honesty: the internal motivation for being honest and an external concern related to social identity for being seen as honest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We isolate the two costs by employing an anonymous environment where the audience cannot identify an agent with a message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Based on the model, we hypothesize that people use vague messages to reduce their internal guilt and increase their monetary payoffs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The absence of the social identity concern in the anonymous environment should lead to more straightforward profit-maximizing message choices, while the prediction is more opaque in the non-anonymous environment because of the multiplicity of equilibria introduced in a richer message space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' To test hypotheses pertaining to vague communication with lying aversion, we compare treatments from an online experiment in which subjects face a variant of the Fischbacher and Föllmi-Heusi (2013) type of reporting task (hereafter “FFH").' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In this experiment, subjects pri- vately observe an integer drawn randomly from a uniform distribution ranging from 1 to 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The subjects are asked to report the number to the experimenter, and their monetary payment increases with the reported number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The basic idea of the experiment is that the discrepancy be- tween the maximum numbers they could have reported and the actual numbers reported should capture subjects’ aversion to lying or, by the same token, their preferences for truth-telling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We generalize the FFH model by allowing subjects to transmit set-valued messages to understand the effect of vagueness in communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In our setup, we allow subjects to be vague by report- ing multiple numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experimenter then chooses one number randomly from the reported numbers and pays the subject accordingly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We use a 2x2 experimental design to test the predictions and, in particular, to separate the intrinsic cost of lying from the social identity cost of appearing dishonest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' First, there are two types of experimental sessions that represent the variation in the anonymity of agents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous session, the responses are recorded under screen names so that the experimenter cannot map a subject’s identity to their response.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the non-anonymous session, on the other hand, the experimenter knows each subject’s response.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The social identity cost should not af- fect their responses in the anonymous treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, the difference in behavior between anonymous and non-anonymous treatments enables us to capture the impact of the social iden- tity cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Second, a subject confronts two reporting tasks within each session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The observation process is identical and independent for the two tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' When the subject reports, however, the set of available messages differs between the two tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the task with restricted com- 3 munication, the set is restricted to single-valued messages only.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the task with unrestricted communication, the subjects are allowed to use both single-valued and set-valued messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experimental data indicate that most subjects use vague messages and that the means of the numbers they report are higher on average when vague messages are allowed but the way they use vague messages differs across the treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' First, in the anonymous treatments, we observe less frequent lying behavior when vague messages are allowed compared with when they are not allowed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As anonymity should suppress subjects’ concern with social identity, the reduction in lying behavior suggests that a vague yet truthful message reduces the internal cost of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Yet, at the same time, they report higher numbers on average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, subjects exploit vagueness to avoid uttering falsehoods while leveraging the imprecision to their own benefit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Second, subjects use vague messages in the non-anonymous treatment more often than in the anonymous counterpart.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The increased frequency of vague messages and the relevance of the concern with social identity suggest that a message’s vagueness affects message choices through both the internal and external costs of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Furthermore, we find that the pattern of reported messages differs substantially between the two types of sessions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' When subjects use vague messages in the anonymous treatment, most do not refrain from the most obvious forms: that is, they report a combination of their true observations and the maximum number (10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We find, however, that only a much smaller fraction of subjects using vague messages employ such obvious messages in the non-anonymous treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' It is possible that, in the non-anonymous treatments, subjects may expect those obvious messages to be interpreted by the audience as lies, which will discredit their social identities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Our paper bridges the literature on lying behavior with a broader set of studies that involve vague communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Existing literature on vague communication focuses on the strategic use of vague messages without considering the behavioral aspect of lying aversion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For instance, Serra-Garcia, Damme, and Potters (2011), analyzing a public-goods provision game between two agents with information asymmetry, assume that a set-valued vague message would incur a much lower lying cost than a precise, single-valued outright lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Our experiment tests this assumption and provides further insight into the cost of lying with respect to the social identity concern as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Agranov and Schotter (2012) show that, in coordination games with multiple equilibria, it might be beneficial for a benevolent sender to use vague communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Zhang and Bayer (2022) show that, in a delegation game, social welfare is higher when messages are intervals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Wood (2022) concludes that transmitted information is more accurate when senders 4 have the option to send either precise or vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Deversi, Ispano, and Schwardmann (2021) study the strategic use of vagueness in a voluntary disclosure game.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' They allow subjects to send an interval that contains their type, and this formal message structure is similar to that in our design.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Yet, there are two distinctive features that separate our design from theirs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' First, we provide subjects with greater flexibility in choosing messages by allowing them to choose any subset of the state space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We also add a between-subject variation on anonymity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This flexibility and the variation of anonymity allow us to explore the extent to which attitudes toward lying and misleading behavior affect the use of vague messages in conjunction with strategic motives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Conceptually, we view lying behavior as an optimization process in which a rational agent chooses the optimal degree of dishonesty, by balancing the marginal monetary benefit of lying with the non-monetary costs that depend on the likelihood that an individual is found to be lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In a more specific context involving the nature of lying behavior, Sobel (2020) provides a comprehensive framework in which to understand lying and deception in games.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' He distin- guishes between three properties —the form, the interpretation, and the consequence —of a message, and our model closely follows the framework suggested in that paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We would like to particularly mention the notions of malleable lies in Turmunkh, Assem, and Dolder (2019) and deniable lies in Tergiman and Villeval (2022) which are closely related to our empirical findings pertaining to the use of vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' However, while the definitions of malleable and deniable lies hinge on their interpretations, we define lying strictly by reference to the for- mal property of a message: a message is a lie only when it does not represent the true state of the world.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The agents in our model face both monetary and, potentially, psychological conse- quences for their message choices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Furthermore, through an analysis of equilibrium strategies, we model how the interpretation of a message affects an agent’s decision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This structure allows us to decompose the internal cost of lying from the concern for social identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Structurally, we study lying behavior within the framework of signaling games.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Since Craw- ford and Sobel (1982), a stream of research has tested the assumption that lying is costless, which implies that individuals will lie whenever there is a material incentive to do so.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The empirical evidence continues, however, to make the case against this notion (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Dickhaut, McCabe, and Mukherji (1995), Blume et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (1998), Gneezy (2005), Sánchez-Pagés and Vor- satz (2007), Shalvi et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (2011) and many more).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Many studies argue that individuals clearly express aversion to lying and accept significantly lower payoffs than the theories have predicted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An empirical consensus regarding the FFH experiments is that people do not lie to the ex- 5 tent that they could, preferring to tell only minor lies, potentially because lying is costly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As the monetary payoff in such an experiment is independent of the drawn number, subjects should report non-maximal numbers only if aversion to lying is present, for otherwise this becomes a simple case of a cheap-talk game.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Abeler, Nosenzo, and Raymond (2019) combine data from 90 such experiments that describe the average reporting behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' They find that the behavior is indeed bounded away from the maximal report but also departs from a complete truth-telling scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, by ruling out other explanations for truth-telling which are popular in the literature (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' inequality aversion or seeking a reputation for not being greedy), they conclude that a preference for being seen as honest and a preference for being honest are the main moti- vations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As Abeler, Nosenzo, and Raymond (2019) note in their conclusion, however, the FFH paradigm has focused on subjects reporting a single number and excludes lies by omission or vagueness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Our paper extends the FFH paradigm by allowing vague communication using set- valued messages and contributes to our knowledge of how a message space in communication plays a role in an agent’s reporting decision when lying entails a cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The studies that are closest to our work are Gneezy, Kajackaite, and Sobel (2018) and Khal- metski and Sliwka (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In these studies, an agent cares not only about obtaining a monetary payoff but also about whether lying takes place as well as how others interpret their report, a finding that is consistent with the empirical findings of the aforementioned study by Abeler, Nosenzo, and Raymond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' These papers adopt the FFH paradigm and simplify the sender/receiver structure of a signaling game into a cheating game between reporting agents and an observing audience.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This idea, in conjunction with the assumption of a non-atomic game, minimizes the role of the receiver and allows a sharper focus on that of the sender.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Moreover, this simplification makes it possible to develop a unique characterization of off-path beliefs in the context of mild conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' While we inherit their assumptions of these assumed motivations for truth-telling —internal guilt and the external concern with one’s social identity —extending the message space to include set-valued messages yields more opaque predictions with multiple equilibria and off-path beliefs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We overcome this difficulty by introducing an anonymous environment to hold the impact of social identity constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We observe that, as long as a message can remain even remotely truthful, most agents exhibit no hesitation in increasing their monetary payoff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This observation is analogous to behavior reflecting “moral wiggle room” (Dana et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2007), the notion that people engage more resolutely in self-interested behavior when excuses for selfish actions are available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' On the other hand, 6 there exists a group of truth-tellers in our experiment whose reports are not only truthful but also precise, thus possibly relinquishing opportunities to obtain monetary gains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The non-trivial size of this group suggests another motivation for truth-telling in addition to internal and external concerns regarding honesty.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Candidates might include a concern for the consequences of choices made or concern with good intentions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The remainder of the paper is organized as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In Section 2 we define our terms and present the model setting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Section 3 provides theoretical analysis, while in Section 4 we present the experiment hypotheses based on the theoretical predictions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In Section 5 we describe our experiment design, in Section 6 we summarize the experimental outcomes, and Section 7 con- cludes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We list all the proofs and additional details regarding the experiment in the appendices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2 Model 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1 A model of lying aversion with vague communication We study lying aversion with vague communication by considering a variant of the Fischbacher and Föllmi-Heusi cheating game with a population of agents and one audience.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An agent pri- vately observes the state of the world i ∈ Ω, where Ω = {1,2,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N} is finite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We assume i is drawn i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' from a uniform distribution over Ω across agents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Each agent has a private type t that represents their intrinsic aversion to lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We also assume t is i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' across agents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Its CDF F(t) is strictly increasing, continuous, and has support [0, T].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Denote an agent who observes the state i and has the intrinsic aversion type t as a type (i, t) agent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An agent reports a message J to the audience after observing the true state of the world i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This reporting takes place only once and there are no repeated interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Like Gneezy, Kajackaite, and Sobel (2018)(GKS) and Khalmetski and Sliwka (2019)(KS), we assume that an agent’s utility consists of three components: a monetary payoff, an internal concern with being honest, and a social identity for being seen as honest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We use the term ‘social identity’ to distinguish it from the typical understanding of reputation in a dynamic game, as we are modeling a one-shot game.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Formally, we write a type (i, t) agent’s utility for reporting a message J as: U(i, J, t) = ¯π(J) ���� monetary payoff −1(i ̸∈ J)[t + c(π({i}), ¯π(J))] � �� � internal guilt + γρ(J) � �� � external social identity .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (1) 7 The message J is a nonempty1 subset of the state space: J ∈ MΩ ≡ 2Ω \\ �.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An important distinction between our model and GKS and KS is that in our model the message space admits set-value messages instead of being isomorphic with the state space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This generalization allows messages to be categorized in multiple ways.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We define the relevant terms below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Definition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A message J is truthful if i ∈ J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A message is a lie if it is not truthful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Definition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A message is called precise if it is a singleton set;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' otherwise it is vague.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let MΩ P ≡ {{1},{2},.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=',{N}} and MΩ V ≡ MΩ \\ MΩ P denote the sets of precise messages and vague messages, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For example, if Ω = {1,2,3}, the set of possible precise messages is {{1},{2},{3}}, and the set of possible vague messages is {{1,2},{2,3},{1,3},{1,2,3}}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2 Note that our definition of lying depends purely on the form of a message, as in the framework proposed in Sobel (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The consequences of a message choice are reflected through effects on the components of the agent’s utility.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The monetary payoff for an agent maps a message to the agent’s utility: π : MΩ → �.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For simplicity, we assume that π(J) is a uniform draw over J and that the agent is a risk-neutral, expected-utility maximizer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let us denote ¯π(J) = E[π(J)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Note that, when a message is precise, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' J = {x} with x ∈ Ω, the monetary payoff π(J) is simply x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition to receiving a monetary payoff, the agent also has two distinct motivations for being honest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' First, the agent has an internal motivation for being honest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' When the agent’s re- port J is not truthful, that is, i ̸∈ J, the dishonesty incurs an internal cost, t+c(π({i}), ¯π(J)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The agent’s private type t captures their sensitivity to the intrinsic (fixed) cost of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The function c(π({i}), ¯π(J)) : �×� → � represents the variable cost of lying depending on the size of the lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The size of the lie is measured as the ex-ante difference between the monetary payoff for a report J and that of the true and precise report {i}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We assume that i) c(·) ≥ 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' ii) c(π({i}),π({i})) = 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' iii) c(π({i}), ¯π(J)) is weakly increasing in |π({i}) − ¯π(J)|;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' iv) c(π({i}),π({i}) + 1) < 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3 and v) c(π({i}), ¯π(J)) + c( ¯π(J), ¯π(K)) ≥ c(π({i}), ¯π(K)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An empty set is often interpreted as silence and plays an interesting role in the literature on vague communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We abstract away from silent messages to compare cases where lying aversion is accom- panied by vague language with cases where lying aversion is accompanied by precise language.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The fully vague message Ω ∈ MΩ is analogous to the notion of evasive lying in Khalmetski, Rock- enbach, and Werner (2017) by which the sender falsely states to have not observed the state of the world.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This assumption excludes the trivial case where the variable cost of lying is so high that no agent chooses to lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 8 The agent also has an external motivation to be ascribed a social identity for being seen as honest;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' that is, the agent’s utility depends on the audience’s belief about how honest the agent is.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We assume that the audience is a rational Bayesian who forms a posterior belief based on the agent’s report J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This belief, in turn, depends on the agent’s mixed strategy in equilibrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A type (i, t) agent’s mixed strategy is a mapping σ : Ω × [0, T] → [0,1]2N−1 such that σ(i, t) = � σ{1} it ,σ{2} it ,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=',σ{1,2,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=',N} it � (2) where σJ it is the probability which the intrinsic aversion type t agent with true observation i assigns to the report J ∈ MΩ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In equilibrium, the audience’s posterior belief about whether an agent’s report J is truthful is computed using Bayes’ rule: ρ(J) = P(agent is honest ∧ agent reports J) P(agent reports J) = � k∈J( � T 0 σJ ktd f (t)) �N k=0( � T 0 σJ ktd f (t)) (3) We normalize the posterior belief ρ in terms of the agent’s utility by parameter γ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The pa- rameter measures the agent’s sensitivity to their social identity reflected in reporting message J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We further assume that i) γ is homogeneous across agents and is common knowledge;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' more- over, ii) N + γ < T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The latter condition ensures that F(N + γ) < 1, or there always exists a positive mass of agents who will be truth-telling for any observation i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Because the agent’s payoff depends on the audience’s belief, we adopt the notion of se- quential equilibria in an induced psychological game in the sense of Battigalli and Dufwenberg (2009) and Geanakoplos, Pearce, and Stacchetti (1989).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An equilibrium is a set of mixed strategies and beliefs satisfying the following conditions: ∀(i, J, t) : σJ it > 0 only if J ∈ argmax J′ U(i, J′, t), (4) ∀(i, t) : � J∈MΩ σJ it = 1, (5) ∀J : ρ(J) = � k∈J( � T 0 σJ tkd f (t)) �N k=0( � T 0 σJ tkd f (t)) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (6) 9 As noted by GKS, the existence of an equilibrium follows Schmeidler (1973) in treating each type (i, t) as a player.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2 The communication environment and the anonymity of agents The baseline model assumes no restriction on message space MΩ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If we restrict message space to MΩ P , on the other hand, we obtain a model with restricted communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We refer to the baseline model as the model with unrestricted communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the risk of abusing the notation, let us denote a message as j in the model with restricted communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Furthermore, for a more comprehensive understanding of the relationship between the use of vague messages and the costs of lying, it is helpful to isolate their effects on internal guilt from their effects on external social identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We can achieve this by employing an anonymous environment where the audience cannot identify an agent with a message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This implies that the agent’s report does not alter the audience’s belief, and hence the agent’s social identity remains constant independent of their reporting choice in the anonymity-preserving environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' On the other hand, we call an environment non-anonymous when the audience can associate a message with its sender.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Formally, we write an agent’s utility in the anonymous environment as: UA(i, J, t) = ¯π(J) − 1(i ̸∈ J)[t + c(π({i}), ¯π(J))].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (7) We thus create four distinct environments by varying the restriction on the message space and the anonymity of the agents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In these environments, the agent is non-anonymous with re- stricted (precise only) communication (NA-R), non-anonymous with unrestricted (potentially vague) communication (NA-UR), anonymous with restricted communication (A-R), and anony- mous with unrestricted communication (A-UR).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Note that the NA-R environment corresponds to the standard FFH model where an agent can send only precise messages and the audience can identify a message’s sender.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Non-anonymous Anonymous Restricted NA-R A-R Unrestricted NA-UR A-UR Table 1: Four environments 10 3 Analysis In this section we describe the equilibria and agents’ behavior in each of the four environments as defined above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We begin by arguing that we should expect to observe vague messages used at positive probabilities in any equilibrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In any equilibrium under unrestricted communication, there exist types of agents who use at least one vague message with positive probability in their mixed strategies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' See Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The intuition behind Proposition 1 is that there exist agents with moderately high intrinsic lying aversion type t who would prefer to be truthful while increasing the monetary payoff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, in their messages, they include higher numbers apart from their true observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For these agents, such messages strictly dominate any precise lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In comparison with the unrestricted communication environment in which we are inter- ested, a particularly useful element when analyzing a restricted communication environment is the assumption that there always exists a positive mass of truth-tellers for each observation i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' While this assumption is sufficiently plausible in many contexts, it aids us especially in elimi- nating less interesting equilibria by uniquely characterizing off-path beliefs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thus, we are able to make sharp equilibrium predictions even in the case of the non-anonymous environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This, however, is unfortunately not the case when we relax the message space to allow vague messages: the richer message space induces multiplicity in agents’ reporting strategies and the audience’s off-path beliefs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We illustrate this point with the following examples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Example 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Consider the case where γ > N−1 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Then this case in which all agents report the vaguest messages {1,2,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N}, such that ρ(J) = 1 only when J = {1,2,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N} and 0 otherwise, is an equilibrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This example describes the scenario in which agents care greatly about social identity and are forced to stick to the (exogeneously) given norm of full vagueness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' While this specific example is similar to situations that might occur in the real world,4 an important implication is that there 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For instance, we may imagine a group of politicians all replying with the same vague message to a politically sensitive question because they know that any message other than a fully vague message will be interpreted as a lie and hurt their social identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Many similar examples can arise in situations where agents value their social identity highly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We appreciate Brian Rogers for suggesting this interpretation of the equilibrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 11 can exist a multiplicity of equilibria depending on the combination of γ and the off-path beliefs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Consider the following example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Example 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' All agents report a message in the form of an interval: [i, N] = {i, i+1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N−1, N}, and the audience assigns a positive posterior belief only to those interval-messages and zero to all other messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We need the following conditions to hold to constitute such equilibria: truth-tellers: i+N 2 +γρ(J) > U(i,{i, x∗, x∗+1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N}, t)5 for all i and an interval message J liars: k+N 2 − 1(i ̸= k)(t + ¯π(J)) + γρ(J) > U(i, J′, t) for any non-interval message J′ The conditions hold trivially when γ → ∞, and the signaling game becomes isomorphic with the NA-R scenario, where each message i is simply replaced with the interval [i, N].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The ending sequence conveys no information to the audience, and the equilibrium follows that of the NA-R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The multiplicity of equilibria obscures our understanding of the agents’ behavior regarding vague communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We thus first turn our attention to anonymous environments to delineate the relationship between one’s internal cost of lying and vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Because we are now free of the social identity component, we can utilize the fact there is a clear mapping between an agent’s type (i, t) and the resulting behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, fixing the observation i, one’s lying behavior simplifies to a monotone function of t in the A-R environ- ment: the agent reports j > i when (j −i)+c(i, j) > t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thus, there exists a threshold t∗ i for each observation i below which the agent reports a lie and above which the agent tells the truth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This threshold, however, becomes much higher in the A-UR environment because now the agent is equipped with vague messages that allow them to remain truthful while increasing the expected payoff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This fact leads to the following observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If a type (i, t) agent reports truthfully after observing state i in the A-R environment, then the agent also reports truthfully when observing i in the A-UR environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For each observation i < N, there exists a positive mass of agents who lie in the A-R environment but report truthfully in the A-UR environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This message maximizes the expected payoff conditional on including the true state i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We later define this kind of message as an optimally vague message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 12 The immediate corollary of these two observations is the following proposition, which states that the mass of liars is greater when communication is restricted in the anonymous environ- ments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proposition 2 (A-R/A-UR).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The set of types (i,t) of agents who lie in equilibrium in the A-UR environment is a subset of liar types in any equilibrium in the A-R environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The expected monetary earnings are greater on average when the communication is not restricted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Another implication of the anonymous environment worth noting is that the absence of the social identity concern simplifies the problem into a straightforward comparison between the solutions to truth-telling-constrained and unconstrained optimizations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, for each type (i, t) agent, there exists both maxJ:i∈J U(i, J, t) and maxJ∈MΩ U(i, J, t), and the agent reports truthfully only if the two maximums coincide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We can add structure to the constrained optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Intuitively, this maximum has the form of a union of the true observation i and some ending sequence x, x + 1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We can solve the optimization problem to find the threshold x∗ = ⌈(N + 2) − � 2N − 2i + 3⌉.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 6 Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The optimal vague message (OVM) for each true observation i is defined as OV Mi ≡ {i, x∗, x∗ + 1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N}, (8) where x∗ the threshold above which including the values maximizes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Then OV Mi ∈ argmaxJ:i∈J U(i, J, t) and weakly dominates all truthful messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This allows us to make the following inferences in the A-UR environment: Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the A-UR environment, all truth-tellers use the optimal vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the A-UR environment, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' no agent’s message contains a number below the true observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' ii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' no precise message except {N} is truthful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We can simplify the optimization problem by approximating it with a continuous uniform distribu- tion: argmaxx � i+1 i u N−x+2 du + � N x u N−x+2 du 13 Lastly, we conclude this section by comparing the behavior of an agent in the non-anonymous environment with that of an agent in the anonymous environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A thought experiment that involves choosing an agent and comparing their behavior in the two environments easily leads to the conjecture that the absence of the social identity concern should only encourage more lies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The following lemma shows that this is indeed the case under restricted communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Lemma 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If a type (i, t) agent lies in an equilibrium in the NA-R environment, then the agent lies in the A-R environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' See Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Using the above lemma, we now generalize the argument to compare the probability that an arbitrarily chosen agent lies in the two environments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proposition 4 (NA-R/A-R).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The set of types (i, t) of agents who lie in an equilibrium in the NA-R environment is a subset of the set of liar types in any equilibrium in the A-R environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The expected monetary earnings are higher on average in the A-R environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' See Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 4 Hypotheses In this section, we have transformed our theoretical analysis into specific hypotheses that can be tested in an experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The results of the experiment, which will be discussed in the following section, will be based on these hypotheses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In both the NA-UR and A-UR environments, agents use vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 1 is a consequence of Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 2 (A-R/A-UR).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous environment, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' more agents lie when communication is restricted (precise): lieA−R ≥ lieA−UR;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' ii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' an agent who is truthful in A-R is also truthful in A-UR conditional on the same observation;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' iii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' some agents who lie in A-R report truthfully in A-UR conditional on the same observation;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 14 iv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' agents earn higher monetary payoffs on average when communication is not restricted (vague): earningA−R ≤ earningA−UR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 2 is a consequence of Proposition 2 and the above observations about behavior in the A-UR environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the A-UR environment, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' all truth-tellers in A-UR use OVM;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' ii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' no message contains a number below the true observation;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' iii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' no precise message except {N} is truthful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 3 is a consequence of Proposition 3 and its corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 4 (NA-R/A-R).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Under restricted communication, agents earn higher monetary payoffs on average in the anonymous environment: earningNA−R ≤ earningA−R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hypothesis 4 is a consequence of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 5 Experimental Design We vary the experimental treatments along two dimensions: 1) we consider precise or vague messages, and 2) we vary the experimenter’s ability to identify responses to an individual sub- ject.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We represent variation in the anonymity of subjects in two types of experiments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Within each session, a subject confronts two stages of reporting tasks that represent the availability of vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In each stage, subjects are incentivized to observe a random number and later report the number to the experimenter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Their earnings depend on the number or numbers they report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous session, responses are recorded under screen names so that the exper- imenter cannot map a subject’s identity to their response.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As the subjects are instructed that the experiment prohibits such mapping by design, this treatment should establish the effect of suppressing the social identity concern and emulate the environment where γ → 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the non- anonymous (identifiable) session, on the other hand, the experimenter knows each subject’s response.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 15 The ‘stage’ is our basic unit of observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In each stage, subjects first observe the realiza- tion of a random integer uniformly distributed between 1 and 10 and later are asked to report the number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous treatment, the number is generated within the experimental software so that the experimenter knows both the true observation what each subject reports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the non-anonymous treatment, subjects use an external website to generate the random num- ber so that the experimenter cannot know the true observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='7 The experimenter is unable to identify whether a particular individual has lied or not in any of the treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This design choice derives from the idea that an environment in which the experimenter can detect cheating at the individual level is overly artificial;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' as result, subjects’ discomfort with lying behavior may be exaggerated in such an environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The observation process for the two tasks is identical and independent but the set of available messages differs between the two tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the restricted stage, the set is restricted to single- valued messages only.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the unrestricted stage, subjects are allowed to use both single-valued and set-valued messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the restricted stage, the subjects are paid the equivalent in dollars of the number they report divided by 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the unrestricted stage, when they report a single number, they are paid the equivalent in dollars to the number they report divided by 2, while when they report multiple numbers, the computer will randomly choose one of the numbers and they are paid the equivalent in dollars of this number divided by 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' To emulate the one-shot game structure of our model, we asked subjects to participate only once in each of the stages and we randomized the order of precise and vague stages within a session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='8 The combination of these between-subject and within-subject variations yields four treat- ments in our experiment: NA-R,9 NA-UR, A-R, and A-UR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We recruited 176 student subjects from the subject pool of the Missouri Social Science Ex- perimental Laboratory (MISSEL) at Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Louis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For recruitment and experiment management, we used the ORSEE system (Greiner 2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Students were invited to participate in a virtual Zoom meeting where the experimenter read the instructions and pro- vided a link to the main experiment web page.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The main experiment was implemented using the Qualtrics online survey platform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We conducted 18 sessions and each session lasted for approximately 30 minutes, including 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We provided a link to a Google search result for the phrase “random number between 1 and 10.” 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Using OLS regressions, we find that there is no order effect on the average report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The NA-R treatment is equivalent to the original experiment conducted by Fischbacher and Föllmi- Heusi (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 16 the instructions read during Zoom meetings, a screening quiz to make sure that subjects under- stood the experimental procedures, and two stages of the main experiment task via Qualtrics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 10 In all cases, subjects received a $2 show-up fee, so the total amount they could earn ranged from $2 to $12, with an average total fee, including the show-up fee, of $9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='84.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' No subject participated in more than one session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' See Appendix D for our experimental instructions and procedural details.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 6 Results We present basic summary statistics in Table 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' There were 8 anonymous sessions and 10 non-anonymous sessions, with an average of 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3 and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2 participants, respectively, in each session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the unrestricted communication treatments, we used the mean of the numbers included in each subject’s report when calculating the average report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We computed the average length of vague messages using the average number of numbers included in the messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Restricted Unrestricted Observations Average Average Vague (%) Length Anonymous 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='619 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='132 48% 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='103 97 Non-anonymous 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='000 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='282 75% 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='810 79 Table 2: Data summary We report first that the majority of subjects used vague messages in both treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Result 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the NA-UR environment, 59 of 79 participants (74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='7%) used vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the A-UR environment, 47 of 97 participants (48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='4%) used vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This result supports Hypothesis 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous treatment, the numbers reported for the restricted and unrestricted treat- ments average 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='619 and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='132, respectively, while in the non-anonymous treatment, the num- bers reported average 7 and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='282, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' These reported numbers are significantly lower 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The first 11 sessions were conducted from August 2020 to October 2020, while the remainder were conducted in September 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The results of a logit regression show no significant differences in the probability of lying between the two groups of sessions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Participants chose time blocks voluntarily, and we provided the anonymous version of the software in larger sessions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 17 than those in the profit-maximizing reports and are in line with findings reported in the liter- ature that have highlighted agents’ preference for truth-telling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Overall, we find that allowing vague messages increased the mean of numbers reported.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We also find that the probability of lying is independent of gender in both the restricted and unrestricted treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Additional details regarding the logit regression outcomes are available in Appendix B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, we find that the impact of anonymity on average reports is statistically not significant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Finally, we show that the introduction of non-anonymity significantly affects the choice to send vague mes- sages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the treatment where social identity concern is relevant, subjects use vague and longer messages more often.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Result 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous environment, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' more participants lied when communication was restricted (43.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3% in A-R and 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='9% in A-UR);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' almost all participants who were truthful in A-R remained truthful in A-UR;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='12;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 52.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='58 % of subjects reported truthfully under both A-R and A-UR, 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='80% always lied, while 16% switched from lying to truth-telling when they were allowed to be vague;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' participants reported higher numbers on average when communication was not restricted (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='619 in A-R and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='282 in A-UR).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A comparison between the results obtained with restricted and unrestricted communication in anonymous sessions confirms Hypothesis 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' More subjects lied in the restricted than in the unrestricted treatment and their behavior can be classified into three types: truth-tellers (re- ported truthfully under both treatments), conditional liars (switched from lying to truth-telling when they were allowed to be vague) and liars (lied under both treatments).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is in line with previous work on lying aversion (Gneezy, Kajackaite, and Sobel (2018), Khalmetski and Sliwka (2019), among others).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' An interesting point is that 15 of the 16 participants who switched from lying to truth-telling reported picking the number 10 in A-R, yet they chose vague and truthful messages when the expected earnings were lower than 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This finding supports our conjecture that the internal cost of lying depends on the inclusion of true states in the reported messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Four subjects were truthful in the A-R and lied in the A-UR, but their true observations were much larger in the A-UR than in the A-R treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' It is likely that the counterfactual would be consistent with our prediction conditional on observing similar numbers in both treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 18 Table 3 presents the average numbers reported in the A-R and A-UR treatments for truth- tellers, conditional liars, and liars.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We find that truth-tellers report on average 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='282 higher numbers when they are allowed to be vague.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This result is significant at the 1% significance level and is consistent with the model’s prediction that truth-tellers seek payoff maximization conditional on including true observations in their reports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Conditional liars reported on av- erage 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='156 lower numbers when they were allowed to be vague.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This result is significant at the 1% significance level and supports Hypothesis 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We interpret this result as implying that subjects with moderate t prefer to use vague messages and include true states to reduce the internal cost of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Consistent with our expectation regarding agents with higher t, we did not find a significant impact of unrestricted communication on liars as reflected in their aver- age reports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, t-tests of the difference between subjects’ true observations and their reports show over-reporting across all three types in the A-UR treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This result is also in line with Hypothesis 2, which states that agents earn higher monetary pay-offs under un- restricted communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Likewise, we also find that a true observation has a negative effect on the probability of lying in the restricted treatment, while it has no effect in the unrestricted one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Table 4 presents the estimated coefficients of the logit model of the effect of true obser- vations on the probability of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This result is intuitive, as a subject who randomly drew a low observation can increase their payoff only by lying when the communication is restricted to precise messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' These subjects can, however, always employ a vague yet truthful message in the unrestricted environment;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' hence the true observation has no effect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 19 Restricted Unrestricted Differences in means Observations Truth-tellers 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='608 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='890 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='282 51 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='416) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='293) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='464) Conditional liars 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='875 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='718 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='156 16 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='125) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='316) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='370) Liars 10 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='962 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='038 26 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='027) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='027) Test (Report = True) Truth-tellers 0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='953 51 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='267) Conditional liars 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='000 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='531 16 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='791) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='432) Liars 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='462 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='076 26 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='521) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='543) Table 3: Anonymous Treatment Observation (Restricted) Observation (Unrestricted) Lying dummy 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='138 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='081 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='072) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='081) Table 4: Logit model of the probability of lying Result 3-1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the A-UR environment, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (a) 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='4% (19 of 67) of truth-tellers reported the optimal messages (either the OVM or the honest 10), 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='9% (18 of 67) reported a pair of true observations and 10;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (b) 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='9% (16 of 67) of truth-tellers used a precise message below {10};' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' only 2 of 97 subjects included a number below the true observation in the report;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' all precise messages reporting numbers below {10} were truthful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This result partially supports Hypothesis 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Figure 1 summarizes the types of messages used in A-UR treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Overall, 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='5% (50 of 97) of participants used precise messages, while 48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='5% (47 of 97) used vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Among the 50 precise messages, 54% (27 of 50) were 20 Figure 1: Message types used in A-UR treatments lies and 46% (23 of 50) were not lies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' All the liars reported drawing the maximum of 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Among the 23 precise truth-tellers, 7 observed 10 and reported so.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The model predicts that all truth-tellers seek payoff maximization conditional on including true observations in their reports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If we combine both optimal vague messages (including honest 10 as the optimal message) and the pair-type messages into a broader set of payoff-increasing truthful messages, we find that the majority (55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2%) of truth-tellers maximized their monetary payoffs conditional on being honest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='13 Yet there remains a noticeable number of precise truth- tellers who reported drawing a number below 10, which contradicts our hypothesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This may suggest the possibility of other motivation for truth-telling that is not captured by our model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Comparing behaviors across the A-UR and the NA-UR treatments, it is interesting to note that a higher percentage of participants used vague messages in the non-anonymous environ- ment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Overall, 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3% (20 of 79) of participants used precise messages while 74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='7% (59 of 79) used vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This difference might reflect the fact that some maximal liars in the 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' When a message is both optimal and precise, namely {10}, we count it as an optimal message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Likewise, when a message is both optimal and contains two numbers, like {8,10}, we label it ‘Optimal.’ We classify sub-optimal messages using two numbers as ‘Pair.’ 21 75 Percent (%) Category 50 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='84 lie not lie 25 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='71 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='03 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='06 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='59 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='43 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='34 0 Precise Optimal (vague) Pair Others Form of reportFigure 2: Message types used in NA-UR treatments anonymous case lie by reporting 10 in the absence of any concern with social identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Also, it is worth noting that vague messages were longer in the non-anonymous treatments (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='81 num- bers reported on average) compared with what occurred in the anonymous environment (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='10 numbers reported on average).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The difference is significant at the 1% significance level and could reflect the impact of vague messages on the external cost of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' To further compare the effect of social identity on vague communication, we categorized vague messages into the following three categories: ‘Pseudo-optimal,’ ‘Pair,’ and ‘Others.’ While we do not know the true observation of each subject in the non-anonymous treatments, we classified optimal-looking messages as ‘Pseudo-optimal’ in a similar manner to how we defined the OVM in Definition 8, using the minimum reported numbers in the messages as pseudo-true-observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For exam- ple, a report consisting of {6,9,10} is considered pseudo-optimal because this message would maximize the expected payoff if the true observation was 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A report {6,7,8,9,10}, on the other hand, is not classified as pseudo-optimal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='14 Figure 2 summarizes the types of messages used in 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' When a message is both pseudo-optimal and pair, like {8,10}, we count it as ‘Pseudo-optimal,’ similar to how we classified vague messages in the A-UR treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Only sub-optimal messages using two numbers are labeled ‘Pair.’ 22 75 Percent (%) Category 50 Interval 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='85 Non-interval 25 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='32 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='25 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='86 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='13 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='59 0 Precise Pseudo-optimal (vague) Pair Others Form of reportNA-UR treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Result 3-2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the NA-UR environment, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0% (13 of 59) and 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='6% (8 of 59) of vague messages are pseudo-optimal and pairs, respectively;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' and 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2% (29 of 59) of vague messages take the form of an interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Note that, compared with the results observed in the anonymous treatment, as displayed in Figure 1, participants used fewer precise and pair-type messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Conditional on using vague messages, only 36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='6% are either pseudo-optimal or pairs in the NA-UR treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This ratio differs starkly from the 53.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2% observed in the A-UR treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, we see that the vast majority of the messages now fall into the ‘Others’ category.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Among the ‘Others’ messages, half take the form of intervals, which was not the case in the anonymous treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In particular, 49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1% of vague messages reported in the NA-UR treatment used intervals, while only 37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2% did in A-UR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We can infer from the forms of the reported messages that subjects understood that pseudo-optimal and pair messages are obvious and can negatively affect the audience’s belief in an equilibrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The distribution of the lengths of messages also indicates that there is a significant differ- ence in reporting patterns between the A-UR and the NA-UR treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In Figure 3 we report the cumulative distribution of message lengths for the A-UR and NA-UR treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The fig- ure shows that subjects use vague and longer messages more often in the NA-UR treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Together with the frequent use of interval messages in the non-anonymous treatment, we inter- pret this pattern as indirect evidence of the impact of vagueness on the external cost of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Result 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Comparison between the A and NA environments 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The average numbers reported in A-R and NA-R are 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='619 and 7, respectively;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' the average numbers reported in A-UR and NA-UR are 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='132 and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='232, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We do not find statistically significant differences between average numbers reported in the anonymous and non-anonymous treatments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Our results do not reject the null hypothe- ses under either restricted or unrestricted communication, with p-values 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='185 and 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='562, re- spectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' While these results run against our Hypothesis 4 prediction, they are in line with 23 Figure 3: The cumulative distribution of lengths of vague messages in the A-UR and NA-UR treatments Fries et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (2021), who find a similarly perplexing outcome according to which increasing the anonymity of a report has no effect on average numbers reported or lying behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' One possi- ble explanation is that the impact of anonymity on behavior creates only a minuscule difference when collapsed into an average and projected into a unidimensional space, and our sample is far from the infinitely large sample needed to provide such statistical power.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The difference we documented across the restricted and unrestricted treatments makes this explanation particu- larly plausible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Despite the similar magnitudes of the average numbers reported, the types of messages used in the two environments differ substantially in form and length.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 24 100 8 4 1 2 3 4 5 6 7 8 9 10 length of messages Non-Anonymous Anonymous7 Concluding Remarks We contribute to the literature by bridging lying behavior with studies that involve vague com- munication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The extensive use of vague messages among our subjects, in both anonymous and non-anonymous sessions, provides us with insights that can be applied to various communica- tion contexts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We find that subjects exploit vagueness to be consistent with the truth, while at the same time leveraging such imprecision to their own benefit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, we find that 38% of liars in the anonymous sessions switch to truth-telling when they are allowed to be vague, which supports the key conjecture that a vague and truthful message reduces the internal cost of lying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We show that subjects use vague and longer messages more often when they are con- cerned with social identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The difference in the reporting patterns across the NA-UR and the A-UR sessions suggests that vagueness still plays an important behavioral role.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Overall, subjects included higher numbers in their messages more often when vague com- munication was allowed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This finding sheds new light on our understanding of lying aversion, suggesting that a restricted message space could be a source of the observed abstention from monetary-payoff maximization in previous experiments reported in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The aver- age numbers reported in the treatments with restricted communication in our experiment are comparable to results reported in the previous experiments, capturing approximately 70% of the maximum numbers available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Insofar, as we allow the use of vague messages, however, the average number reported increases to 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1, reducing the degree of lying aversion from 30% of the maximum monetary payoff amount to less than 20%.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, our findings in the anonymous treatments indicate that our subjects lie less often as they increase their expected monetary payoffs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, the option to use vague messages enables the subjects to maintain their integrity by including true states in their messages at a lower monetary cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Similarly, sub- jects in the non-anonymous treatments also reported higher numbers on average when vague messages were allowed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Yet, the difference in the reporting patterns —vague messages with longer and more sophisticated forms —suggests that the use of vague messages provides sub- jects an alternative means to retain their social identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In other words, when subjects are given a new option to use vague messages to convey their honesty to an audience, they no longer need to sacrifice their monetary payoffs significantly to provide a credible signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Our finding is analogous to the “moral wiggle room” mentioned in Dana et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2007, where dictators care only whether they maintain an image of fairness, but without having a signifi- 25 cant effect on their partners’ payoffs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In our experiment, most subjects exhibit no hesitation in increasing their monetary payoffs as long as their messages can remain even remotely truth- ful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The presence of precise truth-tellers in the anonymous environment, while small, suggests however that the moral-wiggle-room argument alone may not paint the whole picture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The de- composition of the aversion to monetary-payoff maximization in our experimental design calls for a new perspective on misleading behavior by showing that the observed aversion in many individuals is independent of the consequences of their message choices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' It is possible that another motivation for truth-telling, such as a concern with projecting a well-intentioned self- image.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, although subjects might have understood that there was no external observer to judge their behavior, their moral standards may combine the honesty achieved by reporting true observations with the uprightness of reporting the most accurate messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 26 References Abeler, Johannes, Daniele Nosenzo, and Collin Raymond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Preferences for Truth- Telling.” Econometrica 87 (4): 1115–1153.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Agranov, Marina, and Andrew Schotter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Ignorance is bliss: an experimental study of the use of ambiguity and vagueness in the coordination games with asymmetric payoffs.” American Economic Journal: Microeconomics 4 (2): 77–103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Battigalli, Pierpaolo, and Martin Dufwenberg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Dynamic psychological games.” Journal of Economic Theory 144 (1): 1–35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Blume, Andreas, Douglas V DeJong, Yong-Gwan Kim, and Geoffrey B Sprinkle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Experimental Evidence on the Evolution of Meaning of Messages in Sender- Receiver Games.” The American Economic Review 88 (5): 1323–1340.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Crawford, Vincent P, and Joel Sobel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1982.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Strategic Information Transmission.” Econometrica 50 (6): 1431–1451.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Dana, Jason, Roberto A Weber Jason Xi Kuang, Carnegie Mel-lon, J Dana, R A Weber, and J X Kuang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Exploiting moral wiggle room: experiments demonstrating an illusory preference for fairness.” Economic Theory 33:67–80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Deversi, Marvin, Alessandro Ispano, and Peter Schwardmann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Spin doctors: An experiment on vague disclosure.” European Economic Review 139 (October): 103872.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Dickhaut, John W, Kevin A McCabe, and Arijit Mukherji.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1995.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “An experimental study of strategic information transmission.” Economic Theory 6 (3): 389–403.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Fischbacher, Urs, and Franziska Föllmi-Heusi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Lies in Disguise - An Experimental Study on Cheating.” Journal of the European Economic Association 11 (3): 525– 547.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Fries, Tilman, Uri Gneezy, Agne Kajackaite, and Daniel Parra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Observability and lying.” Journal of Economic Behavior & Organization 189:132–149.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 27 Geanakoplos, John, David Pearce, and Ennio Stacchetti.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1989.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Psychological games and sequential rationality.” Games and Economic Behavior 1 (1): 60–79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Gneezy, Uri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Deception: The Role of Consequences.” 95 (1): 384–394.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Gneezy, Uri, Agne Kajackaite, and Joel Sobel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Lying Aversion and the Size of the Lie.” American Economic Review 108 (2): 419–453.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Greiner, Ben.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Subject pool recruitment procedures: organizing experiments with ORSEE.” Journal of the Economic Science Association 1 (1): 114–125.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Hagenbach, Jeanne, and Eduardo Perez-Richet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Communication with evidence in the lab.” Games and Economic Behavior 112 (November): 139–165.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Khalmetski, Kiryl, Bettina Rockenbach, and Peter Werner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Evasive lying in strate- gic communication.” Journal of Public Economics 156 (December): 59–72.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Khalmetski, Kiryl, and Dirk Sliwka.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Disguising Lies—Image Concerns and Partial Lying in Cheating Games.” American Economic Journal: Microeconomics 11 (4): 79–110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Russell, Bertrand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1923.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Vagueness.” Australasian Journal of Psychology and Philosophy 1 (2): 84–92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Sánchez-Pagés, Santiago, and Marc Vorsatz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “An experimental study of truth- telling in a sender–receiver game.” Games and Economic Behavior 61 (1): 86–112.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Schmeidler, David.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 1973.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Equilibrium points of nonatomic games.” Journal of Statistical Physics 7 (4): 295–300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Serra-Garcia, Marta.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Lying in Economics.” Chap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 38 in The Oxford Handbook of Lying, edited by Jörg Meibauer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Oxford University Press.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Serra-Garcia, Marta, Eric van Damme, and Jan Potters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Hiding an inconvenient truth: Lies and vagueness.” Games and Economic Behavior 73 (1): 244–261.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 28 Shalvi, Shaul, Jason Dana, Michel JJ Handgraaf, and Carsten KW De Dreu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Justi- fied ethicality: Observing desired counterfactuals modifies ethical perceptions and behavior.” Organizational Behavior and Human Decision Processes 115 (2): 181– 190.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Sobel, Joel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Lying and Deception in Games.” Journal of Political Economy 128 (3): 907–947.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Tergiman, Chloe, and Marie Claire Villeval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “The Way People Lie in Markets: Detectable Vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Deniable Lies.” Management Science (Forthcoming).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Turmunkh, Uyanga, Martijn J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' van den Assem, and Dennie van Dolder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Mal- leable Lies: Communication and Cooperation in a High Stakes TV Game Show.” Management Science 65 (10): 4795–4812.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Wood, Daniel H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Communication-Enhancing Vagueness.” Games 13 (4): 49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Zhang, Sookie Xue, and Ralph-Christopher Bayer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' “Delegation based on cheap talk.” Theory and Decision, 1–29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 29 Appendices A Proofs Let us first begin with restating the previously known results about the NA-R environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The main intuition is that some agents lie when they see a small number, while some others always report the truthful message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thus, when a reported number is small, it is safe to believe that the agent is being honest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The following equilibrium results establish a baseline for comparison with the other environments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Lemma 2 (KS LM 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In any equilibrium in the NA-R environment, there exists a strict positive probability that agents lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Suppose there exists an equilibrium in the NA-R environment where no agent lies and the audience believes P(j is truthful) = 1 for any report j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The utility of an arbitrary agent with the true observation of i < N and the type t is U(i, j, t) = j − 1(i ̸= j)[t + c(i, j)] + γ · 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' when the agent reports some j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As the social identity is constant under the audience’s belief that everyone is truth-telling, the agent is better off by reporting i + 1 instead of i when U(i, i + 1, t) − U(i, i, t) = ((i + 1) − i) − (t + c(i, i + 1)) > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Because we assumed c(i, i + 1) < 1, there exists some 0 < t < 1 − c(i, i + 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Lemma 3 (KS PR 4;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' GKS PR 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In any equilibrium in the NA-R environment, no agent underre- ports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Suppose there exists an equilibrium where an agent lies by reporting a number j below their true observation i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Then it must be the case ρ(j) > ρ(i).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 30 It also follows that reporting j must yield a larger utility than reporting i: (j − i) + γ(ρ(j) − ρ(i)) ≥ t + c(i, j).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We complete the proof by showing that ρ(i) = 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' that is, if any agent choose to report i, then it must be the case that the report is truthful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Take another agent with the intrinsic aversion type t who observes some i′ ̸= i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We claim that U(i′, j, t′) > U(i′, i, t): (j − i) + [c(i′, i) − c(i′, j)] + γ(ρ(j) − ρ(i)) ≥ c(i′, i) − c(i′, j) + t + c(i, j) > 0 because of the triangular inequality assumption: c(i, i′) + c(i′, j) ≥ c(i, j).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As the choice of this agent is arbitrary, this is the case for all agents who do not observe i never lies by reporting i;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' in turn, this implies ρ(i) = 1, a contradiction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Lemma 4 (KS THM 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' GKS PR 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In any equilibrium in the NA-R environment, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' there exists a threshold 1 < l∗ < N such that ∀j ≥ l∗ ∃i ̸= j, t ∈ � s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' σj it > 0 and ∀j < l∗, i ∈ Ω, t ∈ � σj it = 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' ii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' all agents who observe a value above the threshold report their observed value truthfully.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let LΩ P ⊆ MΩ P be the set of messages that liars use to lie with positive probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let l∗ = min LΩ P .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' By Lemma 2, LΩ P is nonempty and l∗ is well-defined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Also, by the no-underreporting condition, we can deduce that ρ(1) = 1 and 1 < l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We now show LΩ P = {l∗, l∗ + 1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=', N} by contradiction: suppose there exists some elements of Ω greater than l∗ which is not an element of LΩ P .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let n be the minimum of such elements, so that n − 1 ∈ LΩ P .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As ρ(n) = 1, we can see that any agent who lies by reporting n − 1 is strictly better off by reporting n instead: U(i, n, t) − U(i, n − 1, t) = (n − (n − 1)) − (c(i, n) − c(i, n − 1)) + γ(ρ(n) − ρ(n − 1)) > c(n − 1, n) − (c(i, n) − c(i, n − 1)) + γ(ρ(n) − ρ(n − 1)) ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 31 Now it remains to show that the agents who observe i ∈ LΩ P reports truthfully.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Suppose there exists j, j′ ∈ LΩ P such that some agent observing j instead chooses to report j′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, U(j, j′, t) − U(j, j, t) = (j′ − j) − (t + c(j, j′)) + γ(ρ(j′) − ρ(j)) ≥ 0 for any intrinsic aversion type t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This implies that any agent who lies by reporting j is strictly better off by reporting j′ instead: U(i, j′, t) − U(i, j, t) ≥ (j′ − j) + (c(i, j) − c(i, j′)) + γ(ρ(j′) − ρ(j)) ≥ t + c(j, j′) + (c(i, j) − c(i, j′)) > 0 for any intrinsic aversion type t and any true observation i ̸= j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, all agents who observe i ∈ LΩ P reports truthfully.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Based on the three results, we bring a simple comparison of the behavior of an agent in the NA-R environment to that in the A-R environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A thought experiment of choosing an agent and comparing their behavior in the two environments easily leads to a conjecture that the ab- sence of the social identity concern should only facilitate more lies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Applying these observation to the unrestricted communication leads to Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We also obtain Lemma 1 from these three lemmas, which we will use to prove Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof of Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Case 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Anonymous environment: as shown in Proposition 3, all truth-tellers use the optimal vague message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Case 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Non-anonymous environment: suppose there exists an equilibrium where no agent uses a vague message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let ρ(j) = 0 for all j ∈ MΩ V .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Given that all messages used with positive probability are precise, we know all precise messages are used with positive probability in this equilibrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We use Lemma 2 and 3 to argue that there exists a positive probability that agents lie upward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let l∗ be the threshold defined in Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let us first argue that there exists agents who observe l∗ − 1 and lie by reporting l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 32 Suppose not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Then there must exist some l > l∗ such that U(l∗ − 1, l, t) − U(l∗ − 1, l∗, t) = (l − l∗) + (c(l∗ − 1, l∗) − c(l∗ − 1, l)) + γ(ρ(l) − ρ(l∗)) > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Note that c(l∗ −1, l∗)− c(l∗ −1, l) ≤ 0 because l > l∗ and c is increasing in the distance between the two arguments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Also, c(l∗ −1, l∗)− c(l∗ −1, l) ≤ c(i, l∗)− c(i, l) ≤ 0 for all i < l∗ because of the triangular inequality assumption.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This implies that for all agents whose true observation is below l∗ is better off by reporting l instead of l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As no agent would lie by reporting l∗, this is a contradiction to the definition of threshold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Now consider the agent who observes l∗−1 and lies by reporting l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The agent receives the utility of U(l∗ − 1, l∗, t) = l∗ − (t + c(l∗ − 1, l∗)) + γρ(l∗).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' However, if the agent reports {l∗ − 1, l∗ + 1} instead, the agent receives U(l∗ − 1,{l∗ − 1, l∗ + 1}, t) = l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' That is, the agent is better off by reporting {l∗ − 1, l∗ + 1} when t > γρ(l∗) − c(l∗ − 1, l∗), which happens with a positive probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Furthermore, the above analysis is valid for all ρ({l∗ − 1, l∗ + 1}) ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a contradiction to the assumption of an equilibrium with no vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, there exists a positive probability that agents use a vague message in any equilibrium in the NA-UR environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof of Lemma 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let j > i be a part of an equilibrium strategy for an (i, t)-agent in the NA-R environment: U(i, j, t) = j − [t + c(i, j)] + γρ(j) ≥ U(i, j′, t) ∀j′ ∈ MΩ P .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We can infer that j − [t + c(i, j)] + γρ(j) ≥ i + γρ(i), 33 or (j − i) − [t + c(i, j)] ≥ γ(ρ(i) − ρ(j)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As no agent underreports in the A-R environment, it suffices to show that this agent does not tells the truth in the A-R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Suppose not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' UA(i, i, t) = i > UA(i, j, t) = j − [t + c(i, j)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This implies ρ(i) < ρ(j), meaning the agent lied in the NA-R both because there were a monetary gain and a social identity gain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' (j − i) + [c(i′, i) − c(i′, j)] + γ(ρ(j) − ρ(i)) ≥ c(i′, i) − c(i′, j) + t + c(i, j) > 0 because of the triangular inequality assumption: c(i, i′) + c(i′, j) ≥ c(i, j).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As the choice of this agent is arbitrary, this is the case for all agents who do not observe i never lies by reporting i;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' in turn, this implies ρ(i) = 1, a contradiction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, the agent must lie in the A-R if the agent sees the same observation i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Proof of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' WTS P(lieNA−R) < P(lieA−R).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We argue by the law of total probability: P(lieNA−R) = N � i=1 P(observe i)P(lieNA−R|observe i);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' P(lieA−R) = N � i=1 P(observe i)P(lieA−R|observe i).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Because the probability of observing some i is uniform in both environments, it suffices to show that the conditional probability of lying in A-R is greater than or equal to that in NA-R for all true observation i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We learned that there exists some threshold 1 < l∗ < N in Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thus, conditional on that an agent observing i ≥ l∗, we have the conditional probability of the agent reporting a lie as P(lieNA−R|observe i) = 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 34 for all intrinsic aversion type t, while P(lieA−R|observe i) > 0 because of the positive probability that the agent has the type t small enough to report a lie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Now consider the case of i < l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Let Ti be a subset of � such that P(lieNA−R|observe i, t ∈ Ti) > 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' P(lieNA−R|observe i, t ̸∈ Ti) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' By Lemma 1, P(lieA−R|observe i, t ∈ Ti) = 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' and P(lieA−R|observe i, t ̸∈ Ti) ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thus, regardless of P(t ∈ Ti), we have P(lieNA−R|observe i) ≤ P(lieA−R|observe i) for all i < l∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, P(lieNA−R) < P(lieA−R).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Also, because the monetary payment is a monotone mapping of the reports under the re- stricted communication, and because any lying takes the form of reporting upward, agents would earn more monetary payoff on average as the probability of lying is greater in the A- R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' B Gender Difference B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1 Non-anonymous Environment In the non-anonymous environment, 28 males and 51 females attended the experiments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We do not find statistically significant differences between males and females neither on their reports nor on the length of vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Under the unrestricted communication, male and female subjects are almost equally likely to send a precise message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Men are more likely to send a pseudo-optimal vague message, while women are slightly more likely to send a message with contains a pair or an interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 35 Table 5: Gender difference on Reports in the Non-anonymous Environment Dependent variable: Report_P Exp_V vague_size (1) (2) (3) Gender difference 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0553 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1587 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2941 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='7291) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2963) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3582) Constant 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='9804 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3379 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='7059 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='4341) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1764) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2133) Observations 79 79 79 R2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0001 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0037 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0087 Figure 4: Message types male subjects used in NA-UR treatments 36 60 Percent (%) 40 Category 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='86 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='29 Interval 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='57 Not Interval 20 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='86 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='86 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='57 0 Precise Pseudo-optimal(vague) Pair Others Form of reportFigure 5: Message types female subjects used in NA-UR treatments 37 60 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='33 Percent (%) 40 Category Interval Not Interval 20 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='53 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='57 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='73 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='88 0 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='96 Precise Pseudo-optimal(vague) Pair Others Form of reportB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2 Anonymous Environment We used only a subset of anonymous sessions to analyze the gender effect as we could not identify the participants’ gender in principle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We ran four single sexed sessions (two male-only sessions and two female-only sessions) to create a gender dummy variable without interfering the anonymous property of the experiment design.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Under the restricted communication, we do not see statistically significant differences be- tween males and females neither on their reports nor on the length of vague messages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Table 6: Gender difference on Reports in the Anonymous Environment Dependent variable: Report_P Exp_V vague_size (1) (2) (3) Gender difference 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1034 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='6061 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='4867 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='7107) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='5810) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='4461) Constant 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='000 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3290 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3143 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='4784) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3911) (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='3003) Observations 64 64 64 R2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0003 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0172 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='0030 The results show that, 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='29 % of males and 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='03% of females lied in the non-anonymous treatment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Therefore, we do not find substantial difference in the lying behavior between them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addition, we do not see significant differences in the choice of type and length of vague messages, as shown by Figures 6 and 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 38 Figure 6: Message types male subjects used in A-UR treatments 39 60 (%) 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='03 40 Category Percent ( lie not lie 20 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='59 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='45 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='79 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='24 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='9 0 Precise Optimal(vague) Pair Others Form of reportFigure 7: Message types female subjects used in A-UR treatments 40 60 Category Percent !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='29 lie not lie 20 2 86 2 86 20 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='14 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='29 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='57 0 - Precise Optimal(vague) Pair Others Form of reportC Experimental procedure An experiment session contains the instructions, a preliminary quiz, and two stages of choice tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The subjects receive invitations through the Missouri Social Science Experimental Labo- ratory(MISSEL)’s ORSEE system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Once registered, they receive a link to a Zoom meeting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After subjects join the Zoom meeting room, the experimenter reads the instructions, followed by a preliminary quiz to ensure the understanding of the procedures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In an anonymous session, ev- eryone including the experimenter has their video off, and the experimenter does not explicitly take attendance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In order to assure the subjects about anonymity, subjects are asked to use a screen name.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experimenter emphasizes that the screen name is used solely for data analysis and that they cannot associate the screen name with their true identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In an non-anonymous session, on the other hand, the experimenter has their video on and asks the subjects to turn their video on as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Also, the experimenter takes attendance with the experiment roster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The subjects are allowed to participate in the main experiment only after successfully passing the preliminary quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Each subject has three chances to attempt the quiz before turned away.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experimenter provides a link to the main experiment website via the chat window of the Zoom meeting upon the successful completion of the quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the beginning of the main experiment part, subjects enter their identification information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous treatment, subjects are asked to enter their screen name;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' whereas, in the non- anonymous treatment, subjects are asked their name and student ID.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After entering the identification information, the subjects observe the realization of a ran- dom integer uniformly distributed between 1 to 10 and later asked to report the number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous treatment, the number is generated within the experiment software.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the non- anonymous treatment, subjects click on a link to a Google search result for the phrase “random number between 1 to 10.” In the reporting screen, subjects are presented with ten boxes labeled from 1 to 10 and report by selecting the numbered boxes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In a precise stage (Figure 8), which allows for precise messaging only, a subject may select only one box at a time, whereas in a vague stage (Figure 9), they may select multiple boxes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Subjects are told that this selection is deemed as a statement that the numbers represented by the selected box include the number they observed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the case of a precise message, the subject is paid the equivalent in dollars to the number she selected divided by two.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the case of a vague message, the computer randomly chooses 41 Figure 8: A screenshot of the experiment software displaying the precise-message stage one number from the submitted selection of numbers and pay the equivalent in dollars to the randomly chosen number divided by two.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After subjects complete both precise and vague stages, a confirmation screen reviews their compensation information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the anonymous treatment, the experimenter bulk-purchases Ama- zon gift card to match the earnings for each screen name.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experimenter provides a third party (Department of Economics) with a list of the gift codes and matching screen names, ex- cluding the amount associated with each code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The subjects visit an external website and re- trieve their compensation in the form of the gift code by entering their screen name.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This pro- cess is to ensure the anonymity of the experiment session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experiment can map between responses and screen names, but not between screen names and true identities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The third party can only map between true identities and screen names, but not between true identities and their corresponding responses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the non-anonymous treatment, they are asked to enter their email address and taxpayer information to receive the Amazon gift card directly emailed to them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 42 XM Preview - Online Survey Softwa X → 0 https:/wustl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='az1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='qualtrics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='com/fe/preview/SV_2ggP5vixBroObNOF?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Q_SurveyVersionID=current&Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='.:。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='☆ 三 Restart Survey PlaceBookmark Mobile view off Toolsv :Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Louis Stage 2 Whatnumberdidyouobserve?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Pleaseselectonenumber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2 3 4 5 6 9 10 PoweredbyQualtricsFigure 9: A screenshot of the experiment software displaying the vague-message stage 43 XM Preview - Online Survey Softwa X → 0 https:/wustl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='az1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='qualtrics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='com/fe/preview/SV_2ggP5vixBrObNOF?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Q_SurveyVersionID=current&Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='.。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='☆ 三 Restart Survey PlaceBookmark Mobile view off Tools : Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Louis Stage 1 Whatnumberdidyouobserve?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Youmayselectasetof numbersbyclickingmultiple numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='Afteryousubmityourselection,thecomputerwillrandomly chooseonefromthe setof numbersyouselected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 2 3 4 5 9 7 8 9 10 Powered by Qualtrics D Experimental Instructions D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='1 Instructions for Anonymous session (Zoom session;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' no video) Welcome to the experiment!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thank you very much for participating today, and I will first walk you through the instructions for the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We need your full attention during the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you have trouble with hearing the audio or seeing the shared screen, please let me know.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Do not turn the video on during the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you have any questions during the instructions, please raise hand so that I can unmute you.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Your question will be answered out loud, so everyone can hear.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experiment you will be participating in today is an experiment in individual decision making.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the end of the experiment, you will be paid an Amazon gift card.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will receive the show-up fee of $2 for completing the experiment, with the additional amount that depends on your decisions and on chance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The details of the compensation will be described later.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' All instructions and descriptions that you will be given in the experiment are accurate and true.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In accordance with the policy of this lab, at no point will we attempt to deceive you in any way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' I would like to first point out that we want to ensure this experiment is conducted anony- mously, meaning that we cannot connect the responses recorded in this experiment to any par- ticular individual who participated in this research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Qualtrics, the survey platform we are using, provides an option for the researchers to not collect any personal information, such as IP address or geographic location of the participants, for anonymous surveys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Also, your response will be recorded with a SCREEN NAME.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be asked to choose a screen name that is at least 8 characters in length using letters, numbers, and underscore.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This SCREEN NAME is only used in data analysis and distributing your Amazon gift card after the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We cannot and will not attempt to associate SCREEN NAMEs to any particular individual.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' I will now describe the main features of the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' First, there is a short quiz to ensure your understanding of the procedures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be able to repeat the quiz if you make mistakes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will have three chances to attempt the quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you fail to get all questions correct after three attempts, you may not participate in the main experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Even in such a case, please remain connected to the Zoom session until everyone finishes, and you will receive the show-up fee for today’s experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The main part of the experiment after the preliminary quiz consists of two STAGES.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In each STAGE you will observe a number that we ask you to remember and later report to us.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The number you report to us determines how much money you will be paid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 44 At the end of the experiment, a confirmation screen will summarize today’s experiment and provide the information to retrieve your payment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' I will give you more details about each step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the end of this instructions, we will first provide a link for the preliminary quiz using the chat window.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the webpage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please choose a screen name, and make sure you keep this screen name.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After you successfully complete the preliminary quiz, you will be reminded of your screen name once again.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We recommend you copy and paste the screen name.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We cannot recover this information for you, and you will not be able to receive your compensation without the correct screen name.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please wait while everyone else finishes the quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Once everyone finishes, we will provide another link for the main experiment using the chat window.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the first page of the main experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Make sure you use the same screen name you used in the preliminary quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You may not receive your compensation if the screen names do not match.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As previously mentioned, your main task today is to observe a number that we ask you to remember and later report to us.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The observation process is identical in both STAGES.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the beginning of both STAGES, the computer will randomly draw a number between 1 to 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The probabilities are equal across the numbers;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' that is, each number is chosen with the same probability of one-tenth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We ask you to remember the number and report on the next screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' However, the way you can report differs between STAGES.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In one STAGE, you are allowed to select one number after you observe the draw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the experiment stage where you can select one number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' By one number, we mean that you may click only one number on the screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We will interpret this selection as a statement that the number you observed is the number you selected, and we will regard this number as your report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The number you report determines how much money you will be paid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be paid the equivalent in dollars to the number you report divided by 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In other words, if you report “1”, you receive 50 cents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you report “2”, you receive $1, if you report “3”, you receive $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='50 and so on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A confirmation screen after your report will help you review your selection and the corresponding payment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In another STAGE, you are allowed to select a set of numbers after you observe the draw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the experiment stage where you can select a set of numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' By a set of numbers, we mean that you may click multiple numbers on the screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For instance, you may choose to click on four numbers, one number, two numbers, or even all ten numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you select multiple numbers, we will interpret it as a statement that the number you observed is one of the numbers you selected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you select a single number, we will interpret it as a statement that the number you observed is the number you selected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After you 45 submit your selection, the computer will randomly choose one number from the set of numbers you selected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We will regard this randomly chosen number as your report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Again, the number you report determines how much money you will be paid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be paid the equivalent in dollars to the randomly chosen number from the set of numbers you selected divided by 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If the randomly chosen number is “1”, you receive 50 cents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If “2”, you receive $1, if “3”, you receive $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='50 and so on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A confirmation screen after your report will help you review your selection and the corresponding payment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The order of the two STAGES is randomly determined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In other words, it is equally likely that you first participate in the STAGE allowing only a single number and then participate in the STAGE allowing a set of numbers, or first participate in the STAGE allowing a set of numbers and then participate in the STAGE allowing only a single number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In any case, you will play each STAGE only once.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After the completion of both STAGES, a final review screen will provide the information to receive your Amazon gift card code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the review screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This includes your SCREEN NAME you entered, the amount you will receive, and a randomly generated passcode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Because the experiment is anonymous, we have no means to recover this information for you.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please make sure you either print or take the screenshot of this page for your record, because it is very important when you retrieve your compensation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In the final review screen, you will find a link of a google form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' To receive your amazon gift card you have to copy and paste the link in another browser and then fill your personal information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The google form is created by a staff of the department of economics, who won’t have access to the data of this experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='The experimenter only has a list of SCREEN NAMES and the amount associated with them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experimenter will never be able to access your personal information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the google form which you will have to fill to receive your amazon gift card.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Louis recommends student subjects to report their taxpayer identification information for tax purposes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you are an international student and do not have the taxpayer identification information, please indicate so by entering ‘Foreign’ in the form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you do not have or do not wish to provide the identification information, please indicate that you would like to opt out by entering ’Refuse’ in the form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We are sorry for the inconvenience that we are not able to email you with the gift code directly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This payment process is to ensure anonymity in this experiment, and we appreciate 46 your understanding that the anonymity of the reports constitutes a crucial component of our research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is the end of the instructions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you have any questions, please raise hand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Otherwise, I will provide the link via the chat window.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please copy and paste the link to your browser to generate the SCREEN NAME.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='2 Instructions for Non-Anonymous session (Zoom session;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' video) Welcome to the experiment!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Thank you very much for participating today.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Before we start, I will go over the roster to take attendance to make sure I have everyone registered for the session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' During the attendance please turn your video on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' I will now walk you through the instructions for the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We need your full attention during the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you have trouble with hearing the audio or seeing the shared screen, please let me know.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you have any questions during the instructions, please use the hand- raising feature of Zoom and your question will be answered out loud, so everyone can hear.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The experiment you will be participating in today is an experiment in individual decision making.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the end of the experiment, you will be paid an Amazon gift card.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will receive the show-up fee of $2 for completing the experiment, with the additional amount that depends on your decisions and on chance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The details of the payment will be described later.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' All instructions and descriptions that you will be given in the experiment are accurate and true.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In accordance with the policy of this lab, at no point will we attempt to deceive you in any way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the first page in the main experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the end of the instructions, I will provide the link to the experiment using the chat window.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please copy and paste the link to your browser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The first screen will ask your identification information – your first and last name and your student ID.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After you enter your information, you will proceed to the next screen and take a short quiz to ensure your understanding of the procedures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be able to repeat the quiz if you make mistakes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will have three chances to attempt the quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you fail to get all questions correct after three attempts, you may not participate in today’s experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In such case, you will only receive the show-up fee for today’s experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The main part of the experiment consists of two STAGES after the preliminary quiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In each STAGE you will observe a number that we ask you to remember and later report to us.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The 47 number you report to us determines how much money you will be paid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the end of the experiment, a confirmation screen will summarize today’s experiment and provide the informa- tion to retrieve your payment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' I will give you more details about the observation, reporting, and payment processes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The observation process is identical in both STAGES.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' At the beginning of both STAGES, we will provide a link to Google page that randomly generates a number between 1 and 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The probabilities are equal across the numbers;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' that is, each number is chosen with the same probability of one tenth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We ask you to open the link, remember the number, close the Google page, and report the number on the next screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of an example Google page.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will see that this is a search result for ‘random number between 1 and 10’, and the page displays a randomly generated number that matches the search phrase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Do not click on the ‘generate’ button on the Google page, because the number you see is already randomly generated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Any additional generation only distorts the statistical properties of the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' I will now demonstrate how this Google page works.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will find this link during the experiment, and this is equivalent of opening a google page and typing in “random number between 1 and 10.” When you open the link, a new window pops up.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' As you can see, this number is already randomly generated and you should not generate the number again.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The way you can report differs between STAGES.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In one STAGE, you are allowed to select one number after you observe the randomly generated number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the experiment stage where you can select one number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' By one number, we mean that you may click only one number on the screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We will interpret this selection as a statement that the number you observed is the number you selected, and we will regard this number as your report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The number you report determines how much money you will be paid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be paid the equivalent in dollars to the number you report divided by 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In other words, if you report “1”, you receive 50 cents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you report “2”, you receive $1, if you report “3”, you receive $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='50 and so on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A confirmation screen after your report will help you review your selection and the corresponding payment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In another STAGE, you are allowed to select a set of numbers after you observe the randomly generated number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the experiment stage where you can select a set of numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' By a set of numbers, we mean that you may click multiple numbers on the screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' For instance, you may choose to click on four numbers, one number, two numbers, or even all ten numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you select multiple numbers, we will interpret it as a statement that the number you observed is one of the numbers you selected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you select a 48 single number, we will interpret it as a statement that the number you observed is the number you selected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After you submit your selection, the computer will randomly choose one number from the set of numbers you selected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We will regard this randomly chosen number as your report.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Again, the number you report determines how much money you will be paid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' You will be paid the equivalent in dollars to the randomly chosen number from the set of numbers you selected divided by 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If the randomly chosen number is “1”, you receive 50 cents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If “2”, you receive $1, if “3”, you receive $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content='50 and so on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' A confirmation screen after your report will help you review your selection and the corresponding payment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The order of the two STAGES is randomly determined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In other words, it is equally likely that you either participate in the STAGE allowing a single number first and then participate in the STAGE allowing a set of numbers or participate in the STAGE allowing a set of numbers first and then participate in the STAGE allowing a single number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In any case, you will play each STAGE only once.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' After the completion of both STAGES, a final review screen will summarize today’s exper- iment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is a screenshot of the review screen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' The last screen will ask your email address to receive the Amazon Gift Code of the amount that corresponds to your responses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' We will directly send you Amazon gift code to the email address you provide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please allow us a few hours after the completion of the experiment to validate the data and send the email.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' In addi- tion, Washington University in St.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Louis recommends student subjects to report their taxpayer identification information for tax purposes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you are an international student and do not have the taxpayer identification information, please indicate so by entering ‘Foreign’ in the form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you do not have or do not wish to provide the identification information, please indicate that you would like to opt out by entering ’Refuse’ in the form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' This is the end of instructions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' If you have any questions, please raise your hand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Otherwise, I will provide the link via the chat window.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' Please copy and paste the link to your browser and start the experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} +page_content=' 49' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/PtAyT4oBgHgl3EQfhPgp/content/2301.00372v1.pdf'} diff --git a/RdE3T4oBgHgl3EQfDQlE/content/tmp_files/2301.04284v1.pdf.txt b/RdE3T4oBgHgl3EQfDQlE/content/tmp_files/2301.04284v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..3341d2390c7c42fbfc9fdce3316a3e6ee9ca5892 --- /dev/null +++ b/RdE3T4oBgHgl3EQfDQlE/content/tmp_files/2301.04284v1.pdf.txt @@ -0,0 +1,2801 @@ +Highlights +Ejecta cloud distributions for the statistical analysis of impact cratering events onto asteroids’ +surfaces: a sensitivity analysis +Mirko Trisolini,Camilla Colombo,Yuichi Tsuda +• Development of an ejecta cloud distribution-based model +• Characterisation of asteroid ejecta fate via a representative fragments sampling +• Sensitivity analysis of ejecta fate to modelling techniques and assumptions +arXiv:2301.04284v1 [astro-ph.EP] 11 Jan 2023 + +Ejecta cloud distributions for the statistical analysis of impact +cratering events onto asteroids’ surfaces: a sensitivity analysis +Mirko Trisolinia,∗,1, Camilla Colomboa,2 and Yuichi Tsudab,3 +aPolitecnico di Milano, Via La Masa 34, Milano, 20156, Italy +bInstitute of Space and Astronautical Science (ISAS)/Japan Aerospace Exploration Agency (JAXA), 3-1-1 Yoshinodai, Chuo-ku, Sagamihara, +Kanagawa, 252-5210, Japan +A R T I C L E I N F O +Keywords: +Asteroids +Ejecta fate +Sensitivity analysis +Ejecta distribution +Impact processes +Abstract +This work presents the model of an ejecta cloud distribution to characterise the plume generated +by the impact of a projectile onto asteroids surfaces. A continuum distribution based on the +combination of probability density functions is developed to describe the size, ejection speed, +and ejection angles of the fragments. The ejecta distribution is used to statistically analyse the +fate of the ejecta. By combining the ejecta distribution with a space-filling sampling technique, +we draw samples from the distribution and assigned them a number of representative fragments +so that the evolution in time of a single sample is representative of an ensemble of fragments. +Using this methodology, we analyse the fate of the ejecta as a function of different modelling +techniques and assumptions. We evaluate the effect of different types of distributions, ejection +speed models, coefficients, etc. The results show that some modelling assumptions are more +influential than others and, in some cases, they influence different aspects of the ejecta evolution +such as the share of impacting and escaping fragments or the distribution of impacting fragments +on the asteroid surface. +1. Introduction +The study of impact cratering and ejecta generation has been at the forefront of recent exploration missions toward +small bodies of the Solar System. In 2005, NASA’s mission Deep Impact collided with comet Tempel1 at a speed of +more than 6 km/s (Blume, 2003); in 2019, JAXA’s mission Hayabusa2 carried out an impact experiment on asteroid +Ryugu (Tsuda et al., 2013, 2019, 2020), obtaining images and videos of the impact event and the subsequent crater +formation. In 2021, the NASA Double Asteroid Redirection Test (DART) mission (Cheng et al., 2018) has been +launched toward the Didymos binary system as the first part of the Asteroid Impact and Deflection Assessment (AIDA) +joint mission between ESA and NASA (Cheng et al., 2015; Rivkin et al., 2021). DART has impacted Dimorphos on the +26th of September 2022, while the CubeSat LICIACube (Tortora and Di Tana, 2019) has recorded the event, sending +back to Earth the first images of the impact. The ESA HERA mission (Michel and Kueppers, 2020; Michel et al., +2018) will follow DART as the second part of the AIDA mission to investigate in depth the effects of the impact event, +analysing the crater formation, the deflection, and the fate of the ejecta. +Ejecta models are of utmost importance in studying impact phenomena such as the ones associated with these +missions. They are used to predict the size of the crater, the number of generated fragments, and the initial conditions +of the ejected particles. This information is exploited to predict the fate of ejecta in time and estimate the share of +particles re-impacting and escaping the target body. Most of the ejecta models currently available are based on scaling +relationships that have been observed in experimental impacts. These relationships are based on point-source solutions, +which link impacts of different sizes, velocities, and gravitational accelerations. These scaling relationships are based +upon the Buckingham 휋 theorem (Buckingham, 1914) of dimensional analysis and have had an extensive development +over the years (Holsapple and Housen, 2007, 2012; Housen and Holsapple, 2011). Also based on scaling relationships +and on principles derived by the Maxwell Z-model of crater excavation (Maxwell and Seifert, 1975; Maxwell, 1977), +Richardson developed a further set of scaling relationships (Richardson et al., 2007; Richardson, 2009, 2011) for the +∗Corresponding author +mirko.trisolini@polimi.it (M. Trisolini); camilla.colombo@polimi.it (C. Colombo); tsuda.yuichi@jaxa.jp (Y. Tsuda) +www.cradle.polimi.it (M. Trisolini); www.compass.polimi.it (C. Colombo) +ORCID(s): 0000-0001-9552-3565 (M. Trisolini); 0000-0001-9636-9360 (C. Colombo) +https://www.linkedin.com/profile/view?id=linkedin.com/in/mirkotrisolini/ (M. Trisolini) +Trisolini et al.: Preprint submitted to Icarus +Page 1 of 38 + +aURLinanalysis of the Deep Impact mission to comet Tempel1. In his work, Richardson also derived relationships to describe +the in-plane and out-of-plane components of the ejection angles and how they vary with the impact angle. The work +of Housen and Holsapple has been extensively used in recent years to model impact craters. Several works have been +dedicated to predict the dynamical fate of the ejecta resulting from the impact of the DART spacecraft with Dimorphos. +Yu et al. (2017); Yu and Michel (2018) performed full-scale simulations of the DART impact, modelling the shape +of Didymos as a combination of tetrahedral simplices and Dimorphos as an ellipsoid. In their work, the fate of the +ejected particles is studied drawing a few hundred of thousand of samples, assuming a normal impact and studying +two possible types of materials for the target asteroid. Rossi et al. (2022) also analyse the evolution of the ejecta plume +after the DART impact focusing on the fate of the particles at different time scales, in order to assess the possibility +of the future ESA Hera mission (Michel and Kueppers, 2020) to find particles at its arrival. Fahnestock et al. (2022), +simulate the ejecta plume to obtain synthetic images, representative of the camera view of LICIACube and an Earth- +orbiting telescope. Ejecta models have also been used to analyse the effects on the safety and operations of a mission +scenarios, as it was the case for Hayabusa2 (Soldini and Tsuda, 2017). Additionally, when the impact event can be +observed, such as for Hayabusa2, ejecta models can help characterise the properties of the target body like the type and +strength of the material by comparing the predicted and observed effects (Arakawa et al., 2020). The ejecta models are +also fundamental to understanding and analysing the momentum transfer associated with an impact event, which can +then characterise the deflection capabilities of the impact (Holsapple and Housen, 2012; Rossi et al., 2022). +In this work, we present the development of an ejecta cloud distribution-based model used to describe the ejecta +parameters (i.e., size, launch position, ejection speed, and ejection direction) via Probability Density Functions (PDFs). +The formulation we present starts with a review of existing modelling techniques based on experimental correlations +(Holsapple and Housen, 2007, 2012; Housen and Holsapple, 2011; Richardson et al., 2007; Richardson, 2009, 2011; +Sachse et al., 2015). After the review process, we identified common aspects and differences between the various +modelling techniques. The work of Housen and Holsapple (Holsapple and Housen, 2007, 2012; Housen and Holsapple, +2011) is mainly dedicated to scaling relationships describing normal impacts. As we are interested in a generic model, +also valid for oblique impacts, we integrated the work of Richardson (Richardson et al., 2007; Richardson, 2009, 2011) +to extend the distribution-based model to a generic oblique impact. In addition, following the work of Sachse et al. +(2015), we introduce a correlation between the particle size and the ejection speed, which consider larger particles +more likely to be ejected at lower speeds. +The work of synthesis we have performed on previous experimental correlations has resulted in a modular +formulation in which the ejecta cloud distribution is a combination of probability distribution functions and conditional +distributions that describe the parameters characterising the ejecta. By introducing such a modularity, we allow different +models to be plugged-in, as long as they can be described as PDFs (Trisolini et al., 2022a). In this work, we present +three different formulations of the ejecta model and assess how they affect the prediction of the ejecta fate. To do +so, the ejecta correlations have been reformulated into probability distribution functions and analytical solutions +for the corresponding Cumulative Distribution Functions (CDFs) have been obtained. Leveraging the knowledge +of the CDFs of the ejecta cloud distribution, we can directly generate sample that follow the cloud distribution. In +addition, we can exploit the CDF to analytically integrate the ejecta distribution to estimate the number and mass of +particles within specific initial conditions. This feature is exploited in this work to introduce a sampling methodology +based on representative fragments. With this methodology, each sample represents an ensemble of fragments, to +better characterise the overall behaviour of the ejecta cloud. The associated fragments are estimated integrating the +distribution in a neighbourhood of the selected sample. +The ability to predict the fate of the ejecta generated by an impact relies on the knowledge of the impactor and target +properties, and the impact scenario; however, it also depends on the selection of the ejecta model and the definition of +its parameters. Previous works have considered impacts on specific target asteroids and comets (e.g., comet Tempel1, +asteroid Ryugu, and the Didymos system) and focused on the effect of the target material and equivalent strength +(Richardson et al., 2007; Holsapple and Housen, 2007), the impact location (Yu et al., 2017; Yu and Michel, 2018), +or the asteroid environment. Petit and Farinella (Petit and Farinella, 1993) combined different scaling laws to model +the outcome of the impact between asteroids or other small bodies in the solar system. In this work, instead, we focus +on the ejecta modelling decisions and study how they can affect the overall evolution of the ejecta (Trisolini et al., +2022b). In fact, as several formulations are available in literature for the initialisation of the initial conditions of the +ejecta, it is interesting to compare the different modelling techniques available, integrated within the distribution-based +formulation we developed. Specifically, we investigate, among others, the effect of the particle size range, different +models for the ejection speed, and different distributions for the in-plane and out-of-plane components of the ejection +Trisolini et al.: Preprint submitted to Icarus +Page 2 of 38 + +angles. As the study of the motion and fate of the ejecta relies on the models used for their initialisation, it is important +to understand to what extent different modelling techniques and different assumptions can affect our predictions. +Section 2 describes the developed distribution-based formulation for the ejecta model and its mathematical +derivation. Section 3 introduces the sampling methodology based on the representative fragments, while Section 4 +describes the dynamical environment for the simulations. Section 5 shows the results of the sensitivity analysis on the +modelling techniques and Section 6 summarises the conclusions of the study. +2. Ejecta field distribution +This section provides the description of the modelling technique used to represent the characteristics of an ejecta +field after a hypervelocity impact onto celestial bodies. Specifically, we focus the attention on the impact of small- +kinetic impactor onto small-bodies surfaces. The proposed modelling principles is based on the scaling relations +experimentally derived by Housen and Holsapple (2011); Holsapple and Housen (2007); Richardson et al. (2007) and +leverages their work to build a continuous model, where the ejecta field is described using a combination of probability +density functions so that the particle number density is recovered as a function of the ejection variables. In addition, +the relevant cumulative distributions are derived, which can be used to directly sample the distribution and obtain the +relevant initial conditions after the impact event. Finally, the distribution can be integrated to estimate the number of +particles having specified ejection conditions, allowing a better understanding of the overall fate of the ejected particles. +The most general expression of the ejecta distribution is a function 푝(퐱) that represents the particles number density +as a function of the ejection parameters, 퐱. The ejection parameters considered in this work are the particle radius, 푠, +the particle launch position (i.e., the radial distance from the centre of the crater to the ejection location), 푟, the particle +speed, 푢, and the in-plane and out-of-plane components of the ejection angle, 휉 and 휓, respectively. Fig. 1 shows +the physical meaning of the ejection parameters in a local horizontal reference frame, tangent to the asteroid surface +at the impact point. In general, the size and ejection angles are always present in the models; speed and position are +instead mutually exclusive as they can be related according to experimental correlations (Housen and Holsapple, 2011; +Richardson et al., 2007). Both cases have been analysed: Section 2.1 describes a model that considers the particle’s +launch position, 푟, while Sections 2.2 and 2.3 both describe a model based on the ejection speed, 푢. +Figure 1: Schematics of the ejection parameters used for the ejecta distributions. The reference frame xyz is a local +horizontal frame tangent to the asteroid surface. +The following sections describes the different formulations developed for the ejecta distributions. Three main types +of formulations have been considered. A first formulation, identified as position-based where the distribution variables +are 퐱 = {푠, 푟, 휉, 휓} and the speed is then obtained from the particle’s launch position, 푟. A second formulation, identified +as speed-based, which instead drops the dependency on 푟 and for which the distribution variables are 퐱 = {푠, 푢, 휉, 휓}. +A last formulation that is a correlated version of the speed-based formulation in which the size and speed of the ejected +particles are correlated. In fact, it is reasonable to expect that, on average, smaller particles have higher velocities and +vice versa (Sachse et al., 2015). In the other two formulations, size and speed are not correlated and all particle sizes +can assume any ejection velocity in the distribution range. Section 5 will present a comparison between the different +formulations. +Trisolini et al.: Preprint submitted to Icarus +Page 3 of 38 + +2.1. Position-based distribution +The position-based distribution formulation, derives the particle number density as a function of 푠, 푟, 휉, and 휓. As +previously mentioned the model is based on experimental correlations, which have been mainly developed for normal +impacts. In this work, we wish to derive a general distribution for oblique impacts by combining the results from +Housen and Holsapple (2011) and Richardson et al. (2007). For a normal impact, we consider the distribution in 퐱 to +have the following expression: +푝(퐱) = 푝(푠, 푟, 휉푛, 휓푛) = 푝푠(푠) ⋅ 푝휉푛(휉푛) ⋅ 푝휓푛|푟(휓푛|푟) ⋅ 푝푟(푟), +(1) +where 휉푛 and 휓푛 are the in-plane and out-of-plane ejection angles relative to a normal impact (i.e., an impact +perpendicular to the surface of the target). We observe that, even for a normal impact, the out-of-plane ejection angle, +휓푛, depends on the launch position, 푟 (Richardson et al., 2007; Richardson, 2011); therefore, we represent it with a +conditional distribution. +However, to increase the generality of our formulation, we derive the distribution for a generic oblique impact; +therefore we introduce the following transformation (Richardson et al., 2007; Richardson, 2011): + ∶ +⎧ +⎪ +⎨ +⎪⎩ +휉 = 휉푛 +휓 = 휓푛 − 휋 +6 cos 휙 +( +1−cos 휉 +2 +) ( +1 − +푟 +푟max +)2 +, +(2) +where 휙 is the impact angle, measured from a plane tangent to the target surface, and 푟max is the maximum launch +position, which does not necessarily coincide with the transient crater radius (Housen and Holsapple, 2011). The +variable 휉 is measured from the x-axis (as shown in Fig. 1), which is defined as the projection of the impactor’s +incoming surface-relative velocity vector onto the plane tangent to the target surface. The expression of 휓 of Eq. (2) is +derived from the work of Richardson et al. (2007); Richardson (2011) and expresses the variation of the out-of-plane +ejection angle as function of the distance from the crater’s centre and the in-plane ejection angle, 휉. By using this +transformation, we can obtain the ejecta distribution for an oblique impact, starting from a normal impact as follows: +푝(푠, 푟, 휉, 휓) = 푝(푠, 푟, 휉푛, 휓푛) +|퐽( )| +, +(3) +where |퐽( )| is the determinant of the Jacobian of the transformation of Eq. (2), which, in this case, equals to one. +Alongside the transformation of variables, we need to take into account the dependency of the distributions from other +variables. Sections 2.1.3 and 2.1.4 will show that the in-plane and out-of-plane ejection angles depend on a subset of +the ejection variables so that the expression for the ejecta distribution can be re-written as: +푝(푠, 푟, 휉, 휓) = 푝푠(푠) ⋅ 푝휓|휉,푟(휓|휉, 푟; 휙) ⋅ 푝휉|푟(휉|푟; 휙) ⋅ 푝푟(푟), +(4) +where 푝휓|휉,푟(휓|휉, 푟; 휙) is the conditional distribution of 휓 given 휉 and 푟, and having fixed the impact angle 휙; +푝휉|푟(휉|푟; 휙) is the conditional distribution of 휉 given 푟 and fixing 휙. Therefore, we have a combination of probability +distributions in 푠 and 푟, and conditional distributions in 휓 and 휉. We will now describe in detail all these contributions +to the overall ejecta distribution. +2.1.1. Particle size +In this formulation, the size distribution is considered to be independent from all the other variables. The +distribution derives from the power law expression of the reverse cumulative distribution of the number of particles as +a function of their size (Krivov et al., 2003). +퐺(> 푠) = 푁푟 ⋅ 푠−̄훼. +(5) +Trisolini et al.: Preprint submitted to Icarus +Page 4 of 38 + +Here, 푁푟 is a multiplicative factor that can be determined from mass conservation. Following the notation of Sachse +(Sachse et al., 2015), ̄훼 is the exponent defining the slope of the power law. +Differentiating Eq. (5), we obtain the differential density distribution function, which has the following expression: +푝푠(푠) = 푑퐺(< 푠) +푑푠 += +푑 (푁tot − 퐺(> 푠)) +푑푠 += ̄훼푁푟푠−1−̄훼, +(6) +with 푠min ≤ 푠 ≤ 푠max. We can then obtain 푁푟 from mass conservation as follows: +푀tot = 4 +3휋휌 ∫ +푠max +푠min +푠3푝푠(푠) 푑푠 +→ +푁푟 = +3(3 − ̄훼)푀tot +4̄훼 (푠3−̄훼 +max − 푠3−̄훼 +min +) 휋휌 +, +(7) +where 푀tot is the total mass ejected from the crater, 휌 is the density of the asteroid, and 푠min and 푠max are the minimum +and maximum particle radii, respectively. The minimum and maximum radii are free parameters of the model that +can be selected by the user. Commonly used values are 10-100 µm for the minimum diameter and 1-10 cm for the +maximum one (Yu and Michel, 2018). The total mass ejected can instead be derived from experimental correlations +as follows (Housen and Holsapple, 2011): +푀tot = 푘휌 [(푛2푅푐)3 − (푛1푎)3] , +(8) +where 푎 is the impactor diameter, and 푘, 푛1, and 푛2 are coefficients depending on the type of material and impact derived +from experimental correlations. The works of Housen and Holsapple contain extensive coverage for the derivation and +usage of these parameters. The interested reader is referred to their work (Housen and Holsapple, 2011; Holsapple and +Housen, 2007, 2012). +2.1.2. Launch position +The probability distribution of the launch position can be derived from the expression of the mass ejected within a +distance 푟 from the crater origin. This expression has been derived by Housen (Housen and Holsapple, 2011) and has +the following form: +푀(< 푟) = 푘휌 [푟3 − (푛1푎)3] +with +푛1푎 ≤ 푟 ≤ 푛2푅푐. +(9) +We can thus obtain the CDF in 푟 by simply dividing by the total mass (Eq. (8)). +푃푟(< 푟) = +푘휌 +푀tot +(푟3 − 푟3 +min +) , +(10) +where 푟min = 푛1푎. Analogously, we identify 푟max = 푛2푅푐 as the maximum launch position (i.e., the maximum radial +distance from the centre of the crater). By differentiating the CDF, we get the probability distribution of Eq. (11) +(Pishro-Nik, 2016). +푝푟(푟) = 3푘휌 +푀tot +푟2 +with +푟min ≤ 푟 ≤ 푟max. +(11) +2.1.3. In-plane ejection angle +The distribution of the in-plane ejection angle expresses the azimuthal variation of the ejected samples. For normal +impacts, this distribution is uniform and the fragments are ejected with the same probability within the range 0◦-360◦. +For oblique impacts, instead, the uniformity is lost and the distribution starts to assume more complex patterns such +as the one of Fig. 2. +Trisolini et al.: Preprint submitted to Icarus +Page 5 of 38 + +Figure 2: Image of the ejecta field resulting from an oblique impact on the Moon for a crater of about 0.5 km in diameter +(Richardson, 2011). +Richardson proposed an expression for the distribution of such patterns (Richardson, 2011). However, it was +considered too complex for the implementation in the proposed framework as it is not integrable. Therefore, starting +from the work of Richardson, we propose the following expression: +푝휉|푟(휉|푟; 휙) = 1 +2휋 +[ +1 − cos 휙 +( +cos 2휉 ⋅ cos3 휉 − 1 +5 cos 휉 ⋅ cos4 2휉 +) ( +1 − +푟8 +푟8 +max +)] +. +(12) +In addition, differently from Richardson, we have introduced 푟max, instead of the crater radius, to maintain +the generality of the model among both Richardson (Richardson, 2011), and Housen and Holsapple (Housen and +Holsapple, 2011) correlations. As it is possible to observe, Eq. (12) is a conditional distribution of 휉 given 푟. The +angle 휉 is measured from the direction of the incoming projectile; therefore, 휉 = 180◦ is downstream to the incoming +projectile. An example of the distribution as function of the impact angle is shown in Fig. 3, where we see the peak of +the distribution along the projectile direction and two other smaller peaks on the sizes representing symmetrical lobes. +For a 90° impact angle, the distribution degenerates to uniform. +0 +50 +100 +150 +200 +250 +300 +350 + (deg) +0.05 +0.10 +0.15 +0.20 +0.25 +Impact angle +20.0 +45.0 +60.0 +90.0 +Figure 3: Variation of the in-plane ejection angle lobed probability distribution function with respect to the impact angle. +Trisolini et al.: Preprint submitted to Icarus +Page 6 of 38 + +As discussed in Richardson (2011), a distribution of the type of Eq. (12) cannot fully describe the complexity of the +cratering and ejection phenomenon; however, it can be used for a first analysis of the behaviour of the ejected particles +after the impact. +Alongside the distribution of Eq. (12), we can also introduce a simpler and more manageable description, where +only the main lobe is taken into account and the dependency on the launch position, 푟, is dropped. In this case, we +model the in-plane ejection angle with a Gaussian distribution: +푝휉(휉) =  (휇휉, 휎휉), +(13) +where 휇휉 and 휎휉 are the mean and standard deviation of the distribution respectively. As the main lobe is in the +downstream direction with respect to the incoming projectile, we have 휇휉 = 180◦. For the standard deviation, instead, +we assume it varies linearly with the impact angle as follows: +휎휉 = 2 +5휋 ⋅ +휙 − 휙min +휙max − 휙min +with +휙min ≤ 휙 < 휙max, +(14) +where 휙min and 휙min are the minimum and maximum impact angles, respectively. For the presented model, 휙max = 90◦, +while 휙min = 20◦ as the experimental models on which the distribution formulation is based are valid only down to +an impact angle of 20◦ (Fig. 4). The variation fo the standard deviation in Eq. (14) was extrapolated from the work of +Yamamoto (2002). In addition, care must be taken for normal impacts; in this case, the distribution is not Gaussian +anymore and we fall back to a uniform distribution. +0 +50 +100 +150 +200 +250 +300 +350 + (deg) +0.0 +0.5 +1.0 +1.5 +2.0 +2.5 +Impact angle +20.0 +30.0 +45.0 +60.0 +85.0 +90.0 +Figure 4: Variation of the in-plane ejection angle Gaussian probability distribution function with respect to the impact +angle. +Eqs. (12) and (13) describe two options for the characterisation of the in-plane component of the ejection angle +after an oblique impact. These distributions can be seen as a starting point for the characterisation of the ejecta field. +For example, by using images of the impact event these distributions may be tuned to better describe the event in exam. +A different formulation may also be used; for example, a Gaussian mixture model may better be fitted to image data. +Given the modularity of the presented formulation, such options can be explored in future works. +2.1.4. Out-of-plane ejection angle +The out-of-plane component of the ejection angle, 휓, defines how steep the launch angle of the particles is with +respect to the body surface. In several studies, this angle is assumed constant and equal to 45◦ for all the ejected +particles. However, as also shown by experimental results, different particles will possess different ejection angles. +According to Richardson et al. (2007), the ejection angle is typically within 27° and 63°. +Trisolini et al.: Preprint submitted to Icarus +Page 7 of 38 + +We consider two different options for the distribution in 휓. For the distribution in exam, we first derive the +distribution for normal impacts and then apply the transformation of Eq. (2) to obtain the oblique formulation. The +first option is a spherically uniform distribution as follows: +푝휓푛(휓푛) = +cos 휓푛 +sin 휓푛,max − sin 휓푛,min +with 휓푛,min ≤ 휓푛 ≤ 휓푛,max, +(15) +which is defined within the aforementioned limits of 휓푛,min = 27° and 휓푛,max = 63° (Richardson et al., 2007). Inverting +the transformation of Eq. (2) and substituting into Eq. (15), we obtain the distribution in 휓: +푝휓|휉,푟(휓|휉, 푟; 휙) = +cos (휓 + 퐾휓(휉, 푟; 휙)) +sin 휓푛,max − sin 휓푛,min +with 휓min ≤ 휓 ≤ 휓max, +(16) +where 퐾휓(휉, 푟; 휙) = 휋 +6 cos 휙 +( +1−cos 휉 +2 +) ( +1 − +푟 +푟max +)2 +. Note that that now we have a conditional distribution in 휓, given +푟 and 휉. In addition, 휓min = 휓푛,min − 퐾휓(휉, 푟; 휙) and 휓max = 휓푛,max − 퐾휓(휉, 푟; 휙); therefore, also the limits depend on +휉 and 푟. +For the second option, we start from the work of Richardson et al. (2007). Here, it is observed that the ejection +angle tends to decrease with the distance from the impact point, 푟. A simple linear scaling is considered as follows: +휓푛(푟) = 휓0 − 휓푑 ⋅ +푟 +푟max +, +(17) +where 휓0 = 52.4◦±6.1◦ is the starting angle and 휓푑 = 18.4◦±8.2◦ is the total drop angle, using 2휎 errors (Richardson +et al., 2007). The values of 휓0 and 휓푑 have been derived by laboratory shots performed by Cintala et al. (Cintala et al., +1999). These values can be used as starting points; however, they can be changed and tailored to the specific impact, +using for example, direct imaging of the impact event. As the expression of Eq. (17) is expressed as a combination +of means and standard deviations, we can assume they can be treated as Gaussian distributions and combine them to +have: +푝휓푛|푟(휓푛|푟) =  (휇푛, 휎푛), +(18) +where 푝휓푛|푟(휓푛|푟) is a conditional probability distribution of 휓푛, given the position, 푟. The mean and standard deviation +derive from the combination of the characteristics of the distribution of 휓0 and 휓푑, as follows: +⎧ +⎪ +⎨ +⎪⎩ +휇푛 = 휇0 − 휇푑 ⋅ +푟 +푟max +휎2 +푛 = 휎2 +0 + 휎2 +푑 ⋅ +( +푟 +푟max +)2 +(19) +and 휇0 = 52.4◦, 휎0 = 3.05◦, 휇푑 = 18.4◦, 휎푑 = 4.1◦. Similarly to Eq. (16), we obtain the distribution for an oblique +impact as follows: +푝휓|휉,푟(휓|휉, 푟; 휙) =  (휇푛 − 퐾휓(휉, 푟; 휙), 휎푛) +(20) +Trisolini et al.: Preprint submitted to Icarus +Page 8 of 38 + +2.1.5. Ejection speed +In this position-based formulation, the ejection speed is a derived quantity that depends on the other variables, +which are sampled from the distribution described in Sections 2.1.1 to 2.1.4. The expression for the ejection speed for +oblique impacts has the following form (Richardson et al., 2007): +푢ej = +√ +(푢푛 sin 휓푛 +)2 + +(푢푛 sin 휓푛 +tan 휓 +)2 += 푢푛 ⋅ sin 휓푛 +sin 휓 , +(21) +where 푢푛 is the ejection speed associated to an equivalent normal impact. In principle, Eq. (21) expresses the variation +of the ejection speed as a function of the impact angle. This expression is a function of 푟, 휉, and 휓. +The normal ejection speed, 푢푛 can have different expressions; in this work, we selected two commonly used +experimentally derived expressions. The first one has been derived by Housen et al (Housen and Holsapple, 2011; +Holsapple and Housen, 2007) and has the following form: +푢푛(푟) = 퐶1 ⋅ 푈 ⋅ +[ 푟 +푎 ⋅ +(휌 +훿 +)휈]− 1 +휇 ⋅ +( +1 − +푟 +푟max +)푝 +, +(22) +where 퐶1, 휇, 휈, 푝, and 푛2 are coefficients depending on the target material, 훿 is the density of the impactor and 푈 is +the impactor speed. As Eq. (22) refers to normal impacts, 푈 is the impactor velocity normal to the impact surface. +Therefore, for an oblique impact, 푈 = 푈imp ⋅ sin 휙, where 푈imp is the absolute magnitude of the impactor speed. The +second expression, instead, has been derived by Richardson (Richardson et al., 2007) and is the following: +⎧ +⎪ +⎨ +⎪⎩ +푢푛(푟) += +√ +푢2 +푒 − 퐶2 +푣푝푔 ⋅ 푔 ⋅ 푟 − 퐶2 +푣푝푠 ⋅ 푌 +휌 +푢푒 += 퐶푣푝푔 ⋅ √푔 ⋅ 푅푐,푔 ⋅ +( +푟 +푛2푅푐,푔 +)− 1 +휇 , +(23) +where 퐶푣푝푔 and 퐶푣푝푠 are proportionality constants in the gravity and strength regime, respectively, 푔 is the net +acceleration at the impact point on the surface of the small body accounting for both the small body’s gravitation and its +rotation, 푅푐,푔 is the crater radius in the gravity regime, and 푌 is the strength of the small body material. Fig. 5 compares +the two velocity expression of Eq. (22) and Eq. (23) for the same normal impact. The most noticeable feature is the +different magnitude of the speed in the high-velocity region, where the model derived by Richardson returns higher +speed up to about 200 m s−1. +10 +1 +100 +101 +Launch position (m) +0 +200 +400 +600 +800 +Speed (m/s) +Ejecta speed model +Richardson +Housen +Figure 5: Example of the comparison between the ejection speed obtained with Eq. (22) (orange) and Eq. (23) (blue). +Trisolini et al.: Preprint submitted to Icarus +Page 9 of 38 + +2.2. Speed-based distribution +The second formulation we propose is identified as speed-based. Here, we remove the dependency on the launch +position, 푟; therefore, the ejection location on the target surface is the same for all the particles, thus assuming that +the crater size is small with respect to the target radius. The consequence of this assumption is the substitution of +the position distribution, 푝푟(푟), with a speed distribution, 푝푢(푢). In addition, for the in-plane and out-of-plane ejection +angles, we average out the contribution of the launch position (Sections 2.2.1 and 2.2.2). In this case the transformation +from normal to oblique impact is thus: + ∶ +⎧ +⎪ +⎪ +⎨ +⎪ +⎪⎩ +푢 = +√ +(푢푛 sin 휓푛 +)2 + +( 푢푛 sin 휓푛 +tan 휓 +)2 += 푢푛 ⋅ sin 휓푛 +sin 휓 +휉 = 휉푛 +휓 = 휓푛 − 휋 +18 cos 휙 +( +1−cos 휉 +2 +) ( +1 − 푟min +푟max +)2 += 휓푛 − ̄퐾휓(휉; 휙) +, +(24) +where the first equation derives from Eq. (21), observing that sin 휓 and sin 휓푛 are always between zero and 1, because +the out-of-plane ejection angle belongs to the interval [0◦, 90◦]. The third equation derives from the averaging over 푟 +of the corresponding expression of Eq. (2). We will see in the following sections that introducing the transformation +of Eq. (24), the distribution can then be expressed as follows: +푝(푠, 푢, 휉, 휓) = 푝푠(푠) ⋅ 푝푢|휉,휓(푢|휉, 휓; 휙) ⋅ 푝휓|휉(휓|휉; 휙) ⋅ 푝휉(휉). +(25) +In the following sections, the building blocks of Eq. (25) will be described, except for the size distribution, which +is equivalent to Section 2.1.1. +2.2.1. In-plane ejection angle +In Section 2.1.3, we introduced two different expressions for the in-plane ejection angle distribution. For the +expression based on a single lobe and modelled by a Normal distribution, no further modification is needed as the +expression is already independent from the launch position, 푟. The expression of Eq. (12), instead, depends on 푟; +therefore, we remove this contribution by averaging out in 푟 as follows: +푝휉(휉) = +1 +푟max − 푟min ∫ +푟max +푟min +푝휉|푟(휉|푟) 푑푟 = += 1 +2휋 +[ +1 − ̄퐾휉푟 ⋅ cos 휙 +( +cos 2휉 ⋅ cos3 휉 − 1 +5 cos 휉 ⋅ cos4 2휉 +)] +. +(26) +where ̄퐾휉푟 = +[ +8 − 9 푟min +푟max + +( 푟min +푟max +)9] +∕ [9 (푟max − 푟min +)]. +2.2.2. Out-of-plane ejection angle +In a similar fashion, we can obtain the out-of-plane ejection angle distribution by averaging over 푟 the equations +of Section 2.1.4. For the spherically uniform formulation, the contribution of the launch position is limited to the +transformation of Eq. (2). Therefore, to remove the dependency, we simply use the averaged expression of Eq. (24). +The resulting distribution is analogous to Eq. (16), with the only substitution of 퐾휓(휉, 푟; 휙) with ̄퐾휓(휉; 휙). +For the Gaussian distribution formulation, we proceed in a similar fashion by averaging out Eq. (17) as follows: +̄휓 = +1 +푟max − 푟min ∫ +푟max +푟min +( +휓0 − 휓푑 +푟 +푟max +) +푑푟 += 휓0 − 푟max + 푟min +2 ⋅ 푟max +⋅ 휓푑 +. +(27) +Trisolini et al.: Preprint submitted to Icarus +Page 10 of 38 + +Therefore, following the same procedure of Section 2.1.4, we obtain the distribution of the out-of-plane ejection +angle for a normal impact, averaged over the launch position: +푝휓푛(휓푛) =  ( ̄휇푛, ̄휎푛), +(28) +with modified values of the mean and standard deviations as follows: +⎧ +⎪ +⎨ +⎪⎩ +̄휇푛 += 휇0 − 푟max+푟min +2⋅푟max +⋅ 휇푑 +̄휎2 +푛 += 휎2 +0 + +( 푟max+푟min +2⋅푟max +)2 +⋅ 휎2 +푑 +. +(29) +Finally, the Gaussian model of the out-of-plane ejection angle, 휓, distribution of the speed-based formulation for +oblique impacts has the following expression: +푝휓|휉(휓|휉; 휙) =  ( ̄휇푛 − ̄퐾휓(휉; 휙), ̄휎푛). +(30) +2.2.3. Ejection speed +The ejection speed distribution is the main difference between the position-based and the speed-based formulations. +To derive the speed distribution, we assume the distribution is of the form 푝푢푛(푢푛) = 퐶푢 ⋅푢−1−̄훾 +푛 +, where 푢푛 is the ejection +speed after a normal impact (Sachse et al., 2015) and ̄훾 is an exponent that determines the slope of the speed distribution. +The value of ̄훾 depends on the characteristics of the target material. Comparing the speed distribution with experimental +correlations, we derive that ̄훾 = 3휇 (Trisolini et al., 2022a). To compute the constant 퐶푢, we simply impose that the +integral of the probability density function is equal to unity, so that the final expression of the distribution is: +푝푢푛(푢푛) = +̄훾 +푢−̄훾 +푛,푚푖푛 − 푢−̄훾 +푛,푚푎푥 +푢−1−̄훾 +푛 +. +(31) +The values of the minimum and maximum ejection speeds can be provided by the user. Possible values can be +derived from Eq. (22). Values for the minimum speed can also be derived from the simplified expressions for the knee +velocity, 푣∗ that is the approximate value of the slowest ejecta velocity, which is found from laboratory experiment +(Holsapple and Housen, 2012). The knee velocity has two expressions depending whether the impact is in the strength +(푣∗ +푔) or in the gravity (푣∗ +푠) regime: +⎧ +⎪ +⎨ +⎪⎩ +푣∗ +푔 += 퐾vs +√ +푌 +휌 +푣∗ +푠 += 퐾vg푈 +( +푔 +푎푈2 +) +1 +2+휇 . +(32) +It is interesting to note here the main difference between the speed-based and position-based formulation. The speed +distribution of the position-based formulation (Eq. (23), Eq. (22)), has a minimum ejection speed, at the crater rim, +equal to zero. The speed-based formulation, instead, has a minimum speed, which cannot be equal to zero as Eq. (31) +becomes indefinite. The speed distribution is thus characterised by a power-law with a presence of a low-speed cut-off, +which eliminates the low-velocity transition after the knee velocity (Fig. 29). +The expression of Eq. (31) is for a normal impact. We need now to transform it to an oblique impact. The procedure +is similar to the one of the ejection angles, so that we have: +푝푢(푢) = +푝푢푛(푢푛) +|퐽( )| . +(33) +Trisolini et al.: Preprint submitted to Icarus +Page 11 of 38 + +where 퐽( ) = sin 휓푛 +sin 휓 is the Jacobian of the transformation of Eq. (24). Therefore, we have: +푝푢|휓휉(푢|휓, 휉; 휙) = 푝푢푛(푢푛) ⋅ sin 휓 +sin 휓푛 += += 푝푢푛 +( +푢 sin 휓 +sin (휓 − ̄퐾휓(휉; 휙)) +) +⋅ +sin 휓 +sin (휓 − ̄퐾휓(휉; 휙)) +, +(34) +where we have inverted the expressions of 푢푛 and 휓푛 as functions of 푢 and 휓 to complete the transformation. As in the +previous cases, passing from a normal impact to an oblique impact, the distribution becomes conditioned, so that we +now have a distribution in 푢, given 휓 and 휉. +2.3. Correlated speed-based distribution +This third formulation directly derives form Section 2.2; therefore, they share several characteristics. Specifically, +they share the same distributions of the in-plane and out-of-plane ejection angle components so that, also for this +formulation, we can refer to Sections 2.2.1 and 2.2.2. The main difference, instead, resides in the size and speed +distribution, which in this case is correlated. Therefore, we can write the generic expression for our distribution as +follows: +푝(푠, 푢, 휉, 휓) = 푝푠푢|휉,휓(푠, 푢|휉, 휓; 휙) ⋅ 푝휓|휉(휓|휉; 휙) ⋅ 푝휉(휉). +(35) +In Section 2.3.1, we focus on describing in detail the expression for the correlated size vs. speed distribution. +2.3.1. Size vs. speed +The size-speed correlated distribution is derived from the work of Sachse (Sachse et al., 2015) and has the following +expression: +푝푠,푢푛(푠, 푢푛) = 퐴푠−1−̄훼푢−1−̄훾 +푛 +⋅ Θ +[ +푏푠− ̄훽 − 푢푛 +] +(36) +where Θ is the Heaviside step function, and 퐴, 푏, ̄훼, ̄훽, and ̄훾 are parameters that characterise the shape of the distribution +function. Fig. 6 shows an example of the correlated distribution in size and speed for a normal impact. As we can notice, +the maximum possible ejection speed decreases with increasing particle size. A detailed description of the procedure +to select and derive these parameters for the correlated distribution is presented in Appendix A. +Figure 6: Example of the correlated distribution in size and speed for a normal impact. +Trisolini et al.: Preprint submitted to Icarus +Page 12 of 38 + +Table 1 +Summary of the ejecta distribution formulations and the of the available models. +Var. +Model +Position-based +Speed-based +Correlated +(Section 2.1) +(Section 2.2) +(Section 2.3) +푠 +Eq. (6) +Eq. (6) +Eq. (37) +푟 +Eq. (11) +- +- +푢 +Richardson +Housen +Eqs. (21) and (23) +Eqs. (21) and (22) +Eq. (34) +Eq. (37) +휉 +Lobed +Gaussian +Eq. (12) +Eq. (13) +Eq. (26) +Eq. (13) +Eq. (26) +Eq. (13) +휓 +Uniform +Gaussian +Eq. (16) +Eq. (20) +Eq. (16) +Eq. (30) +Eq. (16) +Eq. (30) +Similarly to Section 2.2.3, the starting expression for the distribution only refers to normal impact. As we are +interested in generic impacts, we must perform a transformation to obtain a distribution that is valid also for oblique +impacts. The transformation is analogous to Eq. (33) of Section 2.2.3 so that we have the following expression for the +correlated distribution: +푝푠,푢|휓,휉(푠, 푢|휓, 휉) = 퐴 ⋅ 푠−1−̄훼 ⋅ 푢−1−̄훾 ⋅ +( +sin 휓 +sin (휓 − ̄퐾휓(휉; 휙)) +)−̄훾 +⋅ +⋅ Θ +[ +푏푠− ̄훽 − +푢 sin 휓 +sin (휓 − ̄퐾휓(휉; 휙)) +] +(37) +2.4. Summary +At this point, it is useful to summarise the different formulations and their relevant models. As we have seen in +Sections 2.1 to 2.3, we have three distributions formulations (i.e., position-based, speed-based, and correlated). For +each of these formulations, we have derived the particle density distribution in the form of a combination of conditional +distributions as function of size, 푠, position, 푟, speed, 푢, and ejection angles, 휉 and 휓. For some of these variables, +namely the speed and the ejection angles, we have identified different modelling techniques based on the type of +formulation and on past literature. Table 1 presents a summary of the models used for the different formulations, along +with the relevant equations, in order to increase the understanding of the structure of the different models proposed. In +addition, this highlights the modular nature of the proposed distribution-based formulations, which can accommodate +different types of models as long as they are expressed in terms of density distributions. +3. Sampling methodology +The sampling methodology is a key ingredient to the understanding of the fate of the ejecta after an impact with a +small body. Having defined our ejecta models via continuous distributions, we can exploit this formulation to sample +the ejecta distribution. Because we have expressed the distributions as the product of either independent or conditional +probability distributions (Eqs. (4), (25) and (35)), we can follow the order of these distributions to perform a random +sampling of each of the presented formulations. As an example, let us consider the position-based distribution of +Eq. (4). In this case, we can directly sample the distribution in size and position using the cumulative distributions +derived from Eqs. (6) and (11), because they are independent from the other variables. Then we can obtain the samples +of the ejection angles. First, we perform the sampling of the in-plane distribution that is a conditional distribution in +푟. To do so, we use the already obtained launch positions and we draw samples from the distribution in 휉 inverting +its cumulative distribution. Now that we have both the samples in 푟 and 휉, we can use the CDF of the conditional +distribution in 휓 to sample the out-of-plane component of the ejection angle. Finally, we have the full set of random +samples from the ejecta distribution. Fig. 7a shows an example of ten thousand samples drawn from a size-speed +distribution. We can observe that most of the samples concentrate in the region of small diameters and small speeds. +This is an expected behaviour as both the size and speed distributions are power laws. It is clear that, to fully characterise +Trisolini et al.: Preprint submitted to Icarus +Page 13 of 38 + +an impact, it is necessary to draw a large amount of samples, particularly if we are also interested in the dynamical fate +of larger particles. +10 +5 +10 +4 +10 +3 +Fragment size (m) +10 +2 +10 +1 +100 +101 +102 +Fragment speed (m/s) +(a) +10 +5 +10 +4 +10 +3 +Fragment size (m) +10 +2 +10 +1 +100 +101 +102 +Fragment speed (m/s) +(b) +Figure 7: Examples of sampling strategies of the size-speed distribution. (a) random variates (b) space-filling. +To address this drawback, different sampling strategies can be assessed. Specifically, in this work, a space-filling +strategy is adopted so that the entire domain defined by the ejecta distribution is covered. Fig. 7b shows an example of +such strategy, using a latin-hypercube algorithm (Bouhlel et al., 2019; Jin et al., 2003) for the sample generation and +a log-log filling technique. That is, the samples are created to best fill a space in log-log coordinates. A space-filling +strategy in both logarithmic and linear coordinates can be performed. In this work. the logarithmic strategy is used for +the ejecta size and speed, while a linear strategy for the ejection angles and ejection position. The idea behind this type +of sampling is that we can have samples that better characterise the ejecta field and, therefore, a more representative +prediction of the ejecta fate around asteroids and other small bodies. In addition, we can exploit the ejecta distribution +to assign to each sample a number of representative fragments. That is, the single sample we have drawn carries +the information of an ensemble of fragments that we associate to it. In this way, also during the propagation we can +understand the behaviour of the whole field of fragments generated by the impact. The computation of the number of +representative fragments comprise the following steps: +• Subdivide the ejecta domain into a grid. The grid can be as fine as needed but such that at least few samples are +in each bin. The grid can be specified either in a linear or in a logarithmic space. +• For each bin, we can integrate the ejecta distribution to compute the number of fragments associated to the bin +(this procedure can be performed analytically using the cumulative distribution functions (Appendix B)). +• Assign to each sample inside the bin a number of representative fragments by equally subdividing the total +fragment associated to the bin by the number of samples in the bin: +̃푛푖푗 +푟 = +푁푗 +푓 +푁푗 +푠 +, +(38) +where ̃푛푖푗 +푟 are the representative fragments associated to the i-th sample belonging to the j-th bin, and 푁푗 +푓 and +푁푗 +푠 are the total fragments and samples of the j-th bin. +Fig. 8 shows an example of the results of the aforementioned procedure for a two-dimensional size-speed +distribution. Fig. 8a shows the particle density distribution (it is possible to observe the grid in which the distribution +is subdivided), while Fig. 8b shows ten thousand samples drawn in a space-filling log-log space and the associated +Trisolini et al.: Preprint submitted to Icarus +Page 14 of 38 + +number of representative fragments that is proportional to the integral of the distribution of Fig. 8a and the number of +samples within each bin. +10 +5 +10 +4 +10 +3 +Fragment size (m) +100 +101 +102 +Fragments speed (m/s) +Integration grid +104 +106 +108 +1010 +1012 +Fragments, nr +(a) +10 +5 +10 +4 +10 +3 +Fragment size (m) +100 +101 +102 +Fragments speed (m/s) +100 +102 +104 +106 +108 +Representative Fragments, nr +(b) +Figure 8: (a) Example of a size-speed distribution with highlighted the integration grid for the representative fragments. +(b) Corresponding set of space-filling samples with associated representative fragments. +4. Dynamics +The adopted dynamical model is the Photo-gravitational Hill Problem that is the extension of the classical Hill +problem to a radiating primary (Soldini and Tsuda, 2017). The equations of motion are expressed in non-dimensional +form in a synodic reference frame centred in the asteroid. The x-axis is along the Sun-asteroid direction, pointing +outwards, the z-axis is along the direction of the angular momentum of the asteroid orbit, and the y-axis completes the +right-hand system. +⎧ +⎪ +⎨ +⎪⎩ +̈푥 − 2 ̇푦 = − 푥 +푟3 + 3푥 + 훽 +̈푦 + 2 ̇푥 = − 푦 +푟3 +̈푧 = − 푧 +푟3 − 푧 +(39) +where 푥, 푦, and 푧 are the non-dimensional particle positions with respect to the centre of the asteroid in the synodic +frame, and 푟 = +√ +푥2 + 푦2 + 푧2 is the particle’s distance from the centre of the asteroid. The lightness parameter 훽 can +be expressed as follows (Pinto et al., 2020): +훽 = 푃0 +푐 +퐴푈2 +휇 +1 +3 +푎 휇2∕3 +Sun +3(1 + 푐R) +2휌p푑p +. +(40) +Where 푃0 = 1367 W m−2 is the solar flux at 1 AU, 푐 is the speed of light, AU is the astronomical unit, 휇Sun and 휇푎 +are the gravitational parameter of the Sun and the asteroid, respectively, 휌p is the particle density and 푑p the particle +diameter. The reflectivity coefficient, 푐R, is a number between 0 and 1, where 1 is for fully reflective surfaces. Eclipses +are taken into account using a cylindrical shadow model via a modified lightness parameter, 훽∗: +훽∗ = +{ +훽 +if 푥 ≤ 0 +훽 ⋅ 푓(휎) +otherwise , +(41) +Trisolini et al.: Preprint submitted to Icarus +Page 15 of 38 + +where 푓(휎) = (1 + 푒−푠⋅휎)−1 is a sigmoid function with steepness parameter 푠, which, in this work is equal to 8 (Pinto +et al., 2020). The variable 휎 = 푟푥 −푅푎, with 푟푥 = +√ +푦2 + 푧2 distance to the 푥-axis, and 푅푎 mean radius of the asteroid. +The Photo-gravitational Hill Problem has been selected as the dynamical model for the study because it is a +relatively simple model that allows taking into account the main forces acting on the ejected fragments, such as the +gravity of the asteroid and the solar radiation pressure. At the same time, it allows maintaining the dynamics sufficiently +simple to focus our analysis on the ejecta distribution models and the effects that different modelling choices have on +the fate of the ejecta. +5. Sensitivity analysis +The objective of this work is to assess how the modelling decisions concerning the ejecta models affect the +overall evolution of the particles around the asteroid. As the initial ejection conditions and the size of the particles are +fundamental in shaping the trajectory evolution of the samples, we want to understand what is the impact of specific +modelling choices. Specifically, in this work we focus on the characteristics of the ejecta model, while we neglect (and +leave to a future study) the effect of the impact location, the size, shape, and orbit of the asteroid. The parameters of +the ejecta distributions considered in the sensitivity analysis are: +• the minimum particle size, 푠min; +• the slope coefficient of the particle size distribution, ̄훼; +• the speed distribution formulation: Housen (Eqs. (21) and (22)) or Richardson (Eqs. (21) and (23)); +• the distribution formulation: position-based (Section 2.1) or speed-based (Section 2.2); +• the correlation between fragment size and ejection speed (Sections 2.2 and 2.3); +• the out-of-plane ejection angle distribution: Uniform or Gaussian; +• the value of the impact angle, 휙. +• the effect of the asteroid rotation; +The selection of these parameters is based on their influence on the ejecta models and on the level of uncertainty +associated to them. In the following sections, each of these points is analysed and their effects on the overall ejecta fate +is discussed. +Target and impactor properties. For a better comparison among the different parameters, we define common +characteristics for the target and the impactor. In addition, we fix some of the parameters of the ejecta models. For +the asteroid, we select a spherical body with a diameter of 1 km, a density of 2.6 g cm−3, an albedo of 0.1, and a +rotational period of 16 h. The semimajor axis of the asteroid’s orbit is 1.755 AU. The rotational axis of the asteroid is +perpendicular to its orbital plane. The characteristics of the asteroid are average values from the NASA Small Bodies +Database (NASA, 2021), except the asteroid radius for which a larger radius has been selected. For the impactor, we +select a mass of 2 kg, a diameter of 0.15 m, and an impact speed of 2 km s−1 that is a geometry equivalent to the small +carry-on impactor used by the Hayabusa2 mission (Tsuda et al., 2019, 2020). The decision of fixing these parameters +is here taken to focus the analysis on the modelling decisions relative to the ejecta model and understand the effect they +can have in the prediction of the fate of the ejecta and particularly to identify the most influential modelling choices. +Simulation setup. The simulations have a time span of 2 months. At the end of the simulation, we refer to the particles +as escaping if their distance is greater than the Hill sphere, as impacting if they re-impact the asteroid’s surface, +and orbiting if they have not impacted nor escaped the asteroid. The impact characteristics are kept constant, with +a normal impact (휙 = 90°) on the North pole of the asteroid, for all the simulations except for Section 5.8. A polar +impact is specifically selected to remove the dependency on the asteroid rotation in the sensitivity analysis of the other +parameters. +Trisolini et al.: Preprint submitted to Icarus +Page 16 of 38 + +Table 2 +Material properties (Housen and Holsapple, 2011). +Sand +WCB +휇 +0.41 +0.46 +퐶1 +0.55 +0.18 +푘 +0.3 +0.3 +푛1 +1.2 +1.2 +푛2 +1.3 +1 +푌ref (MPa) +0 +0.45 +Asteroid material and strength. Two types of materials are used in the analysis, a sand-like very low-strength material +and the moderate-strength Weakly Cemented Basalt (WCB). Table 2 shows the reference properties of these materials. +The coefficients 휇, 퐶1, 푘, 푛1, and 푛2 are characteristics of the material and determine the shape of the ejecta +distributions described in Section 2. 푌ref is the reference strength of the material. These values are typically varied +to perform sensitivity analyses with respect to the asteroid’s strength. However, as we are focusing on the influence +of the ejecta models, the strength values have been fixed to a value of 0 MPa for the sand-like material and 5 kPa +for the WCB. In this last case, we reduced the value with respect to the reference in Table 2 to more closely model +weakly cohesive soils similar to regolith (Holsapple and Housen, 2012; Richardson et al., 2007). Finally, we also select +default parameters for the ejecta distribution. Specifically, the default particle size range is between 푠min = 5e−6 m +and 푠max = 5e−3 m (i.e., particle diameters between 10 µm and 1 cm). In addition, we always assume the ̄훾 coefficient +is equal to 3휇 (Section 2.1.5); therefore, it is only dependent on the material. +Sampling and statistical analysis. Given the statistical nature of the sampling procedure described in Section 3 +and that we associate to each sample a number of representative fragments, for each analysis, we perform multiple +runs. Specifically, we perform 20 runs and in each run we draw 100 000 samples. In this way, we can assess the +robustness and variability of the obtained results, particularly concerning the estimate of the number of fragments +(via the representative fragments). Additionally, it is necessary to define the sampling of the distribution. We adopt +the procedure described in Section 3. Because we are interested in the fate of the ejecta in the neighbourhood of +the asteroid, we limit the sampling to the fragments with an ejection speed below the escape speed of the asteroid. +Therefore, the ejection speed is sampled between 푢min and 푢esc. Similarly, as the launch position is connected to the +ejection speed, even the sampling in 푟 must be limited. Specifically, we sample between 푟esc and 푟max, where 푟esc is +the launch position corresponding to the escape speed. All the ejection location closer than the "escape radius" have +ejection speed greater than the escape speed. For both the escape speed and launch location we use a 16 bins grid with +a log-scaling subdivision. In a similar fashion, the size distribution is sampled between the minimum and maximum +provided particle sizes (푠min, 푠max) on a 20 bins grid with a log-scaling subdivision. The in-plane component of the +ejection angle, 휉 is sampled between 0° and 360° on a linear grid with 36 bins. The out-of-plane component, 휓, is +sampled on a linear grid with 8 subdivisions. The sampling range depends on the model adopted. Between a 휓min and +휓max for a Uniform distribution (see Eq. (16) in Section 2.1.4) and in the range 휇 ± 3휎 for the Gaussian model (see +Eq. (20) in Section 2.1.4). +5.1. Slope coefficient (̄훼) of the size distribution +The coefficient ̄훼 of the size distribution (Section 2.1.1) defines the slope of the distribution that is the number of +particles ejected as a function of their size. Larger ̄훼 coefficients produce distributions with a higher portion of small +fragments. Because the size distribution is a power law, it can change significantly with the value of the slope coefficient. +According to Krivov et al. (2003), typical value of the ̄훼 coefficient for basalt and granite targets is between 2.4 and +2.7, and a good choice in many applications is a value between 2.4 and 2.6. Following the results of Krivov, we test the +sensitivity of the ejecta fate as function of three values of ̄훼 ∈ {2.40, 2.55, 2.70}. For this test case, the position-based +formulation has been used (Section 2.1) with a uniform distribution in the in-plane ejection angle, 휉 (normal impact), +and a Gaussian distribution for the out-of-plane ejection angle, 휓 (Section 2.1.4). Both target materials of Table 2 have +been considered. +First, we compare the fraction of particles orbiting, escaping, and impacting after two months, for the three ̄훼 values +considered. Tables 3 and 4 shows the results for the samples and the estimated fragments, respectively, for the three ̄훼 +Trisolini et al.: Preprint submitted to Icarus +Page 17 of 38 + +Table 3 +Samples fractions and corresponding percent relative standard deviations for the three ̄훼 coefficients and two materials in +exam. +Material +̄훼 +⟨푁imp⟩ +RSDimp +⟨푁esc⟩ +RSDesc +⟨푁orb⟩ +RSDorb +Sand +2.40 +98.48% +0.028% +1.51% +1.87% +0.006% +49.36% +2.55 +98.48% +0.028% +1.51% +1.85% +0.007% +37.25% +2.70 +98.48% +0.039% +1.51% +2.58% +0.006% +44.83% +WCB +2.40 +91.07% +0.031% +8.89% +0.31% +0.039% +9.80% +2.55 +91.07% +0.055% +8.89% +0.54% +0.038% +14.08% +2.70 +91.07% +0.060% +8.89% +0.63% +0.038% +16.97% +Table 4 +Fragments fractions and corresponding percent relative standard deviations for the three ̄훼 coefficients and two materials +in exam. +Material +̄훼 +푁tot +⟨푁imp⟩ +RSDimp +⟨푁esc⟩ +RSDesc +⟨푁orb⟩ +RSDorb +Sand +2.40 +3.94e13 +99.74% +1.23% +0.26% +6.80% +8.8e−10% +67.17% +2.55 +8.11e13 +99.75% +1.03% +0.25% +8.56% +1.2e−9% +148.84% +2.70 +1.57e14 +99.73% +1.59% +0.27% +5.67% +9.5e−10% +224.87% +WCB +2.40 +6.38e10 +76.63% +1.04% +23.37% +2.55% +4.52e−7% +46.14% +2.55 +1.31e11 +76.44% +1.02% +23.56% +1.68% +3.65e−5% +297.92% +2.70 +2.54e11 +75.96% +0.57% +24.04% +1.87% +7.68e−8% +54.32% +values and the two materials considered. Because we are have twenty runs for each simulation, we present the results +for the average percentage of samples and fragments (⟨⋅⟩) and percent Relative Standard Deviation (RSD) (i.e., the +standard deviation divided by the mean). Table 3 shows the statistics for the samples. We can observe that for both +materials most of the samples (more than 90%) re-impact the asteroid. Most of the other samples escape, while only +a very small percentage is still orbiting after two months. The results on the Relative Standard Deviation show stable +results for both the impacting and escaping samples, while a larger variability is associated with the orbiting samples +as they are far fewer. It is interesting to observe that the results are very stable for the different values of ̄훼. Therefore, +the influence of the scaling coefficient, ̄훼, on the samples fate appears to be limited. +Table 4 shows equivalent results for the number of fragments estimated via the representative fragments. Comparing +Table 4 with Table 3 we observe few interesting features. First, the percentage of samples impacting, escaping and +orbiting differs from their corresponding samples, in particular for the WCB case. This is a consequence of the +representative fragments methodology as a different weight is associated to each sample. The orbiting fragments are a +very small percentage in both cases and corresponds to few hundreds or thousands of samples, although their variability +is higher as highlighted by the percent RSD. The combination of a distribution-based ejecta model (Section 2) with the +representative fragments sampling methodology (Section 3) is thus able to characterise the fate of the impacting and +escaping ejecta with a small variability. As for the samples of Table 3, the fate of the ejecta is stable with respect to +the change in ̄훼; the main difference resides in the total number of fragments generated by the impact (see Table 4). In +fact, larger ̄훼 generates more fragments. We can thus infer that the selection of the slope coefficient, ̄훼, only minimally +influences the ultimate fate of the ejecta; however, it scales the total amount of particles generated. +Finally, we also look at the fate of the ejecta as a function of time. Fig. 9 shows the results for the WCB material for +both the impacting and escaping fragments. Specifically, Fig. 9a shows the differential distributions of the impacting +and escaping particles in time for the three analysed values of ̄훼. At each time instant, we appreciate how many particles +escape and impact the asteroid. Fig. 9b instead shows the corresponding cumulative distribution that is the percentage +of particles escaping and impacting within a given time, 푡. From Fig. 9b we can observe that the behaviour of the +normalised distribution is almost identical in all the three cases. The only difference is the absolute number of particles +involved, which can be appreciated from Fig. 9a and Table 4. Therefore, a variation of ̄훼 within the studied interval +does not result in a corresponding variation in the overall behaviour of the fragments. +Trisolini et al.: Preprint submitted to Icarus +Page 18 of 38 + +10 +1 +100 +101 +102 +103 +Time, t (hr) +101 +103 +105 +107 +109 +nr (t) +Fate +Impacting +Escaping +Cases +2.40 +2.55 +2.70 +(a) +10 +1 +100 +101 +102 +103 +Time, t (hr) +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +Fnr ( 0), in all situations, the wave +number (K) is real. Also, the steady state is stable in all +small perturbations. On other side, when the group velocity +dispersion becomes negative (β2<0), the wave number is +imaginary, and the perturbation increases. Therefore in the +negative dispersion, the gain is: +g(Ω) = |β2Ω| +� +Ω2c − Ω2 +(5) +The maximum gain (gmax=2/LNL) also occurs at fre- +quency Ωmax = ± Ωc +√ +2. +Fig. 1, shows the gain spectra for four values of the +nonlinear length (LNL = 1km, 4km, 7km, 10km) for an +optical fiber with β2 = −20ps2/km. +-80 +-60 +-40 +-20 +0 +20 +40 +60 +80 +Frequency shift (GHz) +0 +0.2 +0.4 +0.6 +0.8 +1 +1.2 +1.4 +1.6 +1.8 +2 +Instability Gain (1/km) +LNL=1 km +LNL=4 km +LNL=7 km +LNL=10 km +Fig. 1: Nonlinear fiber modulation instability gain diagram +with the dispersion of the group velocity β2 = −20ps2/km +and nonlinear length LNL=1km, 4km, 7km, 10km. +By considering frequency deviation on the laser wave, the +non-linear fiber with a particular design and under modulation +instability amplifies a specific spectral range of carrier side- +band. Even more, it can work as a filter in this spectral range. +This amplifying/filtering is applied and operated with only the +anomalous fiber and does not require a separate pump [22]. +Also, in some long-distance wireless applications that use an +array antenna, the signal level received by the receiver is lower +than the signal level detected by the receiver. An improved +gain performance can maintain the antenna’s high transmission +capabilities. When antenna systems are required to work at +mm-waves for 5G, the communication distances are drastically +reduced due to the short wavelengths. Electromagnetic (EM) +waves experience a higher signal quality and strength loss by +atmospheric attenuations. In this case, achieving high gain +is essential to reduce path loss in communication networks. +Therefore, by the introduced structure, the modulated signal +can be used to be detected by the receiver. +The comparison between the power laser as input and the +output of fiber under MI conditions is demonstrated in fig. 2. +It is essential to note that by the proposed structure, in which +the length of fiber is L = 11km and LNL = 3.23km, the +sideband level is amplified by almost 18 dBm, and also the +accessible bandwidth is nearly 20 GHz, which indicates the +fundamental of the role of the fiber, especially in the gain case. +, +Fig. 2: Comparison between the output of fiber under MI and +power laser wave as input. +III. Vπ CHALLENGE IN MACH ZEHNDER MODULATOR +To design broadband traveling-wave modulators like MZ, +one of the fundamental challenges to having approved per- +formance, is raising the Vπ with respect to increasing the +frequency. however, in electronic devices like CMOS, this +proportion is inverse [23]. +The block diagram of the structure’s performance of boost- +ing output signal with fiber under MI is demonstrated in fig. 3. +With the help of MZ or any similar modulator in this structure, +it is possible to modulate the Gaussian pulse on the carrier. +Improvement of Vπ is one of the advantages of this proposed +structure. +Fig. 3: Block-diagram of EO modulation structure with the +fiber under the MI. +In the improved MZ modulator, the proportional relation +between Vπ, input/output power, and instability gain is the +following: +PRF,out ∝ 1 +V 2 +π +PRF,inGMI +(6) + +Machzehnder +laser +Photodetector +FibberunderMl +modulator +Input signal +Output signalPowerlaseras input +Output fiberunderMi40 +20 +018dBm +193.05 +193.1Power(dBm) +20 +-40 +-60 +-80 +20GHZ +-100 +192.9 +192.95 +193 +Frequency (THz)4 +Where PRF,in(w), PRF,out(w) and GMI are the input RF +power, the output RF power, and modulation instability gain, +respectively. +MI is stimulated by the sidebands, which results in their +expansion at the expense of the carrier. The sideband fields +grow by G1/2 +MI as the sideband power grows by GMI. The RF +output power is proportional to the G1/2 +MI, as the measured +RF current, is caused by carrier-sideband beating. The Vπ +necessary to obtain desirable modulation depth is decreased +by using the MI in the fiber that leads the MZ modulator. +The MI-induced rise of PRF,out(w) by GMI, is similar to +Vπ decreasing by G1/2 +MI providing the amplified modulator Vπ, +according to eq. 7. +Vπ,eff(wRF ) = Vπ(wRF ) +L1/2 +Dep +G1/2 +MI(wRF ) +(7) +In this equation, L1/2 +Dep <1 is the consumption factor for the +power carrier. +However, in transmitting and receiving communication ap- +plications, the modulator performance improves by increasing +the gain that reduces the Vπ,eff and also helps to facilitate +the detection of the received signal. It is important to note +that the pulse amplification created by a fiber under MI, is +spontaneous and does not need a separate amplifier or another +pump. Fig. 4 shows the comparison of the Vπ in the improved +MZ at different non-linear lengths. +Therefore, it is possible to achieve the improvement at Vπ +with MI. In addition, in photonic microwave structures, the +signal-to-noise ratio (SNR) is related to the inverse square +of the Vπ, which, due to its decrease, increases the SNR +and channel capacity. With respect to the bandwidth and the +amount of gain, we should design the fiber with particular +parameters. +5 +10 +15 +20 +25 +30 +Frequency shift (GHz) +0 +5 +10 +15 +20 +25 +30 +Half-wave voltage V (v) +V under MI(LNL=0.3km) +V under MI(LNL=1km) +V under MI(LNL=3km) +V in EO Modulation +Vπ ∝ f +Vπ,eff ∝ Vπ/G1/2 +MI +Fig. 4: Comparison of Vπ between improved MZ with different +LNL and normal MZ. +IV. TRUE TIME DELAY IN PHASE ARRAY ANTENNA +TTD is one of the essential methods for solving beam- +steering and beamforming problems in wideband PAA com- +munication systems. Large bandwidth, immunity to electro- +magnetic and low loss are some of the advantages of the +optical domain that it is possible to implement with the TTD. +In 5G communication wireless, beamforming can be used +to offer high array gains at both the transmitter (TxBF) and +receiver (RxBF), resulting in increased SNR and more radio +link margin that mitigates propagation route loss. Multiple +beams can be merged at the receiver to reduce path loss even +more. Furthermore, interferences can be decreased, and fre- +quency reuse can be improved using appropriate beamforming +technology. +One of the simplest ways to implement TTD in MW is +to use parallel and separate fibers with low dispersion. The +changing of the time delay is based on the differential length +between the fiber corresponding to the adjacent radiation +element in PAA. However, using an individual fiber for each +radiation element (multi-fiber structure) as the exciter makes +the structure bulky and expensive. +A convenient method is proposed in [24] that uses a single +fiber with a frequency-comb source instead of a multiple-fiber +with a single source. It is important to note that the number of +frequency bands is related to the number of radiated elements +(M). In these types of structures, the differential frequency of +modulated pulses on the adjacent frequency combs generates +a time delay between the pulses in fiber with an associated +dispersion. +Eq. 8 displays the relation between the TTD and the +length of the fiber (L), in which D and ∆λ represent the +fiber’s dispersion and the difference between adjacent comb +wavelengths, respectively. +∆τ = DL∆λ +(8) +According to eq. 8, the theoretical value of optical true time +delay (OTTD) with a relatively large free spectral range (FSR) +of 200 GHz (∆λ = 1.606 nm) and assuming the length of +fiber equivalent to 11km, is achieved as 0.136 ns. Therefore, +the time delay depends on the length of the fiber, dispersion, +and the frequency distance between each pulse. +Θ0 = sin−1 c.mT +dP AA +(9) +In eq. 9 it is clear that modifying the steering angle can +be tuned by two methods: first, it can be accomplished by +changing the FSR. Second, based the eq. 8 by changing the +fiber length, and without changing the type of fiber. +In the first method, the desired frequency range is chosen +from the frequency combs by using a filter. On the other hand, +the second method is more practical and considered because +of its facility. +In an array antenna, beamforming based on TTD has an +essential role in communication systems because it shows +excellent operating frequency and bandwidth performance. +Low loss and immunity to electromagnetic interference are +benefits that increase the quality of wireless communications. + +5 +The array antennas consist of the same and uniformly spaced +radiating where θ0, dP AA and c are radiating steering angle, +spacing between each element, and speed of light in the +vacuum space, respectively. + 0.5 + 1 +30 +210 +60 +240 +90 +270 +120 +300 +150 +330 +180 +0 + 0.5 + 1 +30 +210 +60 +240 +90 +270 +120 +300 +150 +330 +180 +0 +Fig. 5: Simulated beam patterns of 3 microwave frequencies +with TTD-based beamformer and a phase shifter based beam- +former at 42 degrees. +Fig. 5 gives us a piece of important information. Without +using the TTD, the antenna’s beam is dispersed and does +not point to the same point. In another case with TTD, it +is possible to coordinate the beam of the array antenna in the +same direction in all three frequencies at 26, 28, and 31 GHz. +190.5 +191 +191.5 +192 +192.5 +193 +193.5 +194 +194.5 +195 +195.5 +Frequency (THz) +-100 +-80 +-60 +-40 +-20 +0 +20 +Power(dBm) +Fig. 6: Output of the frequency comb with FSR=200 GHz. +Fig. 6 shows the system response is improved due to the +frequency comb source with Fsr = 200 GHz with central +carrier frequency 193 THz. Also, fig. 7 demonstrates the +comparison between the output of the central frequency comb +and the output signal without using fiber under MI. So we +can comprehend that level of the power of the output signal is +increased by almost 14 dBm and the bandwidth, in that case, +is 17 GHz. +V. BEAMFORMING BIT-CONTROLLER SYSTEM +Multiplexing techniques in the time, frequency, code do- +mains, and spatial domains have been essential research areas +to meet the increasing demand for spectral efficiency in fifth- +generation mobile wireless communication systems. Because +of its steering capability and compactness, a PAA is now +Fig. 7: Comparison of the output of the Mach-Zehnder mod- +ulator under MI phenomenon vs. without MI phenomenon. +widely utilized to achieve microwave beamforming. With a +sufficient correlation between the PAA, adaptive beamforming +applies to closely spaced antenna arrays. Therefore it is +possible to improve network coverage, increase signal quality +and exploit the array gain. +Fig. 8 demonstrates the microcomb based on a microwave +TTD beamformer by arbitrary bit-control. Pumping a non- +linear microresonator with a continuous wave laser produces +a broadband optical frequency comb. +Fig. 8: Structure of modulation instability with micro comb +based microwave TTD beamformer with the bit-controller- +based length of the fiber. +With the help of the Mach-Zehnder, the microwave signal +modulates on the frequency comb source, which is then carried +by fiber under MI. Also, this fiber can boost input signals level. +A spectral shaper forms a programmable dispersion matrix +in the next step. With respect to exaggerating the time delay +achieved due to fiber under MI, we use a fiber with positive +dispersion β2 > 0. This elaborated solution has two benefits: +first decreases the overstate time delay that is appropriate for + +Input signal +cw laser +Micro resonator +Amplifier +Mach zehnder +modulator +Fibber under Ml +8L +Programmable +Wavelength demultiplex and +Dispersion matrix +photodetection +4L +Array Antenna +A +2L +DMUX +AOutputof theMZ +Input Signal20 +1014dBm +93.04193.06193.080 +Power(dBm) +-10 +20 +-30 +-40 +-50 +192.92192.94192.96192.98193193.021 +Frequency (THz)6 +PAA, and second, controlling the time delay for PAA by tuning +the minimum time step. +The minimum and maximum fiber lengths used in the +control system are L and 15L. To have the minimum time +delay, we proposed the equation D1L1 = D2L2, in which D1 +and L1 are parameters of the fiber under MI. D2 and L2 are the +parameters of the fiber which are used in the control system. +This time delay can be increased with respect to decreasing +the D2 or L2. +The dispersion matrix is built around a binary delay line +of optical switch devices and dispersive fibers (four bits are +illustrated in fig. 8). +By using the controlling system, it is possible to tune the +length of the fiber, so it is available to control the time delay. +After that, the comb line is photo-detected and de- +multiplexed. Then the generated microwave signals are am- +plified and sent to an antenna array. The beam pattern can +be controlled by shaping the comb spectrum with the spectral +shaper, and the beam direction can be controlled by switching +the dispersion matrix. Fig. 9 shows the time delay state for +PAA that minimum time delay is ∆τ = 1.44ps. Therefore by +changing the binary code, we can tune and control the time +delay. So by changing the time delay, the beam of PAA will +change. +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +Comb number +0 +20 +40 +60 +80 +100 +120 +Time delay (ps) +0001 +0010 +0011 +0100 +0101 +0110 +0111 +1000 +Matrix State +Fig. 9: Time delays of the comb lines with the dispersion +matrix through the states from “0001” to “1000”. +For tuning the angle of the PAA’s main beam with the +time delay between two radiating elements, we adjust the fiber +length with the controlling system. +The wavelength of the laser is fixed at 1550 nm. Based on +4-bit, the target controlled fiber in the dispersion matrix is 115 +m, 57.5 m, 28.75 m, and 14.375 m. +Fig. 10 (a) and fig. 10 (b) show the simulated radiation +patterns for 4 and 8 array antennas for three different cases +at 26, 28, and 31 GHz, respectively. The antenna’s phase +offset was adjusted to direct the peak lobe at a specific angle. +The distance between sub-antennas is 5.4 mm, which is half +the microwave wavelength of 1.08 cm. The ω0 represents the +reference angular frequency, where the value is 2π×193 THz. +Fig. 10: Comparison of the simulated pattern at 26, 28, and +31 GHz for three different angles 8, 24, and 40 degrees. +The beam shape is Gaussian. The beam direction is 42.7 +degrees, and the dispersion matrix state is "1010". For compar- +ison, we also calculated the beam patterns for a phase shifter +controlled PAA. All the frequencies in the PAA are assumed +to be the same as the 26, 28, and 31 GHz. +The rotation angle of cooperative beam-forming, which is +constructed and controlled by the bits are 1101, 1001, and +0101 and with respect to the vertical axis, being nearly 8 +(a), 24 (b), and 40 (c) degrees, respectively. By increasing +the number of radiating elements, the pattern of the array +antenna is narrow, and the effect of the back lobe decreases. So +for high-technology of wireless communication, it is better to +increment the number of antennas to achieve accurate results. +VI. CONCLUSION +The Mach-Zehnder modulator’s function and efficiency us- +ing MI are developed for even transmitting and receiving 5G +applications. With respect to the instability gain of the carrier +sideband frequency at the MI phenomenon, we demonstrate +that we can improve the Vπ in MZ modulator. At first, by +using laser power as a carrier and the fiber under MI, we +achieved 18 dBm and 20 GHz for the sideband’s gain and +bandwidth, respectively. +With the help of the MZ, we modulated the frequency comb +on the carrier with the central frequency 193 THz, so the +gain is 14 dBm, and the bandwitdh is 17 GHz. By using the +fiber under MI, we increment the power of the input without +adding any external device like a pump or making the structure +complex. +To avoid the beam-squint, used the TTD technique to create +a constant time delay for every side radiation element in +PAA. In TTD, instead of employing many fibers for the +excitation of PAA’s radiation element, we used a frequency +comb with FSR=200 GHz. The patterns by using TTD and +then comparing them without using TTD for beamforming at +26, 28, and 31 GHz frequencies are shown, respectively. +The beam is controlled with a bit-control system by using +normal fiber with a length of 2.75 km. The bit-control creates + +90 +1 +90 +1 +90 +1 +120 +60 +120 +1 +60 +120 +60 +- +1 +- +150 +0.5/ +30 +150/ +0.5 +30 +150/ +0.5/ +30 +0 +180 +0 +180 +0 +180 +2101 +330 +210 +/330 +210 +330 +1 +- +240 +300 +240 +300 +240 +300 +270 +270 +27006 +1 +06 +06 +120 +60 +120 +60 +120 +60 +- +- +150 +0.5 +30 +150, +0.5 +30 +150/ +0.5 +30 +180 +0 +180 +180 +0 +210 +/330 +210 +/330 +210 +330 +- +- +- +- +240 +300 +240 +300 +240 +300 +270 +270 +2707 +the minimum time delay=1.44 ps showing three patterns in +8, 24, and 40, degrees with respect to 1101, 1001, and 0101 +binary bit-control. +REFERENCES +[1] R. Waterhouse and D. Novack, “Realizing 5g: Microwave photonics for +5g mobile wireless systems,” IEEE Microwave Magazine, vol. 16, no. 8, +pp. 84–92, 2015. +[2] M. Romagnoli, V. Sorianello, M. Midrio, F. H. Koppens, C. Huyghe- +baert, D. Neumaier, P. Galli, W. Templ, A. D’Errico, and A. C. Ferrari, +“Graphene-based integrated photonics for next-generation datacom and +telecom,” Nature Reviews Materials, vol. 3, no. 10, pp. 392–414, 2018. +[3] I. I. Nureev, O. G. Morozov, A. F. Agliyullin, V. V. Purtov, D. L. Ovchin- +nikov, V. I. Anfinogentov, and V. Y. Vinogradov, “Microwave photonic +polyharmonic probing for fiber optical telecommunication structures +and measuring systems sensors monitoring,” in Optical Technologies in +Telecommunications 2017, vol. 10774. +International Society for Optics +and Photonics, 2018, p. 107741J. +[4] T. Li, E. H. W. Chan, X. Wang, X. Feng, B.-O. Guan, and J. Yao, “Broad- +band photonic microwave signal processor with frequency up/down +conversion and phase shifting capability,” IEEE Photonics Journal, +vol. 10, no. 1, pp. 1–12, 2017. +[5] M. Chen, H. Yu, B. Yang, Y. Li, H. Chen, and S. Xie, “A silicon +integrated microwave-photonic transceiver,” in 2017 Optical Fiber Com- +munications Conference and Exhibition (OFC). +IEEE, 2017, pp. 1–3. +[6] X. Zou, W. Bai, W. Chen, P. Li, B. Lu, G. Yu, W. Pan, B. Luo, L. Yan, +and L. Shao, “Microwave photonics for featured applications in high- +speed railways: communications, detection, and sensing,” Journal of +Lightwave Technology, vol. 36, no. 19, pp. 4337–4346, 2018. +[7] J. Hervás, A. L. Ricchiuti, W. Li, N. H. Zhu, C. R. Fernández-Pousa, +S. Sales, M. Li, and J. Capmany, “Microwave photonics for optical +sensors,” IEEE Journal of Selected Topics in Quantum Electronics, +vol. 23, no. 2, pp. 327–339, 2017. +[8] D. Tosi, E. Schena, C. Molardi, and S. Korganbayev, “Fiber optic +sensors for sub-centimeter spatially resolved measurements: Review and +biomedical applications,” Optical Fiber Technology, vol. 43, pp. 6–19, +2018. +[9] S. Mukherjee, L. Udpa, S. Udpa, E. J. Rothwell, and Y. Deng, “A +time reversal-based microwave imaging system for detection of breast +tumors,” IEEE Transactions on Microwave Theory and Techniques, +vol. 67, no. 5, pp. 2062–2075, 2019. +[10] T. Ren, M. Zhang, C. Wang, L. Shao, C. Reimer, Y. Zhang, O. King, +R. Esman, T. Cullen, and M. Lonˇcar, “An integrated low-voltage broad- +band lithium niobate phase modulator,” IEEE Photonics Technology +Letters, vol. 31, no. 11, pp. 889–892, 2019. +[11] G. K. Gopalakrishnan, W. K. Burns, R. W. McElhanon, C. H. Bul- +mer, and A. S. Greenblatt, “Performance and modeling of broadband +linbo/sub 3/traveling wave optical intensity modulators,” Journal of +Lightwave Technology, vol. 12, no. 10, pp. 1807–1819, 1994. +[12] D. A. Presti, V. Guarepi, F. Videla, A. Fasciszewski, and G. A. Torchia, +“Intensity modulator fabricated in linbo3 by femtosecond laser writing,” +Optics and Lasers in Engineering, vol. 111, pp. 222–226, 2018. +[13] A. Rueda, F. Sedlmeir, M. Kumari, G. Leuchs, and H. G. Schwefel, +“Resonant electro-optic frequency comb,” Nature, vol. 568, no. 7752, +pp. 378–381, 2019. +[14] J. Sun, R. Kumar, M. Sakib, J. B. Driscoll, H. Jayatilleka, and H. Rong, +“A 128 gb/s pam4 silicon microring modulator with integrated thermo- +optic resonance tuning,” Journal of Lightwave Technology, vol. 37, no. 1, +pp. 110–115, 2018. +[15] Y. Sobu, T. Simoyama, S. Tanaka, Y. Tanaka, and K. Morito, “70 gbaud +operation of all-silicon mach–zehnder modulator based on forward- +biased pin diodes and passive equalizer,” in 2019 24th OptoElectronics +and Communications Conference (OECC) and 2019 International Con- +ference on Photonics in Switching and Computing (PSC). +IEEE, 2019, +pp. 1–3. +[16] A. Rao, A. Patil, P. Rabiei, A. Honardoost, R. DeSalvo, A. Paolella, +and S. Fathpour, “High-performance and linear thin-film lithium niobate +mach–zehnder modulators on silicon up to 50 ghz,” Optics letters, +vol. 41, no. 24, pp. 5700–5703, 2016. +[17] P. Steglich, C. Mai, D. Stolarek, S. Lischke, S. Kupijai, C. Villringer, +S. Pulwer, F. Heinrich, J. Bauer, S. Meister et al., “Novel ring resonator +combining strong field confinement with high optical quality factor,” +IEEE Photonics Technology Letters, vol. 27, no. 20, pp. 2197–2200, +2015. +[18] I. Demirtzioglou, C. Lacava, K. R. Bottrill, D. J. Thomson, G. T. Reed, +D. J. Richardson, and P. Petropoulos, “Frequency comb generation in +a silicon ring resonator modulator,” Optics express, vol. 26, no. 2, pp. +790–796, 2018. +[19] J. Fatome, B. Kibler, F. Leo, A. Bendahmane, G.-L. Oppo, B. Garbin, +S. G. Murdoch, M. Erkintalo, and S. Coen, “Polarization modulation +instability in a nonlinear fiber kerr resonator,” Optics Letters, vol. 45, +no. 18, pp. 5069–5072, 2020. +[20] G. P. Agrawal, “Nonlinear fiber optics,” in Nonlinear Science at the +Dawn of the 21st Century. +Springer, 2000, pp. 195–211. +[21] N. A. Kudryashov, “Model of propagation pulses in an optical fiber with +a new law of refractive indices,” Optik, vol. 248, p. 168160, 2021. +[22] P. T. DeVore, D. Borlaug, and B. Jalali, “Enhancing electrooptic mod- +ulators using modulation instability,” physica status solidi (RRL)–Rapid +Research Letters, vol. 7, no. 8, pp. 566–570, 2013. +[23] J. Golden and D. O’Malley, “Reverse annealing for nonnegative/binary +matrix factorization,” Plos one, vol. 16, no. 1, p. e0244026, 2021. +[24] X. Xu, J. Wu, T. G. Nguyen, T. Moein, S. T. Chu, B. E. Little, +R. Morandotti, A. Mitchell, and D. J. Moss, “Photonic microwave true +time delays for phased array antennas using a 49 ghz fsr integrated +optical micro-comb source,” Photonics Research, vol. 6, no. 5, pp. B30– +B36, 2018. + diff --git a/VdAzT4oBgHgl3EQfl_1k/content/tmp_files/load_file.txt b/VdAzT4oBgHgl3EQfl_1k/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..7884f00895a7d127a363fe769beb721284368430 --- /dev/null +++ b/VdAzT4oBgHgl3EQfl_1k/content/tmp_files/load_file.txt @@ -0,0 +1,557 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf,len=556 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='01556v1 [physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='optics] 4 Jan 2023 1 Performance Improvement of True Time Delay Based Centralized Beamforming Control with the Modulation Instability Phenomenon for Wireless-Array Antennas Seyed Moin Alavi, Hassan Zakeri, Rasul Azizpour, and Gholamreza Moradi Abstract—In this paper, we develop the performance of the modulator by using modulation instability (MI) for fifth- generation wireless communication (5G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Initially, we demon- strate how to improve the modulator’s bias voltage (Vπ) with the help of carrier side-band gain due to the MI phenomenon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' It’s also possible to obtain acceptable gain and bandwidth without complicating the modulator structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In receive mode of the array antenna, where the signal is very weak, amplification in this frequency range has a critical role.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' However, we present a comfortable and simple microwave- photonic beamforming bit-controller system based on the length of the fiber for receiver-transmitter phased array antennas (PAAs) that are used in high-technology wireless communication like 5G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' We used the true-time delay (TTD) technique based on frequency comb to achieve beamforming of the PAA and show that the beams remain unchanged at a different frequency for different angles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Finally, we offer the bit-control system, that the results and the benefit is illustrated by the patterns of PAA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Index Terms—Array Antennas, Fifth-generation, Mach- Zehnder, Modulation instability, True-time-delay I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' INTRODUCTION M Icrowave Photonics (MWP) is the connection between the opto-electronic and radio-frequency (RF) engineer- ing where, in optical methods and devices, high-speed radio- frequency signals are generated, transmitted, controlled, and measured.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' This technique offers a large amount of processing bandwidth and provides more flexibility for information and communication technology (ICT) systems and networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' As a result, it has become widely used in a variety of appli- cations in recent years, including wireless systems like the fifth-generation mobile communication (5G) [1], [2], optical telecommunications [3], broadband photonic signal processors [4], transceiver circuits [5], medical imaging systems using terahertz (THz) waves [6], and sensors [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The basic principle of MWP is that analog microwave electrical signals are transmitted through the optical fiber or photonic devices like modulators and detectors, with electrical- to-optical (EO) and optical-to-electrical (OE) transformation on the transmitting and receiving sides of the photonic link, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Seyed Moin Alavi, Hassan Zakeri, Rasul Azizpour, and Gholamreza Moradi are with the Department of Electrical Engineering, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran (E-mails: moeinalavi@aut.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ir, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='Zakeri@aut.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ir, ra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='azizpour@aut.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ir and, Gh- moradi@aut.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='ir).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In biomedical applications, Gaussian pulse is used to detect and locate the exact tumor location by utilizing the modulators [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' By reducing the pulse length by expanding bandwidth in these applications, the ability to distinguish close tumors is improved [9], which can be applied to next-generation appli- cations that demand more precise identification and distinction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' For designing the broadband traveling-wave modulators,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' there are four fundamental challenges for achieving approved performance: I) matching microwave-optical velocity to avoid excitation of undesired modes on the substrate and negative effects by mismatch caused by phase velocity in optical response,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' II) reducing losses in microwave electrodes,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' III) increasing the half-wave voltage (Vπ),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' IV) improving the impedance of the electrode (almost 50 Ω) to match better and reducing the loss of structure [10]–[12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Using these methods make it possible to increase the modulation depth, so in this paper, we focus on reducing the Vπ to improve the modulator’s response.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Theoretically, it has been shown and analyzed that the amplitude of electro-optic (EO) modulation remains stable at the sideband frequency without any losses [13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In the experi- mental case, essential factors like plasmonic losses, impedance mismatching, and microwave optical velocity can decrease the signal’s modulation depth and amplitude by increasing sideband frequency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Also, the interaction between modulation voltage and arm length in EO modulators is constant and stable (VπL=α where α is constant).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Consequently, the arm length under the EO effect increases by decreasing Vπ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' According to the bandwidth associated with the time con- stant (τ) due to the EO arm, the loss drops faster by increasing frequency, and the bandwidth decreases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' So in typical cases, between Vπ and bandwidth, there is a directly proportional relationship.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore by increasing the bandwidth, the am- plitude of voltage increases too [11], [14], [15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Decreasing mismatching between microwave-optical veloc- ities is one of the recent methods for boosting the bandwidth range of modulator systems by keeping the Vπ constant [16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In that case, it is also possible to decrease the arm length by changing the structure of the modulator, type of excitation, and material of components.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The effective light and microwave refractive index neared each other when the plasmonic dielectric and the time constant decreased, and bandwidth rose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The complexity of a modulator increases, so the cost of the modulator structure and the range 2 of bandwidth is not acceptable for new applications like 5G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Another way to achieve the same modulation depth is to use a special material with an improved EO factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' It is possible to achieve more bandwidth range and decrease the losses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Nevertheless, this method is not convenient because of the high cost of this particular material, the incompatibility of this material with the EO integrated circuits, and the low effectiveness of the EO factor [17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In [18], another method is proposed that by using different resonators (like ring resonators) and features of the EO circuit, it is possible to create a structure with a low amplitude of Vπ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' These structures have more complexity than others in manufacturing the devices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Because of the deployment of the resonant feature, the modulation response fluctuates in the range of modulation depths, so the output answer is distorted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore, all of the mentioned methods can only slightly improve the Vπ or bandwidth of the modulator and cannot significantly improve both parameters simultaneously, so we need to employ simple and low-cost practical applications to enhance the modulator bandwidth and Vπ without altering the internal structure of the modulator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In this paper, we propose a novel method by which it is pos- sible to use MI phenomena in optic fiber for 5G applications in PAA in transmitting/receiving (T/R) applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The output to this fiber can repay the attenuation of the frequency response and reduces the required Vπ of the modu- lator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In addition, the fiber can increase the power level in an optical link in amplifying and filtering applications [19] [20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Increasing the gain, controlling the beamforming of the array antenna, and having a simple structure are the advantages of this method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' One of the main problems in PAAs is beam-squint, which is related to the divergence from the antenna’s frequency and the structure of PAAs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Because the number of radiating elements controls the beamwidth in PAAs, a large channel number is necessary for the true-time delay (TTD) to solve the beam- squint problem and improve angular resolution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' A convenient method is proposed in [6] that replaces a single fiber and a frequency-comb source with a multiple- fiber and a single source.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' It is important to note that the number of frequency bands is related to the number of radiated elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In these structures, the frequency of the difference pulses modulated on each frequency comb causes a time delay between the pulses in a fiber with a suitable dispersion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore, we demonstrate that using the time delay makes it possible to increase the gain and improve the answer of the optical modulator and the bandwidth without adding any complexity to the structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Also, we propose a new bit- controlled system that can control the directivity angle of PAA only with the length of the fiber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The rest of the paper is organized as follows: In Section II, the MI phenomenon in optical fiber is analyzed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Section III shows the MI effect on Vπ and performance enhances the intrinsically weak electro-optic development.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Section IV introduces the RF and microwave TTD based on Kerr combs and their applications in PAAs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Section V shows a bit-control system based on the fiber length for controlling the beam direction of PAA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' FIBER UNDER MODULATION INSTABILITY Modulation instability (MI) is a nonlinear phenomenon induced by the interaction of non-linearity, dispersion, and diffraction in the presence of an intense optical carrier wave traveling through a nonlinear medium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' At the output, the exponential expansion of the spectral sidebands with the carrier wave’s center band is expected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' While the carrier is a CW laser optical wave, rapid fluctuations appear as modulated pulse trains [20], [21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Because distinct spectral components are associated with the pulse travel at different speeds determined by c n(w), fiber dispersion is crucial in propagating short optical pulses and optical communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The effects of fiber dispersion are compensated for by expanding the mode-propagation constant in a Taylor series around the frequency w0 at which the pulse spectrum is centered on accounting for the effects of fiber dispersion, that the representative equation is as follows: B(w) = ∞ � m=0 Bm m!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' (w − w0)m (1) The parameters n(w) are group indexes, and β1 and β2 represent the group delay and dispersion of the group velocity, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The non-linear response of polarization to a strong optical field causes non-linear effects in optical fibers, which is another issue to consider when studying pulse propagation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The lowest-order non-linear effects in optical fibers are the third-order effects since they are constructed of silica, which has an amorphous microstructure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Third-order non-linear phenomena in optical fibers include: The optical Kerr effect, four-wave mixing, stimulated Brillouin scattering, stimulated Raman scattering, and third-harmonic production.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The optical Kerr effect has a non-linear reaction that occurs instantly and reacts to light intensity variation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' When considering the propagation of strong pulses in op- tical fibers, both fiber dispersion and non-linearity must be regarded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In that case, using the non-linear Schrodinger equation for optical pulse propagation as follows: ı∂A ∂z + ıα 2 A − β2 2 ∂2A ∂T 2 + γ |A|2 A = 0 (2) where A, γ, and α are the slowly varying envelope of the optical pulse, the non-linear parameter, and the loss factor of fiber, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' (2) is easily solved to obtain the steady-state continuous radiation solution when the time derivation is ignored.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In the case of the lossless response of the laser, the continu- ous wave for the above equation is a soliton in the form of √ P 0eiγp0z that P0 and φNL = γp0z are the incident power and the nonlinear phase shift induced by self-phase modulation, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' If the steady-state is stable against small perturbations in the power of the laser, the solution is: A = ( √ P 0 + a1ei(Kz−Ωt) + a2e−i(Kz−Ωt))eiγp0z (3) 3 where K is the wave number and Ω is frequency perturbation at sideband frequency of laser spectrum [20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' By replacing Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' (3) in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' (2) it’s possible to achieve: K = ±1 2 |B2Ω| � Ω2 + sgn(β2)Ω2 c � 1 2 (4) which Ωc has an inverse relation with β2 and the nonlinear length LNL (Ωc = 2/ � |β2| LNL).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' When the group velocity dispersion is positive (β2 > 0), in all situations, the wave number (K) is real.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Also, the steady state is stable in all small perturbations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' On other side, when the group velocity dispersion becomes negative (β2<0), the wave number is imaginary, and the perturbation increases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore in the negative dispersion, the gain is: g(Ω) = |β2Ω| � Ω2c − Ω2 (5) The maximum gain (gmax=2/LNL) also occurs at fre- quency Ωmax = ± Ωc √ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1, shows the gain spectra for four values of the nonlinear length (LNL = 1km, 4km, 7km, 10km) for an optical fiber with β2 = −20ps2/km.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 80 60 40 20 0 20 40 60 80 Frequency shift (GHz) 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='8 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='4 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='6 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='8 2 Instability Gain (1/km) LNL=1 km LNL=4 km LNL=7 km LNL=10 km Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1: Nonlinear fiber modulation instability gain diagram with the dispersion of the group velocity β2 = −20ps2/km and nonlinear length LNL=1km, 4km, 7km, 10km.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' By considering frequency deviation on the laser wave, the non-linear fiber with a particular design and under modulation instability amplifies a specific spectral range of carrier side- band.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Even more, it can work as a filter in this spectral range.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' This amplifying/filtering is applied and operated with only the anomalous fiber and does not require a separate pump [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Also, in some long-distance wireless applications that use an array antenna, the signal level received by the receiver is lower than the signal level detected by the receiver.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' An improved gain performance can maintain the antenna’s high transmission capabilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' When antenna systems are required to work at mm-waves for 5G, the communication distances are drastically reduced due to the short wavelengths.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Electromagnetic (EM) waves experience a higher signal quality and strength loss by atmospheric attenuations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In this case, achieving high gain is essential to reduce path loss in communication networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore, by the introduced structure, the modulated signal can be used to be detected by the receiver.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The comparison between the power laser as input and the output of fiber under MI conditions is demonstrated in fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' It is essential to note that by the proposed structure, in which the length of fiber is L = 11km and LNL = 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='23km, the sideband level is amplified by almost 18 dBm, and also the accessible bandwidth is nearly 20 GHz, which indicates the fundamental of the role of the fiber, especially in the gain case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' , Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 2: Comparison between the output of fiber under MI and power laser wave as input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Vπ CHALLENGE IN MACH ZEHNDER MODULATOR To design broadband traveling-wave modulators like MZ, one of the fundamental challenges to having approved per- formance, is raising the Vπ with respect to increasing the frequency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' however, in electronic devices like CMOS, this proportion is inverse [23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The block diagram of the structure’s performance of boost- ing output signal with fiber under MI is demonstrated in fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With the help of MZ or any similar modulator in this structure, it is possible to modulate the Gaussian pulse on the carrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Improvement of Vπ is one of the advantages of this proposed structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 3: Block-diagram of EO modulation structure with the fiber under the MI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In the improved MZ modulator, the proportional relation between Vπ, input/output power, and instability gain is the following: PRF,out ∝ 1 V 2 π PRF,inGMI (6) Machzehnder laser Photodetector FibberunderMl modulator Input signal Output signalPowerlaseras input Output fiberunderMi40 20 018dBm 193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='05 193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='1Power(dBm) 20 40 60 80 20GHZ 100 192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='9 192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='95 193 Frequency (THz)4 Where PRF,in(w), PRF,out(w) and GMI are the input RF power, the output RF power, and modulation instability gain, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' MI is stimulated by the sidebands, which results in their expansion at the expense of the carrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The sideband fields grow by G1/2 MI as the sideband power grows by GMI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The RF output power is proportional to the G1/2 MI, as the measured RF current, is caused by carrier-sideband beating.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The Vπ necessary to obtain desirable modulation depth is decreased by using the MI in the fiber that leads the MZ modulator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The MI-induced rise of PRF,out(w) by GMI, is similar to Vπ decreasing by G1/2 MI providing the amplified modulator Vπ, according to eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Vπ,eff(wRF ) = Vπ(wRF ) L1/2 Dep G1/2 MI(wRF ) (7) In this equation, L1/2 Dep <1 is the consumption factor for the power carrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' However, in transmitting and receiving communication ap- plications, the modulator performance improves by increasing the gain that reduces the Vπ,eff and also helps to facilitate the detection of the received signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' It is important to note that the pulse amplification created by a fiber under MI, is spontaneous and does not need a separate amplifier or another pump.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 4 shows the comparison of the Vπ in the improved MZ at different non-linear lengths.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore, it is possible to achieve the improvement at Vπ with MI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In addition, in photonic microwave structures, the signal-to-noise ratio (SNR) is related to the inverse square of the Vπ, which, due to its decrease, increases the SNR and channel capacity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With respect to the bandwidth and the amount of gain, we should design the fiber with particular parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5 10 15 20 25 30 Frequency shift (GHz) 0 5 10 15 20 25 30 Half-wave voltage V (v) V under MI(LNL=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='3km) V under MI(LNL=1km) V under MI(LNL=3km) V in EO Modulation Vπ ∝ f Vπ,eff ∝ Vπ/G1/2 MI Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 4: Comparison of Vπ between improved MZ with different LNL and normal MZ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' TRUE TIME DELAY IN PHASE ARRAY ANTENNA TTD is one of the essential methods for solving beam- steering and beamforming problems in wideband PAA com- munication systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Large bandwidth, immunity to electro- magnetic and low loss are some of the advantages of the optical domain that it is possible to implement with the TTD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In 5G communication wireless, beamforming can be used to offer high array gains at both the transmitter (TxBF) and receiver (RxBF), resulting in increased SNR and more radio link margin that mitigates propagation route loss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Multiple beams can be merged at the receiver to reduce path loss even more.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Furthermore, interferences can be decreased, and fre- quency reuse can be improved using appropriate beamforming technology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' One of the simplest ways to implement TTD in MW is to use parallel and separate fibers with low dispersion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The changing of the time delay is based on the differential length between the fiber corresponding to the adjacent radiation element in PAA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' However, using an individual fiber for each radiation element (multi-fiber structure) as the exciter makes the structure bulky and expensive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' A convenient method is proposed in [24] that uses a single fiber with a frequency-comb source instead of a multiple-fiber with a single source.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' It is important to note that the number of frequency bands is related to the number of radiated elements (M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In these types of structures, the differential frequency of modulated pulses on the adjacent frequency combs generates a time delay between the pulses in fiber with an associated dispersion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8 displays the relation between the TTD and the length of the fiber (L), in which D and ∆λ represent the fiber’s dispersion and the difference between adjacent comb wavelengths, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' ∆τ = DL∆λ (8) According to eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8, the theoretical value of optical true time delay (OTTD) with a relatively large free spectral range (FSR) of 200 GHz (∆λ = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='606 nm) and assuming the length of fiber equivalent to 11km, is achieved as 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='136 ns.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore, the time delay depends on the length of the fiber, dispersion, and the frequency distance between each pulse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Θ0 = sin−1 c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='mT dP AA (9) In eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 9 it is clear that modifying the steering angle can be tuned by two methods: first, it can be accomplished by changing the FSR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Second, based the eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8 by changing the fiber length, and without changing the type of fiber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In the first method, the desired frequency range is chosen from the frequency combs by using a filter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' On the other hand, the second method is more practical and considered because of its facility.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In an array antenna, beamforming based on TTD has an essential role in communication systems because it shows excellent operating frequency and bandwidth performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Low loss and immunity to electromagnetic interference are benefits that increase the quality of wireless communications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5 The array antennas consist of the same and uniformly spaced radiating where θ0, dP AA and c are radiating steering angle, spacing between each element, and speed of light in the vacuum space, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 1 30 210 60 240 90 270 120 300 150 330 180 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 1 30 210 60 240 90 270 120 300 150 330 180 0 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5: Simulated beam patterns of 3 microwave frequencies with TTD-based beamformer and a phase shifter based beam- former at 42 degrees.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5 gives us a piece of important information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Without using the TTD, the antenna’s beam is dispersed and does not point to the same point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In another case with TTD, it is possible to coordinate the beam of the array antenna in the same direction in all three frequencies at 26, 28, and 31 GHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 190.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 191 191.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 192 192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 193 193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 194 194.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 195 195.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 Frequency (THz) 100 80 60 40 20 0 20 Power(dBm) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 6: Output of the frequency comb with FSR=200 GHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 6 shows the system response is improved due to the frequency comb source with Fsr = 200 GHz with central carrier frequency 193 THz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Also, fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 7 demonstrates the comparison between the output of the central frequency comb and the output signal without using fiber under MI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' So we can comprehend that level of the power of the output signal is increased by almost 14 dBm and the bandwidth, in that case, is 17 GHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' BEAMFORMING BIT-CONTROLLER SYSTEM Multiplexing techniques in the time, frequency, code do- mains, and spatial domains have been essential research areas to meet the increasing demand for spectral efficiency in fifth- generation mobile wireless communication systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Because of its steering capability and compactness, a PAA is now Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 7: Comparison of the output of the Mach-Zehnder mod- ulator under MI phenomenon vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' without MI phenomenon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' widely utilized to achieve microwave beamforming.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With a sufficient correlation between the PAA, adaptive beamforming applies to closely spaced antenna arrays.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore it is possible to improve network coverage, increase signal quality and exploit the array gain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8 demonstrates the microcomb based on a microwave TTD beamformer by arbitrary bit-control.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Pumping a non- linear microresonator with a continuous wave laser produces a broadband optical frequency comb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8: Structure of modulation instability with micro comb based microwave TTD beamformer with the bit-controller- based length of the fiber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With the help of the Mach-Zehnder, the microwave signal modulates on the frequency comb source, which is then carried by fiber under MI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Also, this fiber can boost input signals level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' A spectral shaper forms a programmable dispersion matrix in the next step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With respect to exaggerating the time delay achieved due to fiber under MI, we use a fiber with positive dispersion β2 > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' This elaborated solution has two benefits: first decreases the overstate time delay that is appropriate for Input signal cw laser Micro resonator Amplifier Mach zehnder modulator Fibber under Ml 8L Programmable Wavelength demultiplex and Dispersion matrix photodetection 4L Array Antenna A 2L DMUX AOutputof theMZ Input Signal20 1014dBm 93.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='04193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='06193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='080 Power(dBm) 10 20 30 40 50 192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='92192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='94192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='96192.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='98193193.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='021 Frequency (THz)6 PAA, and second, controlling the time delay for PAA by tuning the minimum time step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The minimum and maximum fiber lengths used in the control system are L and 15L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' To have the minimum time delay, we proposed the equation D1L1 = D2L2, in which D1 and L1 are parameters of the fiber under MI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' D2 and L2 are the parameters of the fiber which are used in the control system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' This time delay can be increased with respect to decreasing the D2 or L2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The dispersion matrix is built around a binary delay line of optical switch devices and dispersive fibers (four bits are illustrated in fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' By using the controlling system, it is possible to tune the length of the fiber, so it is available to control the time delay.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' After that, the comb line is photo-detected and de- multiplexed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Then the generated microwave signals are am- plified and sent to an antenna array.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The beam pattern can be controlled by shaping the comb spectrum with the spectral shaper, and the beam direction can be controlled by switching the dispersion matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 9 shows the time delay state for PAA that minimum time delay is ∆τ = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='44ps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Therefore by changing the binary code, we can tune and control the time delay.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' So by changing the time delay, the beam of PAA will change.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1 2 3 4 5 6 7 8 9 10 Comb number 0 20 40 60 80 100 120 Time delay (ps) 0001 0010 0011 0100 0101 0110 0111 1000 Matrix State Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 9: Time delays of the comb lines with the dispersion matrix through the states from “0001” to “1000”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' For tuning the angle of the PAA’s main beam with the time delay between two radiating elements, we adjust the fiber length with the controlling system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The wavelength of the laser is fixed at 1550 nm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Based on 4-bit, the target controlled fiber in the dispersion matrix is 115 m, 57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 m, 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='75 m, and 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='375 m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10 (a) and fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10 (b) show the simulated radiation patterns for 4 and 8 array antennas for three different cases at 26, 28, and 31 GHz, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The antenna’s phase offset was adjusted to direct the peak lobe at a specific angle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The distance between sub-antennas is 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='4 mm, which is half the microwave wavelength of 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='08 cm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The ω0 represents the reference angular frequency, where the value is 2π×193 THz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10: Comparison of the simulated pattern at 26, 28, and 31 GHz for three different angles 8, 24, and 40 degrees.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The beam shape is Gaussian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The beam direction is 42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='7 degrees, and the dispersion matrix state is "1010".' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' For compar- ison, we also calculated the beam patterns for a phase shifter controlled PAA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' All the frequencies in the PAA are assumed to be the same as the 26, 28, and 31 GHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The rotation angle of cooperative beam-forming, which is constructed and controlled by the bits are 1101, 1001, and 0101 and with respect to the vertical axis, being nearly 8 (a), 24 (b), and 40 (c) degrees, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' By increasing the number of radiating elements, the pattern of the array antenna is narrow, and the effect of the back lobe decreases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' So for high-technology of wireless communication, it is better to increment the number of antennas to achieve accurate results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' VI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' CONCLUSION The Mach-Zehnder modulator’s function and efficiency us- ing MI are developed for even transmitting and receiving 5G applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With respect to the instability gain of the carrier sideband frequency at the MI phenomenon, we demonstrate that we can improve the Vπ in MZ modulator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' At first, by using laser power as a carrier and the fiber under MI, we achieved 18 dBm and 20 GHz for the sideband’s gain and bandwidth, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' With the help of the MZ, we modulated the frequency comb on the carrier with the central frequency 193 THz, so the gain is 14 dBm, and the bandwitdh is 17 GHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' By using the fiber under MI, we increment the power of the input without adding any external device like a pump or making the structure complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' To avoid the beam-squint, used the TTD technique to create a constant time delay for every side radiation element in PAA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' In TTD, instead of employing many fibers for the excitation of PAA’s radiation element, we used a frequency comb with FSR=200 GHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The patterns by using TTD and then comparing them without using TTD for beamforming at 26, 28, and 31 GHz frequencies are shown, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The beam is controlled with a bit-control system by using normal fiber with a length of 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='75 km.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' The bit-control creates 90 1 90 1 90 1 120 60 120 1 60 120 60 1 150 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5/ 30 150/ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 30 150/ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5/ 30 0 180 0 180 0 180 2101 330 210 /330 210 330 1 240 300 240 300 240 300 270 270 27006 1 06 06 120 60 120 60 120 60 150 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 30 150, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 30 150/ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='5 30 180 0 180 180 0 210 /330 210 /330 210 330 240 300 240 300 240 300 270 270 2707 the minimum time delay=1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='44 ps showing three patterns in 8, 24, and 40, degrees with respect to 1101, 1001, and 0101 binary bit-control.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' REFERENCES [1] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Waterhouse and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Novack, “Realizing 5g: Microwave photonics for 5g mobile wireless systems,” IEEE Microwave Magazine, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 16, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 84–92, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [2] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Romagnoli, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Sorianello, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Midrio, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Koppens, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Huyghe- baert, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Neumaier, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Galli, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Templ, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' D’Errico, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Ferrari, “Graphene-based integrated photonics for next-generation datacom and telecom,” Nature Reviews Materials, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 3, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 392–414, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [3] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Nureev, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Morozov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Agliyullin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Purtov, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Ovchin- nikov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Anfinogentov, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Vinogradov, “Microwave photonic polyharmonic probing for fiber optical telecommunication structures and measuring systems sensors monitoring,” in Optical Technologies in Telecommunications 2017, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10774.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' International Society for Optics and Photonics, 2018, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 107741J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [4] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Li, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Chan, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Feng, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='-O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Guan, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Yao, “Broad- band photonic microwave signal processor with frequency up/down conversion and phase shifting capability,” IEEE Photonics Journal, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1–12, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [5] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Chen, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Yu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Yang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Li, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Chen, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Xie, “A silicon integrated microwave-photonic transceiver,” in 2017 Optical Fiber Com- munications Conference and Exhibition (OFC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' IEEE, 2017, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1–3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [6] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Zou, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Bai, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Chen, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Li, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Lu, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Yu, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Pan, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Luo, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Yan, and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Shao, “Microwave photonics for featured applications in high- speed railways: communications, detection, and sensing,” Journal of Lightwave Technology, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 36, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 19, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 4337–4346, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [7] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Hervás, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Ricchiuti, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Li, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Zhu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fernández-Pousa, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Sales, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Li, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Capmany, “Microwave photonics for optical sensors,” IEEE Journal of Selected Topics in Quantum Electronics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 23, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 327–339, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [8] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Tosi, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Schena, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Molardi, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Korganbayev, “Fiber optic sensors for sub-centimeter spatially resolved measurements: Review and biomedical applications,” Optical Fiber Technology, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 43, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 6–19, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [9] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Mukherjee, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Udpa, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Udpa, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Rothwell, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Deng, “A time reversal-based microwave imaging system for detection of breast tumors,” IEEE Transactions on Microwave Theory and Techniques, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 67, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 2062–2075, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [10] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Ren, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Shao, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Reimer, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Zhang, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' King, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Esman, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Cullen, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Lonˇcar, “An integrated low-voltage broad- band lithium niobate phase modulator,” IEEE Photonics Technology Letters, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 31, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 11, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 889–892, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [11] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Gopalakrishnan, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Burns, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' McElhanon, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Bul- mer, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Greenblatt, “Performance and modeling of broadband linbo/sub 3/traveling wave optical intensity modulators,” Journal of Lightwave Technology, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 12, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 10, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1807–1819, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [12] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Presti, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Guarepi, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Videla, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fasciszewski, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Torchia, “Intensity modulator fabricated in linbo3 by femtosecond laser writing,” Optics and Lasers in Engineering, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 111, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 222–226, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [13] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Rueda, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Sedlmeir, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Kumari, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Leuchs, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Schwefel, “Resonant electro-optic frequency comb,” Nature, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 568, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 7752, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 378–381, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [14] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Sun, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Kumar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Sakib, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Driscoll, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Jayatilleka, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Rong, “A 128 gb/s pam4 silicon microring modulator with integrated thermo- optic resonance tuning,” Journal of Lightwave Technology, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 37, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 110–115, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [15] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Sobu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Simoyama, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Tanaka, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Tanaka, and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Morito, “70 gbaud operation of all-silicon mach–zehnder modulator based on forward- biased pin diodes and passive equalizer,” in 2019 24th OptoElectronics and Communications Conference (OECC) and 2019 International Con- ference on Photonics in Switching and Computing (PSC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' IEEE, 2019, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1–3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [16] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Rao, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Patil, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Rabiei, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Honardoost, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' DeSalvo, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Paolella, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fathpour, “High-performance and linear thin-film lithium niobate mach–zehnder modulators on silicon up to 50 ghz,” Optics letters, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 41, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 24, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5700–5703, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [17] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Steglich, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Mai, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Stolarek, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Lischke, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Kupijai, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Villringer, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Pulwer, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Heinrich, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Bauer, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Meister et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=', “Novel ring resonator combining strong field confinement with high optical quality factor,” IEEE Photonics Technology Letters, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 27, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 20, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 2197–2200, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [18] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Demirtzioglou, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Lacava, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Bottrill, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Thomson, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Reed, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Richardson, and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Petropoulos, “Frequency comb generation in a silicon ring resonator modulator,” Optics express, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 26, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 790–796, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [19] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Fatome, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Kibler, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Leo, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Bendahmane, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content='-L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Oppo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Garbin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Murdoch, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Erkintalo, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Coen, “Polarization modulation instability in a nonlinear fiber kerr resonator,” Optics Letters, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 45, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 18, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5069–5072, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [20] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Agrawal, “Nonlinear fiber optics,” in Nonlinear Science at the Dawn of the 21st Century.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Springer, 2000, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 195–211.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [21] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Kudryashov, “Model of propagation pulses in an optical fiber with a new law of refractive indices,” Optik, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 248, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 168160, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [22] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' DeVore, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Borlaug, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Jalali, “Enhancing electrooptic mod- ulators using modulation instability,” physica status solidi (RRL)–Rapid Research Letters, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 7, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 8, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 566–570, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [23] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Golden and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' O’Malley, “Reverse annealing for nonnegative/binary matrix factorization,” Plos one, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 16, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' e0244026, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' [24] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Xu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Wu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Nguyen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Moein, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Chu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Little, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Morandotti, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Mitchell, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' Moss, “Photonic microwave true time delays for phased array antennas using a 49 ghz fsr integrated optical micro-comb source,” Photonics Research, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 6, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} +page_content=' B30– B36, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAzT4oBgHgl3EQfl_1k/content/2301.01556v1.pdf'} diff --git a/VtA0T4oBgHgl3EQfEv_z/content/tmp_files/2301.02024v1.pdf.txt b/VtA0T4oBgHgl3EQfEv_z/content/tmp_files/2301.02024v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..9f176929543545098ef1eaebd1589037e7b0c361 --- /dev/null +++ b/VtA0T4oBgHgl3EQfEv_z/content/tmp_files/2301.02024v1.pdf.txt @@ -0,0 +1,1066 @@ +arXiv:2301.02024v1 [math.NA] 5 Jan 2023 +Port-Hamiltonian Systems Modelling +in Electrical Engineering +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +Abstract The port-Hamiltonian modelling framework allows for models that +preserve essential physical properties such as energy conservation or dissipa- +tive inequalities. If all subsystems are modelled as port-Hamiltonian systems +and the inputs are related to the output in a linear manner, the overall system +can be modelled as a port-Hamiltonian system (PHS), too, which preserves +the properties of the underlying subsystems. If the coupling is given by a +skew-symmetric matrix, as usual in many applications, the overall system +can be easily derived from the subsystems without the need of introducing +dummy variables and therefore artificially increasing the complexity of the +system. Hence the PHS framework is especially suitable for modelling multi- +physical systems. +In this paper, we show that port-Hamiltonian systems are a natural general- +ization of Hamiltonian systems, define coupled port-Hamiltonian systems as +ordinary and differential-algebraic equations. To highlight the suitability for +electrical engineering applications, we derive PHS models for MNA network +equations, electromagnetic devices and coupled systems thereof. +Andreas Bartel, Michael G¨unther +Bergische Universit¨at Wuppertal, IMACM, Chair of Applied Mathematics, Gaußstraße 20, +D-42119 Wuppertal, e-mail: [bartel,guenther]@uni-wuppertal.de +Markus Clemens +Bergische Universit¨at Wuppertal, IMACM, Chair of Electromagnetic Theory, Rainer- +Gruenter-Straße 21, D-42119 Wuppertal, e-mail: clemens@uni-wuppertal.de +Birgit Jacob +Bergische Universit¨at Wuppertal, IMACM, Chair of Functional Analysis, Gaußstraße 20, +D-42119 Wuppertal, e-mail: bjacob@uni-wuppertal.de +Timo Reis +TU Ilmenau, Fakult¨at f¨ur Mathematik und Naturwissenschaften, Chair of System Theory +and PDEs, PF 10 05 65, D-98684 Ilmenau, e-mail: timo.reis@tu-ilmenau.de +1 + +2 +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +1 Port-Hamiltonian Systems Modelling in a Nutshell +Port-Hamiltonian Systems (PHS) are a generalization of Hamiltonian systems +˙x = J · ∇H(x), +x(0) = x0 +(1) +with x = (p, q) consisting of generalized position q(t) ∈ Rn and momentum +p(t) ∈ Rn (where t ∈ [0, T ]), the skew-symmetric matrix J given by +J = +� 0 −I +I +0 +� +and the Hamiltonian H(x) = H(p, q) = U(p) + V (q) given as the sum of +potential and kinetic energy, which maps Rn × Rn → R and is twice contin- +uously differentiable. The Hamiltonian flow ϕ(t; x0)), i.e., the solution of (1) +at time point t, starting at the initial value x(0) = x0, is characterized by +four geometric properties: +1. Preservation of the Hamiltonian: +d +dtH(ϕ(t; x0)) = (∇H(ϕ(t; x0)))⊤J(∇H(ϕ(t; x0))) = 0. +2. Time-reversibility: +ρ ◦ ϕ(t; x0) ◦ ρ ◦ ϕ(t; x0) = x0, +with ρ(p, q) = (−p, q), which is a direct consequence of the ρ-reversibility +of the Hamiltonian flow: ρ ◦ J∇H(ϕ(t; x0))) = −J∇H(ρ ◦ ϕ(t; x0))). +3. Symplectic structure of the Hamiltonian flow: +Ψ(t)⊤J−1Ψ(t) = J−1, +Ψ(t) := ∂ϕ(t;x0) +∂x0 +, +which is a direct consequence of the skew-symmetry of J. +4. Volume-preservation: +(det Ψ(t))2 = 1, +which follows immediately from the symplectic structure in 3. +First generalization step: arbitrary skew-symmetric matrices J +If we replace in (1) J by an arbitrary skew-symmetric matrix, the Hamiltonian +is still preserved. As x will loose its characterization as generalized positions +and momenta of classical mechanics, time-reversibility will generally not hold +anymore. However, the symplectic structure of the flow still holds in the +case of a regular J, and volume preservation is still a consequence of the +Hamiltonian flow. + +Port-Hamiltonian Systems Modelling in Electrical Engineering +3 +Second generalization step: adding dissipation to the system +Allowing the flow to become dissipative, we may generalize (1) to the dissi- +pative Hamiltonian system +˙x = (J − R) · ∇H(x), +x(0) = x0 +(2) +with R ≥ 0 being symmetric and positive semi-definite. In this case, the flow +will neither be symplectic nor volume preserving, and the preservation of the +Hamiltonian is replaced by the dissipativity condition +d +dtH(x(t)) = (∇H(x))⊤ ˙x = −(∇H(x))⊤R∇H(x) ≤ 0 +⇒ H(x(t)) = H(x0) − +� t +0 +(∇H(x(τ)))⊤R∇H(x(τ)) dτ ≤ H(x0). +Third generalization step: coupling to the environment via inputs and +outputs +Allowing for inputs and outputs to couple the system to the environment, we +end up with linear port-Hamiltonian system characterized by +˙x = (J − R) · ∇H(x) + Bu(t), +x(0) = x0, +y = B⊤∇H(x) +with inputs u(t) ∈ Rp, outputs y(t) ∈ Rp and port-matrices B ∈ Rn×p. The +dissipativity inequality now reads +d +dtH(x(t)) = (∇H(x))⊤ ˙x = −(∇H(x))⊤R∇H(x) + (∇H(x))⊤Bu(t)) += −(∇H(x))⊤R∇H(x) + y(t)⊤u(t) ≤ y(t)⊤u(t) +⇒ H(x(t)) = H(x0) − +� t +0 +(∇H(x(τ)))⊤R∇H(x(τ)) dτ + +� t +0 +y(τ)⊤u(τ)dτ +≤ H(x0) + +� t +0 +y(τ)⊤u(τ)dτ. +Fourth generalization step: PH-DAE systems +Linear PHS can be easily generalized to PH-DAE systems given by +d +dt(Ex) = (J − R) · z(x) + Bu(t), +x(0) = x0, +(3a) +y = B⊤z(x) +(3b) +with a possibly singular matrix E ∈ Rn×n and the nonlinear mapping +z : Rn → Rn fulfilling the compatibility condition E⊤z = ∇H. Now the + +4 +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +dissipativity condition reads +H(x(t)) = H(x0) − +� t +0 +z(x(τ))⊤R∇z(x(τ)) dτ + +� t +0 +y(τ)⊤u(τ) dτ +≤ H(x0) + +� t +0 +y(τ)⊤u(τ) dτ. +The key point in port-Hamiltonian modelling is the following: there is an easy +way to couple arbitrary many PH-DAE system such that the overall system +is still a PH-DAE system, which preserves a dissipativity inequality. +Let us consider r autonomous PH-DAE systems +d +dt(Eixi) = (Ji − Ri)zi(xi) + Biui, +(4a) +yi = B⊤ +i zi(xi) +(4b) +with r Hamiltonians H1, H2, . . . , Hr and compatibility conditions E⊤ +i zi = +∇Hi. If the inputs and outputs fulfill a linear interconnection relation +Mu + Ny = 0 for the aggregated input u = (u1, u2, . . . , ur) and output +y = (y1, y2, . . . , yr), it has been shown in [13] that one can write the aggre- +gated system as a joint PH-DAE system as +d +dt + + + + + + +E 0 0 +0 0 0 +0 0 0 +0 0 0 + + + + +x +ˆu +ˆy + + + + + + = + + +J − R +B +0 +0 +−B⊤ +0 +Im −M ⊤ +0 +−Im 0 −N ⊤ +0 +M +N +0 + + + + +z(x) +ˆu +ˆy +0 + + + + + +0 +0 +Im +0 + + u, +y = ˆy, +with z(x)⊤ = (z1(x1)⊤, z2(x2)⊤, . . . , zr(xr)⊤), new dummy variables ˆu, ˆy +and setting X = diag(X1, X2, . . . , Xr) for X ∈ {E, J, R, B}. This coupling +property of PH-DAE systems makes the port-Hamiltonian modelling frame- +work well suited for multiphysical applications. +Now, we consider external, time dependent inputs. To this end, we split +the inputs and outputs into external (bar-notation) and internal ones (hat- +notation), i.e., Biui is split into ¯Bi¯ui + ˆBiˆui. Then, the subsystem (4) reads +d +dt(Eixi) = (Ji − Ri)zi(xi) + ˆBiˆui + ¯Bi¯ui, +(5a) +ˆyi = ˆB⊤ +i zi(xi), +(5b) +¯yi = ¯B⊤ +i zi(xi). +(5c) +For the coupling relation (of the internal quantities) ˆu + Cˆy = 0 with a +skew-symmetric matrix C = −C⊤ (which often arises in application), these +systems can be written as a joint PH-DAE system in condensed form [8]: + +Port-Hamiltonian Systems Modelling in Electrical Engineering +5 +d +dt(Ex) = ( ˜J − R)z(x) + ¯B¯u, +(6a) +¯y = ¯B⊤z(x) +(6b) +with the condensed skew-symmetric matrix ˜J = J − ˆBC ˆB⊤. Note that in +this case all internal coupling modelled via the port-matrices ˆBi has now +been transferred into the off-block diagonal elements of the skew-symmetric +matrix ˜J, i.e., − ˆBC ˆB⊤. +A systems theoretic treatment of port-Hamiltonian systems goes back to +Bernhard Maschke and Arjan van der Schaft (see [12, 14] for an +overview), where nonlinear systems governed by ordinary differential equa- +tions are treated. For simplicity of presentation, we will (a) not follow the +differential geometric path via Dirac structures, (b) neglect a feed-through +from input to output and (c) only consider finite dimensional systems, i.e., +ordinary (ODEs) and differential-algebraic equations (DAEs), but no partial +differential equations (PDEs). For simulation, the latter are usually trans- +formed into ODEs and DAEs by spatial semi-discretization. For a differential +geometric setting of PHS see [15] and an introduction into PH-PDEs see [11]. +The paper is organized as follows: In the next section we introduce PH- +DAE systems which allow for a general nonlinear dissipative part. A PHS- +DAE formulation of the MNA network equations is derived in Sect. 3, and +for electromagnetic devices in Sect. 4. Section 5 discusses PHS formulation +of coupled EM/circuit systems, which allow for monotolithic as well as weak +coupling simulation approaches. Sect. 6 finishes with conclusions. +2 PH-DAE systems +When dealing with applications in electrical engineering, the concept of port- +Hamiltonian modelling has to be generalized to coupled differential-algebraic +equations, which (a) allow for a general nonlinear resistive part r(z) instead +of a quasilinear setting Rz as in the approach of [13] and (b) has only to be +accretive on a subspace V ⊂ R according to the constraints of the system. +A differential-algebraic equation of the form +d +dtEx(t) = Jz(x(t)) − r(z(x(t))) + Bu(t), +y(t) = B⊤z(x(t)) +(7) +is called a port-Hamiltonian differential-algebraic equation (PH-DAE) [8], if +the following holds: +• E ∈ Rn×n, J ∈ Rn×n and B ∈ Rn×m, z, r : Rn → Rn. +• There exists a subspace V ⊂ Rn with the following properties: + +6 +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +(i) for all intervals I ⊂ R and functions u : I → Rm such that (7) has +a solution x : I → Rn, it holds z(x(t)) ∈ V for all t ∈ I. +(ii) J is skew-symmetric on V. That is, +v⊤Jw = −w⊤Jv for all v, w ∈ V. +(iii) r is accretive on V. That is, v⊤r(v) ≥ 0 for all v ∈ V. +• There exists some function H ∈ C1(Rn, R) such that ∇H(x) = E⊤z(x) +for all x ∈ z−1(V). +Remark 1. a) The PH-DAE (7) system provides the usual energy balance +d +dtH(x(t)) = −z(x(t))⊤r(z(x(t)))) + y(t)⊤u(t) ≤ y(t)⊤u(t). +b) PH-DAE subsystems now read +d +dtEixi(t) =Jizi(xi(t)) − ri +� +zi(xi(t)) +� ++ Biui(t), +(8a) +yi(t) =B⊤ +i zi +� +xi(t) +� +(8b) +instead of (4), and if they are coupled by a skew-symmetric coupling rela- +tion ˆu + Cˆy = 0 with a skew-symmetric matrix C = −C⊤ as before, they +can be condensed into an overall PH-DAE system +d +dtEx = ˆJz − r + ¯B¯u, +(9a) +¯y = ¯B⊤z +(9b) +with the skew-symmetric matrix ˆJ again given by ˆJ = J − ˆB ˆC ˆB⊤. +3 Electrical networks +We consider the classical charge-/flux oriented MNA network equations [8,9] +d +dt + + +0 0 0 AC 0 +0 0 0 0 I +0 0 0 0 0 +0 0 0 0 0 +0 0 0 0 0 + + + + +e +L +V +qC +φL + + += + + +0 −AL −AV 0 0 +A⊤ +L +0 +0 +0 0 +A⊤ +V +0 +0 +0 0 +0 +0 +0 +0 0 +0 +0 +0 +0 0 + + + + +e +L +V +qC +φL + + +− + + +ARg(A⊤ +Re) +0 +0 +qC − q(A⊤ +Ce) +φL − φ(L) + + ++ + + +−AI +0 +0 +0 +0 +−I +0 +0 +0 +0 + + +�ı(t) +v(t) +� + +Port-Hamiltonian Systems Modelling in Electrical Engineering +7 +with e, L and V denoting node potentials and currents through flux storing +elements and voltages sources, qC and Φl charge and flux-storing elements, +i(t) and v(t) independent current and voltage sources, the resistive currents +g and the incidence matrices AC, AL, AR, AV , AI for charge- and flux storing +elements, resistive elements, voltage and current sources, and seek a formu- +lation as a PH-DAE system. For this, we need the following assumptions, +which naturally occur in circuit simulation, see [8]: +(a) Soundness. The circuit graph has at least one branch and is connected. +Furthermore, it contains neither V -loops nor I-cutsets. Equivalently, AV +and (AC AR AL AV )⊤ have full column rank. +(b) Passivity. The functions q, φ and g fulfill +(i) q : +RnC → +RnC and φ : +RnL → +RnL are bijective, continuously differ- +entiable, and their Jacobians +�C(uC) := +dq +duC +(uC), +�L(L) := dφ +dL +(L) +are symmetric and positive definite for all uC ∈ +RnC, L ∈ +RnL. +(ii) g : +RnR → +RnR is continuously differentiable, and its Jacobian has the +property that dg +duR (uR)+ dg +duR (uR)⊤ is positive definite for all uR ∈ +RnR. +If q : +RnC → +RnC and φ : +RnL → +RnL fulfill these assumptions, then +there exist twice continuously differentiable and non-negative functions VC : +RnC → +R, VL : +RnL → +R with the following property: the gradients of VC +and VL are, respectively, the inverse functions of q and φ. That is, +∀qC ∈ +RnC : ∇VC(qC) = q−1(qC), +∀φL ∈ +RnL : ∇VL(φL) = φ−1(φL). +With this setting, the PH-DAE MNA network equations can now be derived +as follows: we first eliminate the equation φL − φ(L) = 0: L = φ−1(φL); +secondly, we replace the equation qC − q(A⊤ +Ce) = 0 by A⊤ +Ce − q−1(qC) = 0. +We end up with +d +dt + + +AC 0 0 0 +0 I 0 0 +0 0 0 0 +0 0 0 0 + + +� +�� +� +E := + + +qC +φL +e +V + + +� �� � +x := += + + +0 −AL 0 −AV +A⊤ +L +0 +0 +0 +0 +0 +0 +0 +A⊤ +V +0 +0 +0 + + +� +�� +� +J := + + +e +φ−1(φL) +q−1(qC) +V + + +� +�� +� +z(x) +− + + +ARg(A⊤ +Re) +0 +A⊤ +Ce − q−1(qC) +0 + + +� +�� +� +r(z(x)) := ++ + + +−AI +0 +0 +0 +0 +0 +0 +−I + + +� +�� +� +B := +� +ı(t) +v(t) +� +� �� � +u(t) := +, +(10) + +8 +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +which is a PH-DAE of type (7) with subspace V and Hamiltonian H(x) given +by H(x) = VC(qC) + VL(φL), V = +��e, L, uC, V +�⊤ ∈ +Rn��� A⊤ +Ce = uC +� +. +Remark 2. a) The PHS-DAE formulation shares the index properties of char- +ge/flux-oriented MNA network equations, if the assumption on soundness +and passivity hold: the index is one if, and only if, it neither contains +LI-cutsets nor CV -loops except for C-loops; otherwise it is two. +b) If r subcircuits given as PH-DAE MNA network equations are coupled via +voltage/current sources, the overall system can be written as a PH-DAE +MNA of type (10). +4 Electromagnetic devices +In [5], the Maxwell grid equations for an electromagnetic device have been de- +veloped as a linear PH-DAE system provided that (a) the three-dimensional +domain of the device is connected, bounded and surrounded by perfectly con- +ducting material, (b) the permittivity ǫ, the permeability µ are symmetric +positive definite, and the conductivity σ is symmetric positive semi-definite, +and (c) finite integration technique [6] has been used for the spatial discretiza- +tion with orthogonal staggered cells: +�Mµ 0 +0 +Mǫ +� d +dt +�ˆh +ˆe +� += +�� 0 −C +C⊤ +0 +� +− +�0 +0 +0 Mσ +�� �ˆh +ˆe +� ++ +� 0 +XS +� +ˆu2, +(11a) +ˆy2 = +� +0 +XS +�⊤ �ˆh +ˆe +� += X⊤ +S ˆe. +(11b) +Here C denotes the discrete curl operator, the material matrices Mǫ, Mµ +and Mσ represent the discretized permittivity, permeability and conductivity +distributions, ˆe is vector of the electric mesh voltages e, ˆh the vector of the +magnetic mesh voltages h, and the (dual grid facet) source current ˆu2 as +input. This input is allocated at positions XS. In fact, XS maps the interior +mesh links onto the exterior mesh nodes. Furthermore, the respective electric +mesh voltage ˆy2 forms the output. The Hamiltonian of the electromagnetic +device is given by H1 = 1 +2(˜e⊤Mǫ˜e + ˜h⊤Mµ˜h). +5 Coupled EM/circuit system +When coupling an electromagnetic device with an electric circuit, it remains +only to define the inputs, outputs and the coupling equation. For the circuit, +the electromagnetic device produces the current E flowing into the network, +which is assembled at the respective nodes of the circuit via an incidence + +Port-Hamiltonian Systems Modelling in Electrical Engineering +9 +matrix AE. Hence the circuit part reads (where we split inputs again in +external inputs ı, v, and internal ones): +d +dt + + +AC 0 0 0 +0 I 0 0 +0 0 0 0 +0 0 0 0 +0 0 0 0 + + + + +qC +φL +e +V +E + + += + + +0 −AL 0 −AV −AE +A⊤ +L +0 +0 +0 +0 +0 +0 +0 +0 +0 +A⊤ +V +0 +0 +0 +0 +A⊤ +E +0 +0 +0 +0 + + + + +e +L +uC +V +E + + +(12a) +− + + +ARg(A⊤ +Re) +0 +A⊤ +Ce − uC +0 +0 + + ++ + + +−AI +0 +0 +0 +0 +0 +0 +−I +0 +0 + + +�ı(t) +v(t) +� ++ + + +0 +0 +0 +0 +1 + + +ˆu1, + + +¯y1,1 +¯y1,2 +ˆy1 + + = + + +−AI +0 0 +0 +0 0 +0 +0 0 +0 +−I 0 +0 +0 1 + + +⊤ +· + + +e +L +uC +V +E + + += + + +−A⊤ +I e +−V +E + + +(12b) +with the Hamiltonian: H2 = VC(qC) + VL(φL). +The coupling is as follows [5]: the input ˆu1 (of the electric circuit) is given +by the voltage drop at the electromagnetic device, which reads +ˆu1 = −X⊤ +S ˜e = −ˆy2; on the other hand, the input ˆu2 (of the magnetic +device) is given by the current ˆu2 = E = ˆy1. Overall, we get the following +skew-symmetric relation between inputs and outputs: +0 = +�ˆu1 +ˆu2 +� ++ +� 0 +I +−I 0 +� �ˆy1 +ˆy2 +� +. +(13) +As we have a system consisting of two PH-DAE systems (11) and (12) with +a skew-symmetric linear coupling condition (13), the overall system can be +written as a condensed PH-DAE system (9) with Hamiltonian H = H1 + H2 +and enlarged matrices as above. +6 Simulation Strategies +Generally, for simulating the coupled EM/circuit system numerically, two +approaches are feasible: +a) Monolithic approach. The condensed system (9) can be solved by any inte- +gration scheme suitable for index-1 and index-2 systems, depending on the +index. To preserve the dissipation inequality also on a discrete level, collo- + +10 +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +cation schemes [13] and discrete gradient schemes tracing back to [7] are +the methods-of choice. This strategy is also referred to as strong coupling. +b) Monolithic multirate approach. In fact, we are facing models, where the +subsystems can have widely separated time scales. This can create so- +called multirate potential, where it is beneficial to employ schemes, which +use inherent step sizes for each subsystem. In this way, each subsystem +can be sampled on its time scale. See e.g. [2,10]. +c) Weak coupling. Since the coupling equations is merely the one-to-one iden- +tification of output and input, we can insert this. Furthermore, omitting +outputs due to external sources, we have +d +dt + + +AC 0 0 0 +0 I 0 0 +0 0 0 0 +0 0 0 0 +0 0 0 0 + + + + +qC +φL +e +V +E + + += + + +0 −AL 0 −AV −AE +A⊤ +L +0 +0 +0 +0 +0 +0 +0 +0 +0 +A⊤ +V +0 +0 +0 +0 +A⊤ +E +0 +0 +0 +0 + + + + +e +L +uC +V +E + + +(14a) +− + + +ARg(A⊤ +Re) +0 +A⊤ +Ce − uC +0 +0 + + ++ + + +−AI +0 +0 +0 +0 +0 +0 +−I +0 +0 + + +� +ı(t) +v(t) +� +− + + +0 +0 +0 +0 +1 + + +ˆy2, +ˆy1 = E +(14b) +and +�Mµ 0 +0 +Mǫ +� d +dt +�˜h +˜e +� += +�� 0 −C +C⊤ +0 +� +− +�0 +0 +0 Mσ +�� �˜h +˜e +� ++ +� 0 +XS +� +ˆy1 +(15a) +ˆy2 =XS˜e. +(15b) +Here dynamic iteration schemes [1] are the methods-of choice, as due to +the ODE-DAE coupling no stability constraints occur [4]. In addition, each +step of a Jacobi or Gauß-Seidel iteration scheme defines a PH-DAE system +by its own [8]. +Operator splitting approaches are not generally feasible for differential- +algebraic equations, which can easily be seen for the linear PH-DAE (3a) +with z(x) = x and B = 0. A Lie-Trotter splitting approach may read +d +dt(Ex) = Jx, +x(0) = x0, +d +dt(Ew) = −Rw. +w(0) = x(T ), +allowing for using a symplectic integrator for the first step, and a dissi- +pative scheme for the second one. However, the matrix pencil {E, J} or +{E, R} may be singular and thus not define a unique solution for the re- + +Port-Hamiltonian Systems Modelling in Electrical Engineering +11 +spective subproblem, even if the matrix pencil {E, J − R} of the overall +system is regular. Even if this does not happen, the first problem, for exam- +ple, may not allow for a unique solution for arbitrary choices of consistent +initial values. For +E = diag(1, 0, 1), +J = + + +0 −1 0 +1 0 0 +0 0 0 + + , +R = diag(0, 1, 1), +x0 = + + +1 +−1 +0 + + , +all matrix pencils {E, J − R}, {E, J} and {E, R} are regular, but the first +step yields x1 ≡ 0 ̸= 1. +One may overcome the problem by rewriting the DAE in terms of an +underlying ODE and subsequent algebraic variables given by explicit eval- +uations. For network equations a branch oriented loop-cutset approach +is an option for defining such a PH-DAE system, see [5]. Another way to +avoid the problems above is to follow an operator splitting based approach +for dynamic iteration. In the latter case, no stability problems occur and +a monotone convergence can be obtained [3]. +7 Conclusions +PHS provide a modelling framework which preserves essential physical prop- +erties. It is especially suited for multiphysical applications, as the proper +coupling of port-Hamiltonian subsystems yields an overall PHS. In electrical +engineering, we have shown that electrical networks and electromagnetic de- +vices can be written as PHS, and coupled EM/circuit system yield coupled +PHS with a skew-symmetric coupling, which can be rewritten as an overall +PHS. For simulation, a monolithic approach is suitable for the former, and +weak coupling methods for the latter. There are still many unresolved ques- +tions, such as how to adequately integrate distributed ports into the PHS +modeling. +Acknowledgements Michael G¨unther is indebted to the funding given by the European +Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska- +Curie Grant Agreement No. 765374, ROMSOC. +References +1. Arnold, M., G¨unther, M.: Preconditioned dynamic iteration for coupled differential- +algebraic equations. BIT 41 (2001), 1–25. + +12 +A. Bartel, M. Clemens, M. G¨unther, B. Jacob, and T. Reis +2. A. Bartel, M. G¨unther, Multirate schemes — an answer of numerical analysis to a +demand from applications, in: M. G¨unther, W. Schilders (Eds.), Novel Mathematics +Inspired by Industrial Challenges, Springer, 2022, pp. 5–27. +3. Bartel, B., G¨unther, M., Jaob, B., Reis, T.: Operator Splitting Based Dynamic Itera- +tion for Linear Port-Hamiltonian Systems. Submitted for publication +4. Bartel, A., Brunk, M., G¨unther, M., Sch¨ops, S.: Dynamic Iteration for Coupled Prob- +lems of Electric Circuits and Distributed Devices. SIAM Journal on Scientific Com- +puting 35(2), B315-B335 (2013) year = 2013, doi = 10.1137/120867111, +5. Diab, M.: Splitting Methods for Partial Differential-Algebraic Systems with Appli- +cation on Coupled Field-Circuit DAEs. PhD thesis, Humboldt Universit¨at zu Berlin +(2022) +6. Weiland, T.: Time Domain Electromagnetic Field Computation with Finite Difference +Methods, International Journal on Numerical Modeling: Electric Networks, Devices +and Fields 9, 295-319 (1996). +7. Gonzalez, O.: Time integration and discrete Hamiltonian systems. Journal of Nonlinear +Science 6, 1432-1467 (1996). +8. G¨unther, M., Bartel, A., Jacob, B., Reis, T.: Dynamic iteration schemes and port- +Hamiltonian formulation in coupled DAE circuit simulation. International Journal of +Circuit Theory and Applications 49, 430-452 (2021). +9. G¨unther, M., Feldmann, U.: CAD based electric circuit modeling I: mathematical +structure and index of network equations. Surv. Math. Ind. 8, 97–129 (1999). +10. G¨unther, M., Sandu, A., Numer. Math., 133, 497–524 (2016). +11. Jacob, B., Zwart, H.J.: Linear Port-Hamiltonian Systems on Infinite-dimensional +Spaces. Birkh¨auser Verlag, Basel (2012). +12. D. Jeltsema and A.J. van der Schaft. +Port-Hamiltonian systems theory: An intro- +ductory overview. Foundations and Trends in Systems and Control, 1(2-3):173–387, +2014. +13. Mehrmann, V., Morandin, R.: Structure-preserving discretization for port-Hamiltonian +descriptor systems. 2019 IEEE 58th Conference on Decision and Control (CDC), 6863- +6868 (2019). +14. A.J. van der Schaft. Port-Hamiltonian systems: Network modeling and control of non- +linear physical systems. In K. Schlacher and H. Irschnik, editors, Advanced Dynamics +and Control of Structures and Machines, volume 444 of CISM courses and lectures, +pages 127–167. Springer, Vienna, 2004. +15. van der Schaft, A.: Port-Hamiltonian systems: an introductory survey. In M. Sanz-Sole, +J. Soria, J. L. Varona, J. Verdera (Eds.), Proceedings of the International Congress of +Mathematicians Vol. III (pp. 1339-1365). European Mathematical Society Publishing +House (2006). + diff --git a/VtA0T4oBgHgl3EQfEv_z/content/tmp_files/load_file.txt b/VtA0T4oBgHgl3EQfEv_z/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..f16dbbb857bbb41a1c67eb7ee2bd1838e96ed3b6 --- /dev/null +++ b/VtA0T4oBgHgl3EQfEv_z/content/tmp_files/load_file.txt @@ -0,0 +1,323 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf,len=322 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='02024v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='NA] 5 Jan 2023 Port-Hamiltonian Systems Modelling in Electrical Engineering A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis Abstract The port-Hamiltonian modelling framework allows for models that preserve essential physical properties such as energy conservation or dissipa- tive inequalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' If all subsystems are modelled as port-Hamiltonian systems and the inputs are related to the output in a linear manner, the overall system can be modelled as a port-Hamiltonian system (PHS), too, which preserves the properties of the underlying subsystems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' If the coupling is given by a skew-symmetric matrix, as usual in many applications, the overall system can be easily derived from the subsystems without the need of introducing dummy variables and therefore artificially increasing the complexity of the system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Hence the PHS framework is especially suitable for modelling multi- physical systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In this paper, we show that port-Hamiltonian systems are a natural general- ization of Hamiltonian systems, define coupled port-Hamiltonian systems as ordinary and differential-algebraic equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' To highlight the suitability for electrical engineering applications, we derive PHS models for MNA network equations, electromagnetic devices and coupled systems thereof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Andreas Bartel, Michael G¨unther Bergische Universit¨at Wuppertal, IMACM, Chair of Applied Mathematics, Gaußstraße 20, D-42119 Wuppertal, e-mail: [bartel,guenther]@uni-wuppertal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='de Markus Clemens Bergische Universit¨at Wuppertal, IMACM, Chair of Electromagnetic Theory, Rainer- Gruenter-Straße 21, D-42119 Wuppertal, e-mail: clemens@uni-wuppertal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='de Birgit Jacob Bergische Universit¨at Wuppertal, IMACM, Chair of Functional Analysis, Gaußstraße 20, D-42119 Wuppertal, e-mail: bjacob@uni-wuppertal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='de Timo Reis TU Ilmenau, Fakult¨at f¨ur Mathematik und Naturwissenschaften, Chair of System Theory and PDEs, PF 10 05 65, D-98684 Ilmenau, e-mail: timo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='reis@tu-ilmenau.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='de 1 2 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis 1 Port-Hamiltonian Systems Modelling in a Nutshell Port-Hamiltonian Systems (PHS) are a generalization of Hamiltonian systems ˙x = J · ∇H(x),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x(0) = x0 (1) with x = (p,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' q) consisting of generalized position q(t) ∈ Rn and momentum p(t) ∈ Rn (where t ∈ [0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' T ]),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' the skew-symmetric matrix J given by J = � 0 −I I 0 � and the Hamiltonian H(x) = H(p,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' q) = U(p) + V (q) given as the sum of potential and kinetic energy,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' which maps Rn × Rn → R and is twice contin- uously differentiable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The Hamiltonian flow ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0)), i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', the solution of (1) at time point t, starting at the initial value x(0) = x0, is characterized by four geometric properties: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Preservation of the Hamiltonian: d dtH(ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0)) = (∇H(ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0)))⊤J(∇H(ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0))) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Time-reversibility: ρ ◦ ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0) ◦ ρ ◦ ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0) = x0, with ρ(p, q) = (−p, q), which is a direct consequence of the ρ-reversibility of the Hamiltonian flow: ρ ◦ J∇H(ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0))) = −J∇H(ρ ◦ ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' x0))).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Symplectic structure of the Hamiltonian flow: Ψ(t)⊤J−1Ψ(t) = J−1, Ψ(t) := ∂ϕ(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='x0) ∂x0 , which is a direct consequence of the skew-symmetry of J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Volume-preservation: (det Ψ(t))2 = 1, which follows immediately from the symplectic structure in 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' First generalization step: arbitrary skew-symmetric matrices J If we replace in (1) J by an arbitrary skew-symmetric matrix, the Hamiltonian is still preserved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' As x will loose its characterization as generalized positions and momenta of classical mechanics, time-reversibility will generally not hold anymore.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' However, the symplectic structure of the flow still holds in the case of a regular J, and volume preservation is still a consequence of the Hamiltonian flow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Port-Hamiltonian Systems Modelling in Electrical Engineering 3 Second generalization step: adding dissipation to the system Allowing the flow to become dissipative, we may generalize (1) to the dissi- pative Hamiltonian system ˙x = (J − R) · ∇H(x), x(0) = x0 (2) with R ≥ 0 being symmetric and positive semi-definite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In this case, the flow will neither be symplectic nor volume preserving, and the preservation of the Hamiltonian is replaced by the dissipativity condition d dtH(x(t)) = (∇H(x))⊤ ˙x = −(∇H(x))⊤R∇H(x) ≤ 0 ⇒ H(x(t)) = H(x0) − � t 0 (∇H(x(τ)))⊤R∇H(x(τ)) dτ ≤ H(x0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Third generalization step: coupling to the environment via inputs and outputs Allowing for inputs and outputs to couple the system to the environment, we end up with linear port-Hamiltonian system characterized by ˙x = (J − R) · ∇H(x) + Bu(t), x(0) = x0, y = B⊤∇H(x) with inputs u(t) ∈ Rp, outputs y(t) ∈ Rp and port-matrices B ∈ Rn×p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The dissipativity inequality now reads d dtH(x(t)) = (∇H(x))⊤ ˙x = −(∇H(x))⊤R∇H(x) + (∇H(x))⊤Bu(t)) = −(∇H(x))⊤R∇H(x) + y(t)⊤u(t) ≤ y(t)⊤u(t) ⇒ H(x(t)) = H(x0) − � t 0 (∇H(x(τ)))⊤R∇H(x(τ)) dτ + � t 0 y(τ)⊤u(τ)dτ ≤ H(x0) + � t 0 y(τ)⊤u(τ)dτ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Fourth generalization step: PH-DAE systems Linear PHS can be easily generalized to PH-DAE systems given by d dt(Ex) = (J − R) · z(x) + Bu(t), x(0) = x0, (3a) y = B⊤z(x) (3b) with a possibly singular matrix E ∈ Rn×n and the nonlinear mapping z : Rn → Rn fulfilling the compatibility condition E⊤z = ∇H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Now the 4 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis dissipativity condition reads H(x(t)) = H(x0) − � t 0 z(x(τ))⊤R∇z(x(τ)) dτ + � t 0 y(τ)⊤u(τ) dτ ≤ H(x0) + � t 0 y(τ)⊤u(τ) dτ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The key point in port-Hamiltonian modelling is the following: there is an easy way to couple arbitrary many PH-DAE system such that the overall system is still a PH-DAE system, which preserves a dissipativity inequality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Let us consider r autonomous PH-DAE systems d dt(Eixi) = (Ji − Ri)zi(xi) + Biui, (4a) yi = B⊤ i zi(xi) (4b) with r Hamiltonians H1, H2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' , Hr and compatibility conditions E⊤ i zi = ∇Hi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' If the inputs and outputs fulfill a linear interconnection relation Mu + Ny = 0 for the aggregated input u = (u1, u2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' , ur) and output y = (y1, y2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' , yr), it has been shown in [13] that one can write the aggre- gated system as a joint PH-DAE system as d dt \uf8eb \uf8ec \uf8ec \uf8ed \uf8ee \uf8ef\uf8ef\uf8f0 E 0 0 0 0 0 0 0 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fb \uf8ee \uf8f0 x ˆu ˆy \uf8f9 \uf8fb \uf8f6 \uf8f7 \uf8f7 \uf8f8 = \uf8ee \uf8ef\uf8ef\uf8f0 J − R B 0 0 −B⊤ 0 Im −M ⊤ 0 −Im 0 −N ⊤ 0 M N 0 \uf8f9 \uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8f0 z(x) ˆu ˆy 0 \uf8f9 \uf8fa\uf8fa\uf8fb + \uf8ee \uf8ef\uf8ef\uf8f0 0 0 Im 0 \uf8f9 \uf8fa\uf8fa\uf8fb u, y = ˆy, with z(x)⊤ = (z1(x1)⊤, z2(x2)⊤, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' , zr(xr)⊤), new dummy variables ˆu, ˆy and setting X = diag(X1, X2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' , Xr) for X ∈ {E, J, R, B}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' This coupling property of PH-DAE systems makes the port-Hamiltonian modelling frame- work well suited for multiphysical applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Now, we consider external, time dependent inputs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' To this end, we split the inputs and outputs into external (bar-notation) and internal ones (hat- notation), i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Biui is split into ¯Bi¯ui + ˆBiˆui.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Then, the subsystem (4) reads d dt(Eixi) = (Ji − Ri)zi(xi) + ˆBiˆui + ¯Bi¯ui, (5a) ˆyi = ˆB⊤ i zi(xi), (5b) ¯yi = ¯B⊤ i zi(xi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (5c) For the coupling relation (of the internal quantities) ˆu + Cˆy = 0 with a skew-symmetric matrix C = −C⊤ (which often arises in application), these systems can be written as a joint PH-DAE system in condensed form [8]: Port-Hamiltonian Systems Modelling in Electrical Engineering 5 d dt(Ex) = ( ˜J − R)z(x) + ¯B¯u, (6a) ¯y = ¯B⊤z(x) (6b) with the condensed skew-symmetric matrix ˜J = J − ˆBC ˆB⊤.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Note that in this case all internal coupling modelled via the port-matrices ˆBi has now been transferred into the off-block diagonal elements of the skew-symmetric matrix ˜J, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', − ˆBC ˆB⊤.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' A systems theoretic treatment of port-Hamiltonian systems goes back to Bernhard Maschke and Arjan van der Schaft (see [12, 14] for an overview), where nonlinear systems governed by ordinary differential equa- tions are treated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For simplicity of presentation, we will (a) not follow the differential geometric path via Dirac structures, (b) neglect a feed-through from input to output and (c) only consider finite dimensional systems, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', ordinary (ODEs) and differential-algebraic equations (DAEs), but no partial differential equations (PDEs).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For simulation, the latter are usually trans- formed into ODEs and DAEs by spatial semi-discretization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For a differential geometric setting of PHS see [15] and an introduction into PH-PDEs see [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The paper is organized as follows: In the next section we introduce PH- DAE systems which allow for a general nonlinear dissipative part.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' A PHS- DAE formulation of the MNA network equations is derived in Sect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 3, and for electromagnetic devices in Sect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Section 5 discusses PHS formulation of coupled EM/circuit systems, which allow for monotolithic as well as weak coupling simulation approaches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Sect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 6 finishes with conclusions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 2 PH-DAE systems When dealing with applications in electrical engineering, the concept of port- Hamiltonian modelling has to be generalized to coupled differential-algebraic equations, which (a) allow for a general nonlinear resistive part r(z) instead of a quasilinear setting Rz as in the approach of [13] and (b) has only to be accretive on a subspace V ⊂ R according to the constraints of the system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' A differential-algebraic equation of the form d dtEx(t) = Jz(x(t)) − r(z(x(t))) + Bu(t), y(t) = B⊤z(x(t)) (7) is called a port-Hamiltonian differential-algebraic equation (PH-DAE) [8], if the following holds: E ∈ Rn×n, J ∈ Rn×n and B ∈ Rn×m, z, r : Rn → Rn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' There exists a subspace V ⊂ Rn with the following properties: 6 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis (i) for all intervals I ⊂ R and functions u : I → Rm such that (7) has a solution x : I → Rn, it holds z(x(t)) ∈ V for all t ∈ I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (ii) J is skew-symmetric on V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' That is, v⊤Jw = −w⊤Jv for all v, w ∈ V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (iii) r is accretive on V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' That is, v⊤r(v) ≥ 0 for all v ∈ V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' There exists some function H ∈ C1(Rn, R) such that ∇H(x) = E⊤z(x) for all x ∈ z−1(V).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Remark 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' a) The PH-DAE (7) system provides the usual energy balance d dtH(x(t)) = −z(x(t))⊤r(z(x(t)))) + y(t)⊤u(t) ≤ y(t)⊤u(t).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' b) PH-DAE subsystems now read d dtEixi(t) =Jizi(xi(t)) − ri � zi(xi(t)) � + Biui(t), (8a) yi(t) =B⊤ i zi � xi(t) � (8b) instead of (4), and if they are coupled by a skew-symmetric coupling rela- tion ˆu + Cˆy = 0 with a skew-symmetric matrix C = −C⊤ as before, they can be condensed into an overall PH-DAE system d dtEx = ˆJz − r + ¯B¯u, (9a) ¯y = ¯B⊤z (9b) with the skew-symmetric matrix ˆJ again given by ˆJ = J − ˆB ˆC ˆB⊤.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 3 Electrical networks We consider the classical charge-/flux oriented MNA network equations [8,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='9] d dt \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 0 0 0 AC 0 0 0 0 0 I 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 e \uf6beL \uf6beV qC φL \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 0 −AL −AV 0 0 A⊤ L 0 0 0 0 A⊤ V 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 e \uf6beL \uf6beV qC φL \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb − \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 ARg(A⊤ Re) 0 0 qC − q(A⊤ Ce) φL − φ(\uf6beL) \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb + \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 −AI 0 0 0 0 −I 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb �ı(t) v(t) � Port-Hamiltonian Systems Modelling in Electrical Engineering 7 with e,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' \uf6beL and \uf6beV denoting node potentials and currents through flux storing elements and voltages sources,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' qC and Φl charge and flux-storing elements,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' i(t) and v(t) independent current and voltage sources,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' the resistive currents g and the incidence matrices AC,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' AL,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' AR,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' AV ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' AI for charge- and flux storing elements,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' resistive elements,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' voltage and current sources,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' and seek a formu- lation as a PH-DAE system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For this, we need the following assumptions, which naturally occur in circuit simulation, see [8]: (a) Soundness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The circuit graph has at least one branch and is connected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Furthermore, it contains neither V -loops nor I-cutsets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Equivalently, AV and (AC AR AL AV )⊤ have full column rank.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (b) Passivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The functions q, φ and g fulfill (i) q : RnC → RnC and φ : RnL → RnL are bijective, continuously differ- entiable, and their Jacobians �C(uC) := dq duC (uC), �L(\uf6beL) := dφ d\uf6beL (\uf6beL) are symmetric and positive definite for all uC ∈ RnC, \uf6beL ∈ RnL.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (ii) g : RnR → RnR is continuously differentiable, and its Jacobian has the property that dg duR (uR)+ dg duR (uR)⊤ is positive definite for all uR ∈ RnR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' If q : RnC → RnC and φ : RnL → RnL fulfill these assumptions, then there exist twice continuously differentiable and non-negative functions VC : RnC → R, VL : RnL → R with the following property: the gradients of VC and VL are, respectively, the inverse functions of q and φ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' That is, ∀qC ∈ RnC : ∇VC(qC) = q−1(qC), ∀φL ∈ RnL : ∇VL(φL) = φ−1(φL).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' With this setting, the PH-DAE MNA network equations can now be derived as follows: we first eliminate the equation φL − φ(\uf6beL) = 0: \uf6beL = φ−1(φL);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' secondly, we replace the equation qC − q(A⊤ Ce) = 0 by A⊤ Ce − q−1(qC) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' We end up with d dt \uf8ee \uf8ef\uf8ef\uf8f0 AC 0 0 0 0 I 0 0 0 0 0 0 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fb � �� � E := \uf8ee \uf8ef\uf8ef\uf8f0 qC φL e \uf6beV \uf8f9 \uf8fa\uf8fa\uf8fb � �� � x := = \uf8ee \uf8ef\uf8ef\uf8f0 0 −AL 0 −AV A⊤ L 0 0 0 0 0 0 0 A⊤ V 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fb � �� � J := \uf8ee \uf8ef\uf8ef\uf8f0 e φ−1(φL) q−1(qC) \uf6beV \uf8f9 \uf8fa\uf8fa\uf8fb � �� � z(x) − \uf8ee \uf8ef\uf8ef\uf8f0 ARg(A⊤ Re) 0 A⊤ Ce − q−1(qC) 0 \uf8f9 \uf8fa\uf8fa\uf8fb � �� � r(z(x)) := + \uf8ee \uf8ef\uf8ef\uf8f0 −AI 0 0 0 0 0 0 −I \uf8f9 \uf8fa\uf8fa\uf8fb � �� � B := � ı(t) v(t) � � �� � u(t) := , (10) 8 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis which is a PH-DAE of type (7) with subspace V and Hamiltonian H(x) given by H(x) = VC(qC) + VL(φL), V = ��e, \uf6beL, uC, \uf6beV �⊤ ∈ Rn��� A⊤ Ce = uC � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Remark 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' a) The PHS-DAE formulation shares the index properties of char- ge/flux-oriented MNA network equations, if the assumption on soundness and passivity hold: the index is one if, and only if, it neither contains LI-cutsets nor CV -loops except for C-loops;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' otherwise it is two.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' b) If r subcircuits given as PH-DAE MNA network equations are coupled via voltage/current sources, the overall system can be written as a PH-DAE MNA of type (10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 4 Electromagnetic devices In [5],' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' the Maxwell grid equations for an electromagnetic device have been de- veloped as a linear PH-DAE system provided that (a) the three-dimensional domain of the device is connected,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' bounded and surrounded by perfectly con- ducting material,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (b) the permittivity ǫ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' the permeability µ are symmetric positive definite,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' and the conductivity σ is symmetric positive semi-definite,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' and (c) finite integration technique [6] has been used for the spatial discretiza- tion with orthogonal staggered cells: �Mµ 0 0 Mǫ � d dt �ˆh ˆe � = �� 0 −C C⊤ 0 � − �0 0 0 Mσ �� �ˆh ˆe � + � 0 XS � ˆu2,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (11a) ˆy2 = � 0 XS �⊤ �ˆh ˆe � = X⊤ S ˆe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (11b) Here C denotes the discrete curl operator, the material matrices Mǫ, Mµ and Mσ represent the discretized permittivity, permeability and conductivity distributions, ˆe is vector of the electric mesh voltages e, ˆh the vector of the magnetic mesh voltages h, and the (dual grid facet) source current ˆu2 as input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' This input is allocated at positions XS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In fact, XS maps the interior mesh links onto the exterior mesh nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Furthermore, the respective electric mesh voltage ˆy2 forms the output.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The Hamiltonian of the electromagnetic device is given by H1 = 1 2(˜e⊤Mǫ˜e + ˜h⊤Mµ˜h).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 5 Coupled EM/circuit system When coupling an electromagnetic device with an electric circuit, it remains only to define the inputs, outputs and the coupling equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For the circuit, the electromagnetic device produces the current \uf6beE flowing into the network, which is assembled at the respective nodes of the circuit via an incidence Port-Hamiltonian Systems Modelling in Electrical Engineering 9 matrix AE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Hence the circuit part reads (where we split inputs again in external inputs ı,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' v,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' and internal ones): d dt \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 AC 0 0 0 0 I 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 qC φL e \uf6beV \uf6beE \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 0 −AL 0 −AV −AE A⊤ L 0 0 0 0 0 0 0 0 0 A⊤ V 0 0 0 0 A⊤ E 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 e \uf6beL uC \uf6beV \uf6beE \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb (12a) − \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 ARg(A⊤ Re) 0 A⊤ Ce − uC 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb + \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 −AI 0 0 0 0 0 0 −I 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb �ı(t) v(t) � + \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 0 0 0 0 1 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb ˆu1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' \uf8ee \uf8f0 ¯y1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='1 ¯y1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='2 ˆy1 \uf8f9 \uf8fb = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 −AI 0 0 0 0 0 0 0 0 0 −I 0 0 0 1 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb ⊤ \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 e \uf6beL uC \uf6beV \uf6beE \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb = \uf8ee \uf8f0 −A⊤ I e −\uf6beV \uf6beE \uf8f9 \uf8fb (12b) with the Hamiltonian: H2 = VC(qC) + VL(φL).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The coupling is as follows [5]: the input ˆu1 (of the electric circuit) is given by the voltage drop at the electromagnetic device, which reads ˆu1 = −X⊤ S ˜e = −ˆy2;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' on the other hand, the input ˆu2 (of the magnetic device) is given by the current ˆu2 = \uf6beE = ˆy1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Overall, we get the following skew-symmetric relation between inputs and outputs: 0 = �ˆu1 ˆu2 � + � 0 I −I 0 � �ˆy1 ˆy2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (13) As we have a system consisting of two PH-DAE systems (11) and (12) with a skew-symmetric linear coupling condition (13), the overall system can be written as a condensed PH-DAE system (9) with Hamiltonian H = H1 + H2 and enlarged matrices as above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 6 Simulation Strategies Generally, for simulating the coupled EM/circuit system numerically, two approaches are feasible: a) Monolithic approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' The condensed system (9) can be solved by any inte- gration scheme suitable for index-1 and index-2 systems, depending on the index.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' To preserve the dissipation inequality also on a discrete level, collo- 10 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis cation schemes [13] and discrete gradient schemes tracing back to [7] are the methods-of choice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' This strategy is also referred to as strong coupling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' b) Monolithic multirate approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In fact, we are facing models, where the subsystems can have widely separated time scales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' This can create so- called multirate potential, where it is beneficial to employ schemes, which use inherent step sizes for each subsystem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In this way, each subsystem can be sampled on its time scale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' See e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' [2,10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' c) Weak coupling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Since the coupling equations is merely the one-to-one iden- tification of output and input, we can insert this.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Furthermore,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' omitting outputs due to external sources,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' we have d dt \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 AC 0 0 0 0 I 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 qC φL e \uf6beV \uf6beE \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 0 −AL 0 −AV −AE A⊤ L 0 0 0 0 0 0 0 0 0 A⊤ V 0 0 0 0 A⊤ E 0 0 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 e \uf6beL uC \uf6beV \uf6beE \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb (14a) − \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 ARg(A⊤ Re) 0 A⊤ Ce − uC 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb + \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 −AI 0 0 0 0 0 0 −I 0 0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb � ı(t) v(t) � − \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 0 0 0 0 1 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fb ˆy2,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' ˆy1 = \uf6beE (14b) and �Mµ 0 0 Mǫ � d dt �˜h ˜e � = �� 0 −C C⊤ 0 � − �0 0 0 Mσ �� �˜h ˜e � + � 0 XS � ˆy1 (15a) ˆy2 =XS˜e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' (15b) Here dynamic iteration schemes [1] are the methods-of choice, as due to the ODE-DAE coupling no stability constraints occur [4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In addition, each step of a Jacobi or Gauß-Seidel iteration scheme defines a PH-DAE system by its own [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Operator splitting approaches are not generally feasible for differential- algebraic equations, which can easily be seen for the linear PH-DAE (3a) with z(x) = x and B = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' A Lie-Trotter splitting approach may read d dt(Ex) = Jx, x(0) = x0, d dt(Ew) = −Rw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' w(0) = x(T ), allowing for using a symplectic integrator for the first step, and a dissi- pative scheme for the second one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' However, the matrix pencil {E, J} or {E, R} may be singular and thus not define a unique solution for the re- Port-Hamiltonian Systems Modelling in Electrical Engineering 11 spective subproblem, even if the matrix pencil {E, J − R} of the overall system is regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Even if this does not happen, the first problem, for exam- ple, may not allow for a unique solution for arbitrary choices of consistent initial values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For E = diag(1, 0, 1), J = \uf8ee \uf8f0 0 −1 0 1 0 0 0 0 0 \uf8f9 \uf8fb , R = diag(0, 1, 1), x0 = \uf8ee \uf8f0 1 −1 0 \uf8f9 \uf8fb , all matrix pencils {E, J − R}, {E, J} and {E, R} are regular, but the first step yields x1 ≡ 0 ̸= 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' One may overcome the problem by rewriting the DAE in terms of an underlying ODE and subsequent algebraic variables given by explicit eval- uations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For network equations a branch oriented loop-cutset approach is an option for defining such a PH-DAE system, see [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Another way to avoid the problems above is to follow an operator splitting based approach for dynamic iteration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In the latter case, no stability problems occur and a monotone convergence can be obtained [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 7 Conclusions PHS provide a modelling framework which preserves essential physical prop- erties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' It is especially suited for multiphysical applications, as the proper coupling of port-Hamiltonian subsystems yields an overall PHS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In electrical engineering, we have shown that electrical networks and electromagnetic de- vices can be written as PHS, and coupled EM/circuit system yield coupled PHS with a skew-symmetric coupling, which can be rewritten as an overall PHS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' For simulation, a monolithic approach is suitable for the former, and weak coupling methods for the latter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' There are still many unresolved ques- tions, such as how to adequately integrate distributed ports into the PHS modeling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Acknowledgements Michael G¨unther is indebted to the funding given by the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska- Curie Grant Agreement No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 765374, ROMSOC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' References 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Arnold, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', G¨unther, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Preconditioned dynamic iteration for coupled differential- algebraic equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' BIT 41 (2001), 1–25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 12 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Clemens, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Reis 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, Multirate schemes — an answer of numerical analysis to a demand from applications, in: M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Schilders (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' ), Novel Mathematics Inspired by Industrial Challenges, Springer, 2022, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 5–27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', G¨unther, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Jaob, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Reis, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Operator Splitting Based Dynamic Itera- tion for Linear Port-Hamiltonian Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Submitted for publication 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Bartel, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Brunk, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', G¨unther, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Sch¨ops, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Dynamic Iteration for Coupled Prob- lems of Electric Circuits and Distributed Devices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' SIAM Journal on Scientific Com- puting 35(2), B315-B335 (2013) year = 2013, doi = 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='1137/120867111, 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Diab, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Splitting Methods for Partial Differential-Algebraic Systems with Appli- cation on Coupled Field-Circuit DAEs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' PhD thesis, Humboldt Universit¨at zu Berlin (2022) 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Weiland, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Time Domain Electromagnetic Field Computation with Finite Difference Methods, International Journal on Numerical Modeling: Electric Networks, Devices and Fields 9, 295-319 (1996).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Gonzalez, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Time integration and discrete Hamiltonian systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Journal of Nonlinear Science 6, 1432-1467 (1996).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Bartel, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Jacob, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Reis, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Dynamic iteration schemes and port- Hamiltonian formulation in coupled DAE circuit simulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' International Journal of Circuit Theory and Applications 49, 430-452 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Feldmann, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': CAD based electric circuit modeling I: mathematical structure and index of network equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Surv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Ind.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 8, 97–129 (1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' G¨unther, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Sandu, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Numer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', 133, 497–524 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jacob, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Zwart, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' : Linear Port-Hamiltonian Systems on Infinite-dimensional Spaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Birkh¨auser Verlag, Basel (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Jeltsema and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' van der Schaft.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Port-Hamiltonian systems theory: An intro- ductory overview.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Foundations and Trends in Systems and Control, 1(2-3):173–387, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Mehrmann, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=', Morandin, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Structure-preserving discretization for port-Hamiltonian descriptor systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 2019 IEEE 58th Conference on Decision and Control (CDC), 6863- 6868 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' van der Schaft.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Port-Hamiltonian systems: Network modeling and control of non- linear physical systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Schlacher and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Irschnik, editors, Advanced Dynamics and Control of Structures and Machines, volume 444 of CISM courses and lectures, pages 127–167.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Springer, Vienna, 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' van der Schaft, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=': Port-Hamiltonian systems: an introductory survey.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' In M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Sanz-Sole, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Soria, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Varona, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' Verdera (Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' ), Proceedings of the International Congress of Mathematicians Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' III (pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' 1339-1365).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} +page_content=' European Mathematical Society Publishing House (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtA0T4oBgHgl3EQfEv_z/content/2301.02024v1.pdf'} diff --git a/VtE3T4oBgHgl3EQfbAqu/content/tmp_files/2301.04512v1.pdf.txt b/VtE3T4oBgHgl3EQfbAqu/content/tmp_files/2301.04512v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..23e1e1d4e3ed8a494e8095c7dde520bef6b10537 --- /dev/null +++ b/VtE3T4oBgHgl3EQfbAqu/content/tmp_files/2301.04512v1.pdf.txt @@ -0,0 +1,4388 @@ +Partial Conditioning for Inferential Models +Jiasen Yang, Xiao Wang, Chuanhai Liu +Department of Statistics +Purdue University +Abstract +Inferential models have been proposed for valid and efficient prior-free probabilistic inference. +As it gradually gained popularity, this theory is subject to further developments for practically +challenging problems. This paper considers the many-normal-means problem with the means +constrained to be in the neighborhood of each other. A new method, called partial conditioning, +is proposed to generate valid and efficient marginal inference about the individual means. It +is shown that the method outperforms both a fiducial-counterpart in terms of validity and a +conservative-counterpart in terms of efficiency. +Keywords— Fiducial argument, H¨older space, Lipschitz space, Many-normal-means +1 +Introduction +The framework of inferential models (IMs) has been developed to provide valid and efficient prior-free +probabilistic inference. In its simplest form (Martin and Liu, 2013), it is obtained by modifying R. A. Fisher’s +fiducial argument, which suffers from the mathematical difficulties discussed, for example, in Liu and Martin +(2015). It has also been extended for efficient inference by combining information (Martin and Liu, 2015a) +and marginalization (Martin and Liu, 2015b). For certain practically challenging problems, however, this +theory is subject to further developments, similar to other existing schools of thought. One noticeable class +of such challenging problems is inference on constrained parameters, which is challenging for all existing +inferential methods (see, e.g., Leaf and Liu (2012) and references therein). +In this paper, we consider the many-normal-means problem with the means constrained to be in the +neighborhood of each other. For both conceptual and representational simplicity, we describe the problem as +a special case of the familiar classic formulation of a nonparametric regression model (Lepskii, 1991). More +precisely, to motivate the problem, we consider the following model +Yi = ϑ0(ti) + σUi, +i = 1, · · · , n, +(1.1) +where the ti denotes the design points, the Yi denotes the responses, and Ui +i.i.d. +∼ N(0, 1) denote the random +error terms. Here, ϑ0 is the unknown regression function that is assumed to reside in the H¨older space +ΘM,γ = +� +ϑ : [0, 1] → R +��� |ϑ(t) − ϑ(s)| ≤ M|t − s|γ, ∀t, s ∈ [0, 1] +� +, +(1.2) +with the H¨older exponent 0 < γ ≤ 1. We assume M, γ, and σ to be known constants and, without loss of +generality, we set σ = 1. In subsequent discussions, we suppress the subscripts and denote the H¨older space +ΘM,γ simply by Θ, unless otherwise noted. +The objective is to perform statistical inference on the unknown function ϑ0 within the IM framework. +We propose a new method, called partial conditioning, to generate valid and efficient marginal inference +1 +arXiv:2301.04512v1 [math.ST] 11 Jan 2023 + +about the individual means. We show that the method outperforms both a fiducial-counterpart in term of +validity and a conservative-counterpart in terms of efficiency. +It should be noted that IM approaches to the general nonparametric problem itself deserve in-depth +investigations. This explains why our focus here is on a simple case. Our goal is to introduce an innovative +idea of conditioning in the IM framework to combine information when multiple parameters entangle with +each other via known constraints. +The rest of the paper is arranged as follows. Section 2 reviews basic IMs (Martin and Liu, 2013) and +conditional IMs (Martin and Liu, 2015a) with the simple cases of n = 1 and n = 2 with M = 0 in (1.1) +and (1.2). Section 3 addresses the difficulties in the case of n = 2 and M ̸= 0 and introduces the partial +conditioning method, with the general n case considered in Section 4. Finally, Section 5 concludes with a +few remarks. +2 +The IMs: an overview +2.1 +The basic IMs +To illustrate the basic IM framework under the current problem setting, let us first consider the simple case +of n = 1, where we only make use of a single pair of observations (t1, y1). Thus, we have +y1 = ϑ0(t1) + u⋆ +1, +(2.1) +where u⋆ +1 represents an unobserved realization of U1 ∼ N(0, 1). Using the terminology of IMs, (2.1) describes +an underlying sampling model which involves the unknown parameter (function) ϑ ∈ Θ and generates +observed data X1 = (t1, Y1) using auxiliary variable U1 ∼ N(0, 1). +Note that the auxiliary variable is +unobserved but predictable, since its distribution is fully specified. +In general, constructing an IM consists of three steps, namely, an association (A) step, a prediction (P) +step, and a combination (C) step. These three steps are explained below in more detail in the context of the +one-point nonparametric regression problem (2.1): +A-step. The association step can be achieved via some function or procedure F as +X1 = F(U1, ϑ), +(X ∈ X, ϑ ∈ Θ, U ∼ N(0, 1)). +This association allows for direct reasoning with the source of uncertainty U1, which is missing. If +U1 = u⋆ +1 were observed, we would then be able to obtain the best possible inference for ϑ, which is +given by the set-valued “inverse” mapping +G : u⋆ +1 → ΘX1(u⋆ +1) = {ϑ ∈ Θ : ϑ(t1) = Y1 − u⋆ +1}. +P-step. For inference on ϑ, the discussion in the A-step suggests that we should focus our attention on +accurately predicting the unobserved quantity u⋆ +1. To predict u⋆ +1 with a certain desired accuracy, we +utilize a predictive random set S, for example, +S(U1) = {˜u ∈ R : −|U1| ≤ ˜u ≤ |U1|}, +(U1 ∼ N(0, 1)). +(2.2) +C-step. To transfer the available information about u⋆ to the ϑ-space, the last step is to combine the +information in the association, the observed x1 = (t1, y1), and the predictive random set S. For this +purpose, consider the expanded set +Θx1(S) = +� +u1∈S +Θx1(u1) = {ϑ ∈ Θ : |ϑ(t1) − y1| ≤ |U1|}, +(U1 ∼ N(0, 1)) +which contains those values of ϑ that are consistent with the observed data and the sampling model +for at least one candidate u⋆ +1 value u1 ∈ S. +2 + +The random sets obtained in the C-step are in the space of the unknown parameter, and we are ready to +produce uncertainty assessment for assertions of interest. Consider an assertion A about the parameter of +interest ϑ. The assertion A corresponds to a set A ⊆ Θ, and acts as a hypothesis in the context of classical +statistics. To summarize the evidence in x that supports the assertion A, we evaluate the belief function +defined by +belx(A; S) = PS{Θx(S) ⊆ A | Θx(S) ̸= ∅}, +(2.3) +and the plausibility function defined by +plx(A; S) = 1 − belx(Ac, S) = PS{Θx(S) ∩ A ̸= ∅ | Θx(S) ̸= ∅}. +(2.4) +The belief function is subadditive in the sense that if ∅ ̸= A ⊆ Θ, then bel(A; S)+bel(Ac; S) ≤ 1 with equality +if and only if Θx(S) is a singleton with PS-probability 1. Therefore, it follows that belx(A; S) ≤ plx(A; S), +and they are also referred to as the lower and upper probabilities, respectively. Incidentally, we note that +there are continuing interests in using lower and upper probabilities for statistical inference (Gong and Meng, +2021; Liu and Martin, 2021). +To establish the frequency-calibration properties of the IM in the context of our current problem, we +briefly review several results regarding validity as introduced by Martin and Liu (2013). +Definition 2.1. A predictive random set S is valid for predicting the unobserved auxiliary variable U if for +each α ∈ (0, 1), +PU{PS{S ̸∋ U} ≥ 1 − α} ≤ α. +(2.5) +Definition 2.2. The IM is valid if, for all assertions A and for any α ∈ (0, 1), +sup +ϑ∈A +PX|ϑ{plX(A; S) ≤ α} ≤ α. +(2.6) +Theorem 2.1. Suppose that the predictive random set S is valid, and Θx(S) ̸= ∅ with PS-probability 1 for +all x. Then the IM is valid. +In other words, a valid inference in the sense of frequency calibration is obtained as long as predictive +random sets are valid to predict unobserved auxiliary variables. Formally, for the one-point case, we have +the following result, followed by a numerical example to illustrate plausibility-based confidence intervals. +Proposition 2.1. The predictive random set defined by (2.2) is valid and, therefore, the IM with the belief +function given by (2.3) and the plausibility function given by (2.4) is valid. +Example 2.1 (An numerical illustration). Consider the assertion Aθ0 = {ϑ0 : ϑ0(t1) = θ0} ⊆ Θ1,1/2 and +the predictive random set S = [−|U1|, |U1|], U1 ∼ N(0, 1). Since this assertion constitutes a singleton in the +parameter space, the belief function bely1(Aθ0; S) = 0, but the plausibility function +ply1(Aθ0; S) = PS{θ0 ∈ Θx1(S)} = PU1{|U1| ≥ |θ0 − y1|} = 2(1 − Φ(|θ0 − y1|)). +(2.7) +Figure 1 shows the graphs of ply1(Aθ0; S) as a function of θ0 for y1 = 0. +Remark 2.1 (Plausibility intervals). In Example 2.1, the assertion A = {ϑ0 = θ0}, ϑ0 ∈ Θ constitutes a +singleton in the parameter space. In this case, we can define the 100(1 − α)% plausibility region given by +Πx(α) = {ϑ : plx(ϑ) ≥ α}, +(2.8) +where plx(ϑ) := plx({ϑ}; S). It was shown in Martin and Liu (2013) that the plausibility region (2.8) provides +an exact 100(1 − α)% confidence interval. +3 + +−3 +−2 +−1 +0 +1 +2 +3 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +θ +plausibility +Figure 1: The plausibility function ply1(Aθ0; S) defined in (2.7) as a function of θ0 with y1 = 0. +2.2 +The conditional IMs +As seen in the previous section for the simple one-point case, the basic IMs framework is similar to the +frequentist pivotal method of constructing confidence intervals. However, the way to make inference by +predicting unobserved auxiliary variables can help develop efficient inference. Two such methods, called +conditional IMs and marginal IMs, are first explained here for efficient inference in the special cases of the +two-point problem. A new method of taking the strengths of conditional IMs and marginal IMs is proposed +for the general case of the two-point problem in Section 3. +In the n = 2 case, we have two pairs of observations (t1, y1) and (t2, y2). The sampling model for this +case can be written as +� +y1 = ϑ0(t1) + u⋆ +1; +y2 = ϑ0(t2) + u⋆ +2, +(2.9) +where u⋆ +1 and u⋆ +2 represent unobserved realizations of U1, U2 +iid +∼ N(0, 1). The IM framework for the current +problem repeats the three steps introduced in the previous section. However, we notice that in this two- +point case, we have two auxiliary variables U1, U2, which would need to be predicted using a two-dimensional +predictive random set. As Martin and Liu (2015a) pointed out, a more efficient inference procedure can be +obtained by reducing the dimension of the auxiliary variable. In particular, if some functions of the original +auxiliary variable are fully observed, we can condition on the fully observed information to sharpen our +prediction of the unobserved. +Notice that from (2.9) we can obtain +y2 − y1 = ϑ0(t2) − ϑ0(t1) + u⋆ +2 − u⋆ +1, +(2.10) +which motivates us to introduce a new auxiliary variable V = U2 − U1 ∼ N(0, 2). By the H¨older condition +(1.2), we have +|ϑ0(t1) − ϑ0(t2)| ≤ M|t1 − t2|γ. +(2.11) +Letting B := M|t1 − t2|γ, from (2.10) and (2.11) we obtain +v⋆ = u⋆ +2 − u⋆ +1 = y2 − y1 − (ϑ0(t2) − ϑ0(t1)) ∈ [y2 − y1 − B, y2 − y1 + B]. +(2.12) +For each fixed v ∈ [y2 − y1 − B, y2 − y1 + B], the conditional distributions of a linear function of U1 and U2, +for example, U2, given V = v can be derived as follows: +(U2, V ) ∼ N2 +�� 0 +0 +� +, +� 1 +1 +1 +2 +�� +=⇒ +U2|V = v ∼ N +�v +2, 1 +2 +� +. +(2.13) +4 + +Thus, these conditional distributions, sharper than the corresponding marginal distributions, motivate us to +construct more efficient predictive random sets for more efficient inference about (ϑ0(t1), ϑ0(t2)). +For the special case of B = 0, the equation (2.10) simplifies to +y2 − y1 = u⋆ +2 − u⋆ +1. +Most important, u⋆ +2 − u⋆ +1 in this extreme case is fully observed and, therefore, can be easily used to predict +u⋆ +1 and u⋆ +2. Formally, we can construct a conditional inferential model (CIM) as proposed by Martin and Liu +(2015a). Similar to basic IMs, conditional IMs also have their three steps: +A-step. Under the original IM framework, the association step is achieved via the baseline association (2.9). +In the case of B = 0, this baseline association can be decomposed as +� +y2 − y1 = u⋆ +2 − u⋆ +1; +y2 = ϑ(t2) + u⋆ +2. +(2.14) +This decomposition immediately suggests an alternative association. Let PU2|y2−y1 denote the condi- +tional distribution of U2, given U2 − U1 = y2 − y1. Since y2 − y1 does not provide information on the +parameter ϑ, we can establish a new association +Y2 = ϑ(t2) + ˜U2, +( ˜U2 ∼ PU2|y2−y1), +(2.15) +which is referred to as the conditional association. Using (2.15), we can associate the observed infor- +mation Y2 and the parameter ϑ with the new auxiliary variable ˜U2 ∼ PU2|Y2−Y1 to get the collection +of sets +Θy2(˜u⋆ +2) = {ϑ ∈ Θ : y2 = ϑ(t2) + ˜u⋆ +2}. +P-step. Fixing the observed value Y2 − Y1 = y2 − y1, we predict the unobserved value ˜u⋆ +2 of ˜U2 with a +conditionally admissible predictive random set S2 ∼ PS2|y2−y1. Notice that equation (2.13) implies +that ˜U2 ∼ N +� y2−y1 +2 +, 1 +2 +� +, which gives rise to the default predictive random set given by +S2 = +� +˜u2 : +����˜u2 − y2 − y1 +2 +���� ≤ +���� ˜U2 − y2 − y1 +2 +���� +� +. +C-step. We combine the results of the association and prediction steps to get +Θy2(S2) = +� +u2∈S2 +Θy2(u2) = +� +ϑ ∈ Θ : +����ϑ(t2) − y1 + y2 +2 +���� ≤ +���� ˜U2 − y2 − y1 +2 +���� +� +, +where ˜U2 ∼ N((y2−y1)/2, 1/2). For any assertion A ⊆ Θ, the corresponding conditional belief function +is given by +cbely2|y2−y1(A; S2) = PS2|y2−y1{Θy2(S2) ⊆ A | Θy2(S2) ̸= ∅}, +(2.16) +and the conditional plausibility function is given by +cply2|y2−y1(A; S2) = 1 − cbely2|y2−y1(Ac; S2) = PS2|y2−y1{Θy2(S2) ∩ A ̸= ∅ | Θy2(S2) ̸= ∅}. +(2.17) +Note that, for any singleton assertion A = {ϑ0}, ϑ0 ∈ Θ, the conditional belief function cbely2|y2−y1(A; S2) = +0, while the conditional plausibility function takes the form of +cply2(A; S2) = PS2|y2−y1{ϑ0 ∈ Θy2(S2)} = 2 +� +1 − Φ +�√ +2 +����ϑ0(t2) − y1 + y2 +2 +���� +�� +. +5 + +3 +Partial Conditioning +3.1 +The general B = M|t2 − t1| case and its challenges +When B is large enough, neither observation can provide information on the location of the other observation. +In this case, we may simply choose to make inference based on associated single observation alone. The idea +is supported by the method of marginal IMs (Martin and Liu, 2015b). However, for the general case with +0 < B < ∞, it remains challenging to construct efficient IMs. +A seemingly attractive approach is the fiducial-type approach. In this case, inference is obtained using the +conditional distribution of U2 given U2 − U1 in the constrained interval by taking the distribution of U2 − U1 +as its original distribution restricted in the constrained interval. Since this constrained interval is implicitly +dependent on the unobserved realizations of U2 and U1, such a probability operation is questionable. In fact, +it generates an uncertainty assessment without the guarantee of the desired frequency calibration (Liu and +Martin, 2015). +A conservative but simple approach to making valid inference is to weaken the conditional belief function +(2.16) by taking its infimum and the conditional plausibility function (2.17) by taking its supremum over +all possible values of the conditioning variable U2 − U1. While valid inference is produced, this leads to +inefficient inference, which can be even worse than that based on a single data point alone. +3.2 +A new approach based on partial conditioning +We aim to construct nested predictive random sets by introducing the predictive distribution of the form +N +�λ +2 v⋆, 1 − λ + λ2 +2 +� +, +(3.1) +where λ ∈ [0, 1] is some function of B and v⋆ is a realization of V = U2 − U1. Note that the conditional +and marginal IMs are obtained as two extreme cases with λ = 1 for the fully conditional and λ = 0 for the +completely marginal. The case of 0 < λ < 1 thus resembles an approach we refer to as partial conditioning +or, more exactly, partial regression of, say, U2 on V = U2 − U1. The intuition and theoretical support for +the use of (3.1) for IM-based inference are explained below. +The key idea is to take into account the strengths of both conditional and marginal IMs. That is, we +consider conditional IMs but conservative for validity when B is small, and marginal IMs when conditioning- +based conservative IMs are inefficient. Technically, this amounts to utilizing the partial regression +U2 = λ +2 V + ε +to predict U2 through prediction of ε in such a way that the prediction of ε is valid marginally, and thereby +prediction of U2 is valid conditionally, given V in some interval. Marginally, ε is normal with mean zero and +variance +Var(ε) = Var +� +U2 − λ +2 V +� += +� +1 − λ +2 +�2 ++ +�λ +2 +�2 += 1 − λ + λ2 +2 . +This leads to the use of (3.1) as a valid inference. +To predict U2, we utilize the predictive random sets +Sv∗ = +� +z : +����z − λ +2 v⋆ +���� ≤ +����Z − λ +2 v⋆ +���� +� +, +Z ∼ N +�λ +2 v⋆, 1 − λ + λ2 +2 +� +when V is known to be v∗. For the two-point problem, V is known to be some v∗ in the interval [y2 − y1 − +B, y2 − y1 + B]. Thus, the use of the conservative prediction random set +S = +� +v∗∈[y2−y1−B, y2−y1+B] +Sv∗ +6 + +provides valid inference. For constructing confidence intervals, for example, the plausibility region +� +v⋆∈[y2−y1−B, y2−y1+B] +� +λ +2 v⋆ − z1−α/2 +� +1 − λ + λ2 +2 , λ +2 v⋆ + z1−α/2 +� +1 − λ + λ2 +2 +� +covers u⋆ +2 with probability at least 100(1 − α)%, for all λ ∈ [0, 1]. +Note that the length of the plausibility interval is given by +� λ +2 (y2 − y1 + B) − λ +2 (y2 − y1 − B) +� ++ 2z1−α/2 +� +1 − λ + λ2 +2 = λB + 2z1−α/2 +� +1 − λ + λ2 +2 . +(3.2) +Using the usual measure of efficiency in terms of interval length, we minimize (3.2) over λ ∈ [0, 1]. This leads +to the choice of λ: +ˆλB = +� +� +� +� +� +� +� +1 − +B +� +2z2 +1−α/2 − B2 , +if 0 ≤ B < z1−α/2; +0, +if B ≥ z1−α/2. +The corresponding widths are equal to B + +� +2z2 +1−α/2 − B2 and 2z1−α/2, respectively. +0.0 +0.2 +0.4 +0.6 +0.8 +3.0 +3.5 +4.0 +B +Width +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +Marginal +Mixture +Conditional +Figure 2: Comparison of the widths of the plausibility intervals constructed for the assertion +A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = +√ +t} ⊆ Θ1,1/2 using the three different approaches: Marginal IMs, +Partial Conditioning (“Mixture”), and Conservative Conditional IMs. +The optimal value of ˆλB suggests a full conditional IM when B is small relative to the width of the +target confidence interval or the confidence level. Figure 2 shows the numerical results of a simulation with +100 trials. We observe that the plausibility intervals obtained using partial conditioning are the narrowest. +4 +The general n case +For better understanding, we provide detailed investigation on the three-point case in Subsection 4.1 below. +The general n case follows as a straightforward generalization of the three-point problem and is summarized +in Section 4.2. +7 + +4.1 +The n = 3 case +For the case of n = 3, we have three pairs of observations (ti, yi), i = 1, 2, 3. Without loss of generality, +assume that t1 ≤ t2 ≤ t3. In this case, we have the data generation model. +� +� +� +� +� +y1 = ϑ0(t1) + u⋆ +1; +y2 = ϑ0(t2) + u⋆ +2; +y3 = ϑ0(t3) + u⋆ +3 +(4.1) +where u⋆ +1, u⋆ +2, u⋆ +3 represent unobserved realizations of U1, U2, U3 +iid +∼ N(0, 1). The problem of interest is to +make a (marginal) inference on ϑ0(ti), i = 1, 2, 3. To proceed, we first find the relevant conditional distribu- +tions in Subsection 4.1.1, and then provide the optimal solution under a partial conditioning framework in +Subsection 4.1.2. +4.1.1 +Conditional Distributions +The system of pairwise differences from (4.1), that is, +� +� +� +� +� +y2 − y1 = ϑ0(t2) − ϑ0(t1) + u⋆ +2 − u⋆ +1; +y3 − y1 = ϑ0(t3) − ϑ0(t1) + u⋆ +3 − u⋆ +1; +y3 − y2 = ϑ0(t3) − ϑ0(t2) + u⋆ +3 − u⋆ +2 +(4.2) +motivates us to introduce the potential conditioning auxiliary variables +V21 = U2 − U1, V31 = U3 − U1, V32 = U3 − U2. +Clearly, we have that +(V21, V31, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +2 +1 +−1 +1 +2 +1 +−1 +1 +2 +� +� +� +� . +By the H¨older condition (1.2), we have +� +� +� +� +� +|ϑ0(t2) − ϑ0(t1)| ≤ M|t2 − t1|γ; +|ϑ0(t3) − ϑ0(t1)| ≤ M|t3 − t1|γ; +|ϑ0(t3) − ϑ0(t2)| ≤ M|t3 − t2|γ. +(4.3) +For notational convenience, define +Bij := M|ti − tj|γ, 1 ≤ i, j ≤ 3. +From (4.2) and (4.3) we obtain the observed constraints on the pairwise differences v⋆ +i,: +v⋆ +ij = u⋆ +i − u⋆ +j = yi − yj − (ϑ0(ti) − ϑ0(tj)) ∈ [yi − yj − Bij, yi − yj + Bij], 1 ≤ j < i ≤ 3. +(4.4) +A complete characterization of the conditional distributions of the original auxiliary variables U1, U2, U3 +given the new auxiliary variables V21, V31, V32 is provided in Propositions 4.1 and 4.2, with proofs given in +the appendix. Direct applications of these propositions yield the corresponding results with respect to the +parameter ϑ ∈ Θ, which are summarized in Corollaries 4.1 and 4.2. +8 + +Proposition 4.1. For any fixed vij = ui − uj with i > j, the conditional distributions of U1, U2, U3 given +Vij = vij take the following forms: +U1|v21 ∼ N +�u1 − u2 +2 +, 1 +2 +� +, U1|v31 ∼ N +�u1 − u3 +2 +, 1 +2 +� +, U1|v32 ∼ N(0, 1); +(4.5) +U2|v21 ∼ N +�u2 − u1 +2 +, 1 +2 +� +, U2|v32 ∼ N +�u2 − u3 +2 +, 1 +2 +� +, U2|v31 ∼ N(0, 1); +(4.6) +U3|v31 ∼ N +�u3 − u1 +2 +, 1 +2 +� +, U3|v32 ∼ N +�u3 − u2 +2 +, 1 +2 +� +, U3|v21 ∼ N(0, 1). +(4.7) +Corollary 4.1. Proposition 4.1 yields the following results regarding the parameter ϑ ∈ Θ: +ϑ(t1) = Y1 − U1 ∈ N +�Y1 + Y2 +2 +± B21 +2 , 1 +2 +� � +N +�Y1 + Y3 +2 +± B31 +2 , 1 +2 +� +; +(4.8) +ϑ(t2) = Y2 − U2 ∈ N +�Y1 + Y2 +2 +± B21 +2 , 1 +2 +� � +N +�Y2 + Y3 +2 +± B32 +2 , 1 +2 +� +; +(4.9) +ϑ(t3) = Y3 − U3 ∈ N +�Y1 + Y3 +2 +± B31 +2 , 1 +2 +� � +N +�Y2 + Y3 +2 +± B32 +2 , 1 +2 +� +. +(4.10) +Proposition 4.2. For any fixed vi1j1 = ui1 − uj1 and vi2j2 = ui2 − uj2 with i1 > j1 and i2 > j2, the +conditional distributions of U1, U2, U3 given Vi1j1 = vi1j1 and Vi2j2 = vi2j2 take the following forms: +U1|v21, v31, U1|v21, v32, U1|v31, v32 ∼ N +�2u1 − u2 − u3 +3 +, 1 +3 +� +; +(4.11) +U2|v21, v31, U2|v21, v32, U2|v31, v32 ∼ N +�2u2 − u1 − u3 +3 +, 1 +3 +� +; +(4.12) +U3|v21, v31, U3|v21, v32, U3|v31, v32 ∼ N +�2u3 − u1 − u2 +3 +, 1 +3 +� +. +(4.13) +Corollary 4.2. Proposition 4.2 yields the following results with respect to parameter ϑ ∈ Θ: +ϑ(t1) = Y1 − U1 ∈ N +�Y1 + Y2 + Y3 +3 +± B21 + B31 +3 +, 1 +3 +� +; +ϑ(t2) = Y2 − U2 ∈ N +�Y1 + Y2 + Y3 +3 +± B21 + B32 +3 +, 1 +3 +� +; +ϑ(t3) = Y3 − U3 ∈ N +�Y1 + Y2 + Y3 +3 +± B31 + B32 +3 +, 1 +3 +� +. +These results can be restated more compactly as +ϑ(ti) ∈ N +� +¯Y ± ¯Bi, 1 +3 +� +, +i = 1, 2, 3. +where ¯Y = (Y1 + Y2 + Y3)/3, ¯Bi = (Bi1 + Bi2 + Bi3)/3, and we define Bii = 0 for i = 1, 2, 3. +9 + +Remark 4.1. For the sake of completeness, we remark that the covariance matrices in +(U1, V21, V31, V32) ∼ +� +� +� +� +� +��� +0 +0 +0 +0 +� +��� , +� +��� +1 +−1 +−1 +0 +−1 +2 +1 +−1 +−1 +1 +2 +1 +0 +−1 +1 +2 +� +��� +� +� +� +� ; +(U2, V21, V31, V32) ∼ +� +� +� +� +� +��� +0 +0 +0 +0 +� +��� , +� +��� +1 +1 +0 +−1 +1 +2 +1 +−1 +0 +1 +2 +1 +−1 +−1 +1 +2 +� +��� +� +� +� +� ; +(U3, V21, V31, V32) ∼ +� +� +� +� +� +��� +0 +0 +0 +0 +� +��� , +� +��� +1 +0 +1 +1 +0 +2 +1 +−1 +1 +1 +2 +1 +1 +−1 +1 +2 +� +��� +� +� +� +� , +are all degenerate, due to the fact that V21, V31, V32 are linearly dependent. +4.1.2 +Prediction via partial conditioning +Recall that our goal is to predict u⋆ +2 to make a valid probabilistic inference. Without loss of generality, +assume that t2 − t1 ≤ t3 − t2. Motivated by the idea of partial conditioning for the two-point problem, we +construct nested predictive random sets by introducing the partial regression of u⋆ +2 on pairwise differences: +N +� +λ1 +�u⋆ +2 − u⋆ +1 +2 +� ++ λ2 +�2u⋆ +2 − u⋆ +1 − u⋆ +3 +3 +� +, +�λ1 +2 + λ2 +3 +�2 ++ +� +1 − λ1 +2 − 2 +3λ2 +�2 ++ +�λ2 +3 +�2� +, +(4.14) +where the mixing proportions λ1, λ2 are functions of Bij satisfying λ1, λ2 ≥ 0 and λ1 + λ2 ≤ 1. Note that +the variance expression arises naturally from the fact that +Var +� +U2 − λ1 +2 (U2 − U1) − λ2 +3 (2U2 − U1 − U3) +� += +�λ1 +2 + λ2 +3 +�2 ++ +� +1 − λ1 +2 − 2 +3λ2 +�2 ++ +�λ2 +3 +�2 +. +To predict U2, we take the predictive random sets +Sv⋆ +21,v⋆ +32 = +� +z : +����z − +� +λ1 +�v⋆ +21 +2 +� ++ λ2 +�v⋆ +21 − v⋆ +32 +3 +������ ≤ +����Z − +� +λ1 +�v⋆ +21 +2 +� ++ λ2 +�v⋆ +21 − v⋆ +32 +3 +������ +� +, +where we recall that +v⋆ +ij = u⋆ +i − u⋆ +j = yi − yj − (ϑ0(ti) − ϑ0(tj)) ∈ [yi − yj − Bij, yi − yj + Bij], 1 ≤ j < i ≤ 3, +and Z follows the Gaussian distribution (4.14). The predictive random sets are clearly marginally valid, and +thus the corresponding IMs are valid. Therefore, the we utilize the conservative predictive random set +S = +� +v⋆ +21∈[y2−y1±B21] +v⋆ +32∈[y3−y2±B32] +Sv⋆ +21,v⋆ +32 +For constructing confidence intervals, this suggests the plausibility region +� +v⋆ +21∈[y2−y1±B21] +v⋆ +32∈[y3−y2±B32] +� +� +� +��λ1 +2 + λ2 +3 +� +v⋆ +21 − +�λ2 +3 +� +v⋆ +32 +� +± z1−α/2 +��λ1 +2 + λ2 +3 +�2 ++ +� +1 − λ1 +2 − 2 +3λ2 +�2 ++ +�λ2 +3 +�2 +� +� +� , +10 + +which covers u⋆ +2 with probability at least 100(1 − α)%. +It can be easily verified that the width of this +plausibility region is given by +2 +��λ1 +2 + λ2 +3 +� +B21 + +�λ2 +3 +� +B32 +� ++ 2z1−α/2 +��λ1 +2 + λ2 +3 +�2 ++ +� +1 − λ1 +2 − 2 +3λ2 +�2 ++ +�λ2 +3 +�2 +. +The optimal mixing proportions λ1, λ2 are the solutions to the constrained optimization problem +min +λ1, λ2 +��λ1 +2 + λ2 +3 +� +B21 + +�λ2 +3 +� +B32 +� ++ z1−α/2 +��λ1 +2 + λ2 +3 +�2 ++ +� +1 − λ1 +2 − 2 +3λ2 +�2 ++ +�λ2 +3 +�2 +s.t. +λ1, λ2 ≥ 0; λ1 + λ2 ≤ 1. +(4.15) +Solving optimization problem (4.15) analytically requires finding the roots of a set of quadratic equations, +a process which can become quite burdensome. Instead, we adopt numerical optimization methods such as +BFGS to solve problem (4.15) in order to obtain the optimal mixing proportions which minimize the length +of the plausibility interval. Figure 3 shows the numerical results of a simulation with 500 trials. Again, we +find that the plausibility intervals obtained using optimal mixture distributions are the narrowest. +0.1 +0.2 +0.3 +0.4 +0.5 +0.6 +2.5 +3.0 +3.5 +4.0 +B21 +Width +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +GG +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +GG +G +G +G +G +G +G +G +GG +G +G G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GGG +G +G +G +G +G +G +G +G +G +G GGG +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +Marginal +Mixture +Cond'l (1 pt) +Cond'l (2 pts) +(a) Width vs B12 +0.2 +0.4 +0.6 +0.8 +2.5 +3.0 +3.5 +4.0 +B32 +Width +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +GG +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +G +Marginal +Mixture +Cond'l (1 pt) +Cond'l (2 pts) +(b) Width vs B23 +Figure 3: Comparison of the widths of the plausibility intervals constructed for the assertion +A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = +√ +t} ⊆ Θ1,1/2 using the four different approaches. +4.2 +The general n case +In this section, we extend our previous discussions to the general case of n observations, and provide a generic +method to construct valid and efficient pointwise plausibility intervals. Denote our sequence of observations +by (ti, yi), i = 1, · · · , n, where it is not necessary to assume that the ti’s are sorted. With the assumption +σ = 1 made in Section 1, the association (1.1) can be written as +yi = ϑ0(ti) + u⋆ +i , i = 1, · · · , n. +(4.16) +11 + +In the rest of this section, we outline the process of conducting inference on Ui for an arbitrary 1 ≤ i ≤ n. As +a preprocessing step, we sort the n observations in ascending order of their distance from the i-th observation, +such that 0 = |ti − t(i) +0 | ≤ |ti − t(i) +1 | ≤ · · · ≤ |ti − t(i) +n−1|, correspondingly. Denote by U (i) +0 , U (i) +1 , · · · , U (i) +n−1 +the corresponding U variables, and by y(i) +0 , y(i) +1 , · · · , y(i) +n−1 the corresponding y observations. Notice that +U (i) +0 += Ui holds trivially. +To predict Ui, we can utilize the marginal distribution N(0, 1), and the conditional distributions Ui|ui − +u(i) +1 ; Ui|ui − u(i) +1 , ui − u(i) +2 ; · · · ; Ui|ui − u(i) +1 , ui − u(i) +2 , · · · , ui − u(i) +n−1. This motivates us to consider the +partial regression of Ui on the pairwise differences of the U variables: +N +� +� +� +n−1 +� +k=1 +λk +k + 1 +� +�k ui − +k +� +j=1 +u(i) +j +� +� , +� +1 − +n−1 +� +k=1 +λk k +k + 1 +�2 ++ +n−1 +� +j=1 +� +� +n−1 +� +k=j +λk +k + 1 +� +� +2� +� +� , +(4.17) +where the mixing proportions λ1, λ2, · · · , λn−1 are functions of Bij satisfying λ1, λ2, · · · , λn−1 ≥ 0 and +�n−1 +k=1 λk ≤ 1. The variance expression in Equation (4.17) is obtained by routine algebraic operations: +Var +� +� +�Ui − +n−1 +� +k=1 +λk +k + 1 +� +�k Ui − +k +� +j=1 +U (i) +j +� +� +� +� +� += +� +1 − +n−1 +� +k=1 +λk k +k + 1 +�2 ++ Var +� +� +� +n−1 +� +k=1 +� +� λk +k + 1 +k +� +j=1 +U (i) +j +� +� +� +� +� , +(4.18) +To deal with the second term on the right-hand-side of the last equation, we notice that the summation signs +can be exchanged in the following manner: +n−1 +� +k=1 +� +� λk +k + 1 +k +� +j=1 +U (i) +j +� +� = +n−1 +� +k=1 +k +� +j=1 +λk +k + 1U (i) +j += +n−1 +� +j=1 +n−1 +� +k=1 +λk +k + 1U (i) +j += +n−1 +� +j=1 +� +� +n−1 +� +k=j +λk +k + 1 +� +� U (i) +j . +Therefore, we have +Var +� +� +� +n−1 +� +k=1 +� +� λk +k + 1 +k +� +j=1 +U (i) +j +� +� +� +� +� = Var +� +� +� +n−1 +� +j=1 +� +� +n−1 +� +k=j +λk +k + 1 +� +� U (i) +j +� +� +� = +n−1 +� +j=1 +� +� +n−1 +� +k=j +λk +k + 1 +� +� +2 +, +where the last step follows from the independence of U (i) +j , j = 1, · · · , n. Substituting this result into equation +(4.18) yields the variance expression shown in equation (4.17). +To predict Ui, we take the predictive random sets +S{v(i) +j +:j=1,...,k} = +� +� +�z : +������ +z − +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +v(i) +j +������ +≤ +������ +Z − +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +v(i) +j +������ +� +� +� , Z ∼ (4.17), +where +v(i) +j +:= ui − u(i) +j += yi − y(i) +j +− (ϑ0(ti) − ϑ0(t(i) +j )) ∈ [yi − y(i) +j +− B(i) +j , yi − y(i) +j ++ B(i) +j ], +with B(i) +j +:= |t(i) +j +− ti|, 1 ≤ i, j ≤ n. For valid inference, we take the (conservative) predictive random sets +S = +� +{v(i) +j +∈ +� +yi−y(i) +j +±B(i) +j +� +,j=1,...,k} +S{v(i) +j +:j=1,...,k}. +This predictive random set is clearly marginally valid. Let +∆λ = +� +� +� +� +� +� +1 − +n−1 +� +k=1 +λk k +k + 1 +�2 ++ +n−1 +� +j=1 +� +� +n−1 +� +k=j +λk +k + 1 +� +� +2 +, +12 + +the standard deviation of the partial regression. The plausibility region +� +v(i) +j +∈ +� +yi−y(i) +j +±B(i) +j +� +� +� +� +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +v(i) +j +± z1−α/2∆λ +� +� +� , +or equivalently, +� +� +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +� +yi − y(i) +j +− B(i) +j +� +− z1−α/2∆λ, +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +� +yi − y(i) +j ++ B(i) +j +� ++ z1−α/2∆λ +� +� , +covers u⋆ +i with probability at least 100(1 − α)%. The width of the plausibility region is given by +2 +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +B(i) +j ++ 2z1−α/2∆λ. +The optimal set of mixing proportions λ1, λ2, · · · , λn−1 are the solutions to the constrained optimization +problem +min +λ1,λ2,··· ,λn−1 +n−1 +� +k=1 +λk +k + 1 +k +� +j=1 +B(i) +j ++ z1−α/2∆λ +subject to +λ1, λ2, · · · , λn−1 ≥ 0; +n−1 +� +k=1 +λk ≤ 1. +(4.19) +Solving optimization problem (4.19) analytically requires finding the roots of a set of quadratic equations, +a process which can become quite burdensome. Instead, we adopt numerical optimization methods such as +BFGS to solve problem (4.19) in order to obtain the optimal mixing proportions which minimize the width +of the plausibility interval. This is demonstrated using numerical results in Figures 4, 5, and 6 for different +configurations of n, M, and γ. +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +−2 +−1 +0 +1 +2 +3 +t +y +(a) n = 20, σ = 1 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +−2 +−1 +0 +1 +2 +3 +t +y +(b) n = 20, σ = 1/2 +Figure 4: Visualization of the plausibility intervals constructed using the mixture distribution for +the assertion A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = +√ +t} ⊆ Θ1,1/2 under various settings. +13 + +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +−3 +−2 +−1 +0 +1 +2 +t +y +(a) n = 15, M = 1/2, γ = 1/5 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +−3 +−2 +−1 +0 +1 +2 +t +y +(b) n = 15, M = 1/2, γ = 1 +Figure 5: Visualization of the plausibility intervals constructed using the mixture distribution for +the assertion A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = 0} ⊆ Θ1,1/2 under various settings. +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +−3 +−2 +−1 +0 +1 +2 +t +y +(a) n = 10 +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +−3 +−2 +−1 +0 +1 +2 +t +y +(b) n = 100 +Figure 6: Visualization of the plausibility intervals constructed using the mixture distribution for +the assertion A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = 0} ⊆ Θ1,1/2 for various number of observations. +5 +Discussion +We have developed a partial conditioning method to extend IMs for valid and efficient inference about many- +normal-means when the means are subject to constraints. While the basic idea and method are developed +for many-normal-means, they can be extended to handle many-binomial-means and many-Poisson-means +with similiar constraints. Nevertheless, a more general mathematical theory deserves future development. +14 + +Appendix: Proofs +Proof of Proposition 2.1 +Proof. For the predictive random set S = [−|U1|, |U1|], U1 ∼ N(0, 1), we have +PS{S ̸∋ u⋆} = P{|U1| < |u⋆ +1|} = 2Φ(|u⋆ +1|) − 1, +where Φ(·) denotes the standard Normal CDF. It follows that for each α ∈ (0, 1), +PU1{PS{S ̸∋ U1} ≥ 1 − α} = PU1{2Φ(|U1|) − 1 ≥ 1 − α} = PU1{Φ(|U1|) ≥ 1 − α/2} = α, +which verifies that (2.5) holds. Thus, we have shown that the predictive random set S is valid, and Theo- +rem 2.1 implies that the IM previously defined is valid as well. +Proof of Proposition 4.1 +Proof. The conditional distributions for U1 can be derived as follows: +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +(U1, V21) ∼ N2 +�� 0 +0 +� +, +� +1 +−1 +−1 +2 +�� +⇒ +U1|V21 = v21 ∼ N +� +−v21 +2 , 1 +2 +� +; +(U1, V31) ∼ N2 +�� 0 +0 +� +, +� +1 +−1 +−1 +2 +�� +⇒ +U1|V31 = v31 ∼ N +� +−v31 +2 , 1 +2 +� +; +(U1, V32) ∼ N2 +�� 0 +0 +� +, +� 1 +0 +0 +2 +�� +⇒ +U1|V32 = v32 ∼ N(0, 1). +The conditional distributions for U2 can be derived as follows: +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +(U2, V21) ∼ N2 +�� 0 +0 +� +, +� 1 +1 +1 +2 +�� +⇒ +U2|V21 = v21 ∼ N +�v21 +2 , 1 +2 +� +; +(U2, V31) ∼ N2 +�� 0 +0 +� +, +� 1 +0 +0 +2 +�� +⇒ +U2|V31 = v31 ∼ N(0, 1); +(U2, V32) ∼ N2 +�� 0 +0 +� +, +� +1 +−1 +−1 +2 +�� +⇒ +U2|V32 = v32 ∼ N +� +−v32 +2 , 1 +2 +� +. +The conditional distributions for U3 can be derived as follows: +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +(U3, V21) ∼ N2 +�� 0 +0 +� +, +� 1 +0 +0 +2 +�� +⇒ +U3|V21 = v21 ∼ N(0, 1); +(U3, V31) ∼ N2 +�� 0 +0 +� +, +� 1 +1 +1 +2 +�� +⇒ +U3|V31 = v31 ∼ N +�v31 +2 , 1 +2 +� +; +(U3, V32) ∼ N2 +�� 0 +0 +� +, +� 1 +1 +1 +2 +�� +⇒ +U3|V32 = v32 ∼ N +�v32 +2 , 1 +2 +� +. +The proof is complete by recalling that vij = ui − uj for 1 ≤ j < i ≤ 3. +15 + +Proof of Corollary 4.1 +Proof. We notice that the results in (4.4) imply the following: +u1 − u2 = −v21 ∈ [Y1 − Y2 − B21, Y1 − Y2 + B21], +u1 − u3 = −v31 ∈ [Y1 − Y3 − B31, Y1 − Y3 + B31]; +u2 − u1 = v21 ∈ [Y2 − Y1 − B21, Y2 − Y1 + B21], +u2 − u3 = −v32 ∈ [Y2 − Y3 − B32, Y2 − Y3 + B32]; +u3 − u1 = v31 ∈ [Y3 − Y1 − B31, Y3 − Y1 + B31], +u3 − u2 = v32 ∈ [Y3 − Y2 − B32, Y3 − Y2 + B32]. +Combining these results with Proposition 4.1 completes the proof of the corollary. +Proof of Proposition 4.2 +Proof. The conditional distributions for U1 can be derived as follows: +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +(U1, V21, V31) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +−1 +−1 +−1 +2 +1 +−1 +1 +2 +� +� +� +� +⇒ +U1|v21, v31 ∼ N +� +−v21 + v31 +3 +, 1 +3 +� +; +(U1, V21, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +−1 +0 +−1 +2 +−1 +0 +−1 +2 +� +� +� +� +⇒ +U1|v21, v32 ∼ N +� +−2v21 + v32 +3 +, 1 +3 +� +; +(U1, V31, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +−1 +0 +−1 +2 +1 +0 +1 +2 +� +� +� +� +⇒ +U1|v31, v32 ∼ N +�v32 − 2v31 +3 +, 1 +3 +� +. +The conditional distributions for U2 can be derived as follows: +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +(U2, V21, V31) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +1 +0 +1 +2 +1 +0 +1 +2 +� +� +� +� +⇒ +U2|v21, v31 ∼ N +�2v21 − v31 +3 +, 1 +3 +� +; +(U2, V21, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +1 +−1 +1 +2 +−1 +−1 +−1 +2 +� +� +� +� +⇒ +U2|v21, v32 ∼ N +�v21 − v32 +3 +, 1 +3 +� +; +(U2, V31, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +0 +−1 +0 +2 +1 +−1 +1 +2 +� +� +� +� +⇒ +U2|v31, v32 ∼ N +�v31 − 2v32 +3 +, 1 +3 +� +. +The conditional distributions for U3 can be derived as follows: +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +� +(U3, V21, V31) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +0 +1 +0 +2 +1 +1 +1 +2 +� +� +� +� +⇒ +U3|v21, v31 ∼ N +�2v31 − v21 +3 +, 1 +3 +� +; +(U3, V21, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +0 +1 +0 +2 +−1 +1 +−1 +2 +� +� +� +� +⇒ +U3|v21, v32 ∼ N +�v21 + 2v32 +3 +, 1 +3 +� +; +(U3, V31, V32) ∼ N3 +� +� +� +� +0 +0 +0 +� +� , +� +� +1 +1 +1 +1 +2 +1 +1 +1 +2 +� +� +� +� +⇒ +U3|v31, v32 ∼ N +�v31 + v32 +3 +, 1 +3 +� +. +16 + +It can be easily verified that the three conditional distributions in each group are actually equivalent, and +that the results (4.11) – (4.13) hold. +Proof of Corollary 4.2 +Proof. We notice that the results in (4.4) implies the following: +2u1 − u2 − u3 = −(v21 + v31) ∈ [2Y1 − Y2 − Y3 − (B21 + B31), 2Y1 − Y2 − Y3 + (B21 + B31)]; +2u2 − u1 − u3 = v21 − v32 ∈ [2Y2 − Y1 − Y3 − (B21 + B32), 2Y2 − Y1 − Y3 + (B21 + B32)]; +2u3 − u1 − u2 = v31 + v32 ∈ [2Y3 − Y1 − Y2 − (B31 + B32), 2Y1 − Y2 − Y3 + (B31 + B32)]. +Combining these results with Proposition 4.2 completes the proof of the corollary. +References +R. Gong and X.-L. Meng. Judicious judgment meets unsettling updating: Dilation, sure loss and Simpson’s +paradox. Statistical Science, 36(2):169–190, 2021. +D. E. Leaf and C. Liu. Inference about constrained parameters using the elastic belief method. International +Journal of Approximate Reasoning, 53(5):709–727, 2012. +O. Lepskii. +On a problem of adaptive estimation in gaussian white noise. +Theory of Probability & Its +Applications, 35(3):454–466, 1991. +C. Liu and R. Martin. Frameworks for prior-free posterior probabilistic inference. Wiley Interdisciplinary +Reviews: Computational Statistics, 7(1):77–85, 2015. +C. Liu and R. Martin. Comment: Settle the unsettling: An inferential models perspective. Statistical Science, +36(2):196–200, 2021. +R. Martin and C. Liu. +Inferential models: A framework for prior-free posterior probabilistic inference. +Journal of the American Statistical Association, 108(501):301–313, 2013. +R. Martin and C. Liu. Conditional inferential models: combining information for prior-free probabilistic +inference. +Journal of the Royal Statistical Society: Series B (Statistical Methodology), 77(1):195–217, +2015a. +R. Martin and C. Liu. Marginal inferential models: prior-free probabilistic inference on interest parameters. +Journal of the American Statistical Association, 110(512):1621–1631, 2015b. +17 + diff --git a/VtE3T4oBgHgl3EQfbAqu/content/tmp_files/load_file.txt b/VtE3T4oBgHgl3EQfbAqu/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..8dec00a3cbc3495c9628d268d8eb73829155f385 --- /dev/null +++ b/VtE3T4oBgHgl3EQfbAqu/content/tmp_files/load_file.txt @@ -0,0 +1,2512 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf,len=2511 +page_content='Partial Conditioning for Inferential Models Jiasen Yang, Xiao Wang, Chuanhai Liu Department of Statistics Purdue University Abstract Inferential models have been proposed for valid and efficient prior-free probabilistic inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' As it gradually gained popularity, this theory is subject to further developments for practically challenging problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This paper considers the many-normal-means problem with the means constrained to be in the neighborhood of each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' A new method, called partial conditioning, is proposed to generate valid and efficient marginal inference about the individual means.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' It is shown that the method outperforms both a fiducial-counterpart in terms of validity and a conservative-counterpart in terms of efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Keywords— Fiducial argument, H¨older space, Lipschitz space, Many-normal-means 1 Introduction The framework of inferential models (IMs) has been developed to provide valid and efficient prior-free probabilistic inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In its simplest form (Martin and Liu, 2013), it is obtained by modifying R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Fisher’s fiducial argument, which suffers from the mathematical difficulties discussed, for example, in Liu and Martin (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' It has also been extended for efficient inference by combining information (Martin and Liu, 2015a) and marginalization (Martin and Liu, 2015b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For certain practically challenging problems, however, this theory is subject to further developments, similar to other existing schools of thought.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' One noticeable class of such challenging problems is inference on constrained parameters, which is challenging for all existing inferential methods (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=', Leaf and Liu (2012) and references therein).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In this paper, we consider the many-normal-means problem with the means constrained to be in the neighborhood of each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For both conceptual and representational simplicity, we describe the problem as a special case of the familiar classic formulation of a nonparametric regression model (Lepskii, 1991).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' More precisely, to motivate the problem, we consider the following model Yi = ϑ0(ti) + σUi, i = 1, · · · , n, (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) where the ti denotes the design points, the Yi denotes the responses, and Ui i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' ∼ N(0, 1) denote the random error terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Here, ϑ0 is the unknown regression function that is assumed to reside in the H¨older space ΘM,γ = � ϑ : [0, 1] → R ��� |ϑ(t) − ϑ(s)| ≤ M|t − s|γ, ∀t, s ∈ [0, 1] � , (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) with the H¨older exponent 0 < γ ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We assume M, γ, and σ to be known constants and, without loss of generality, we set σ = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In subsequent discussions, we suppress the subscripts and denote the H¨older space ΘM,γ simply by Θ, unless otherwise noted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The objective is to perform statistical inference on the unknown function ϑ0 within the IM framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We propose a new method, called partial conditioning, to generate valid and efficient marginal inference 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='04512v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='ST] 11 Jan 2023 about the individual means.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We show that the method outperforms both a fiducial-counterpart in term of validity and a conservative-counterpart in terms of efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' It should be noted that IM approaches to the general nonparametric problem itself deserve in-depth investigations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This explains why our focus here is on a simple case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Our goal is to introduce an innovative idea of conditioning in the IM framework to combine information when multiple parameters entangle with each other via known constraints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The rest of the paper is arranged as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Section 2 reviews basic IMs (Martin and Liu, 2013) and conditional IMs (Martin and Liu, 2015a) with the simple cases of n = 1 and n = 2 with M = 0 in (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) and (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Section 3 addresses the difficulties in the case of n = 2 and M ̸= 0 and introduces the partial conditioning method, with the general n case considered in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Finally, Section 5 concludes with a few remarks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 2 The IMs: an overview 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 The basic IMs To illustrate the basic IM framework under the current problem setting, let us first consider the simple case of n = 1, where we only make use of a single pair of observations (t1, y1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Thus, we have y1 = ϑ0(t1) + u⋆ 1, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) where u⋆ 1 represents an unobserved realization of U1 ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Using the terminology of IMs, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) describes an underlying sampling model which involves the unknown parameter (function) ϑ ∈ Θ and generates observed data X1 = (t1, Y1) using auxiliary variable U1 ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Note that the auxiliary variable is unobserved but predictable, since its distribution is fully specified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In general, constructing an IM consists of three steps, namely, an association (A) step, a prediction (P) step, and a combination (C) step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' These three steps are explained below in more detail in the context of the one-point nonparametric regression problem (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1): A-step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The association step can be achieved via some function or procedure F as X1 = F(U1, ϑ), (X ∈ X, ϑ ∈ Θ, U ∼ N(0, 1)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This association allows for direct reasoning with the source of uncertainty U1, which is missing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' If U1 = u⋆ 1 were observed, we would then be able to obtain the best possible inference for ϑ, which is given by the set-valued “inverse” mapping G : u⋆ 1 → ΘX1(u⋆ 1) = {ϑ ∈ Θ : ϑ(t1) = Y1 − u⋆ 1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' P-step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For inference on ϑ, the discussion in the A-step suggests that we should focus our attention on accurately predicting the unobserved quantity u⋆ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To predict u⋆ 1 with a certain desired accuracy, we utilize a predictive random set S, for example, S(U1) = {˜u ∈ R : −|U1| ≤ ˜u ≤ |U1|}, (U1 ∼ N(0, 1)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) C-step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To transfer the available information about u⋆ to the ϑ-space, the last step is to combine the information in the association, the observed x1 = (t1, y1), and the predictive random set S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For this purpose, consider the expanded set Θx1(S) = � u1∈S Θx1(u1) = {ϑ ∈ Θ : |ϑ(t1) − y1| ≤ |U1|}, (U1 ∼ N(0, 1)) which contains those values of ϑ that are consistent with the observed data and the sampling model for at least one candidate u⋆ 1 value u1 ∈ S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 2 The random sets obtained in the C-step are in the space of the unknown parameter, and we are ready to produce uncertainty assessment for assertions of interest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Consider an assertion A about the parameter of interest ϑ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The assertion A corresponds to a set A ⊆ Θ, and acts as a hypothesis in the context of classical statistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To summarize the evidence in x that supports the assertion A, we evaluate the belief function defined by belx(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) = PS{Θx(S) ⊆ A | Θx(S) ̸= ∅}, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='3) and the plausibility function defined by plx(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) = 1 − belx(Ac, S) = PS{Θx(S) ∩ A ̸= ∅ | Θx(S) ̸= ∅}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4) The belief function is subadditive in the sense that if ∅ ̸= A ⊆ Θ, then bel(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S)+bel(Ac;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) ≤ 1 with equality if and only if Θx(S) is a singleton with PS-probability 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Therefore, it follows that belx(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) ≤ plx(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S), and they are also referred to as the lower and upper probabilities, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Incidentally, we note that there are continuing interests in using lower and upper probabilities for statistical inference (Gong and Meng, 2021;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu and Martin, 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To establish the frequency-calibration properties of the IM in the context of our current problem, we briefly review several results regarding validity as introduced by Martin and Liu (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Definition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' A predictive random set S is valid for predicting the unobserved auxiliary variable U if for each α ∈ (0, 1), PU{PS{S ̸∋ U} ≥ 1 − α} ≤ α.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5) Definition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The IM is valid if, for all assertions A and for any α ∈ (0, 1), sup ϑ∈A PX|ϑ{plX(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) ≤ α} ≤ α.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6) Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Suppose that the predictive random set S is valid, and Θx(S) ̸= ∅ with PS-probability 1 for all x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Then the IM is valid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In other words, a valid inference in the sense of frequency calibration is obtained as long as predictive random sets are valid to predict unobserved auxiliary variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Formally, for the one-point case, we have the following result, followed by a numerical example to illustrate plausibility-based confidence intervals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The predictive random set defined by (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) is valid and, therefore, the IM with the belief function given by (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='3) and the plausibility function given by (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4) is valid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Example 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 (An numerical illustration).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Consider the assertion Aθ0 = {ϑ0 : ϑ0(t1) = θ0} ⊆ Θ1,1/2 and the predictive random set S = [−|U1|, |U1|], U1 ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Since this assertion constitutes a singleton in the parameter space, the belief function bely1(Aθ0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) = 0, but the plausibility function ply1(Aθ0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) = PS{θ0 ∈ Θx1(S)} = PU1{|U1| ≥ |θ0 − y1|} = 2(1 − Φ(|θ0 − y1|)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='7) Figure 1 shows the graphs of ply1(Aθ0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) as a function of θ0 for y1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Remark 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 (Plausibility intervals).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In Example 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1, the assertion A = {ϑ0 = θ0}, ϑ0 ∈ Θ constitutes a singleton in the parameter space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In this case, we can define the 100(1 − α)% plausibility region given by Πx(α) = {ϑ : plx(ϑ) ≥ α}, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8) where plx(ϑ) := plx({ϑ};' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' It was shown in Martin and Liu (2013) that the plausibility region (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8) provides an exact 100(1 − α)% confidence interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 3 −3 −2 −1 0 1 2 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 θ plausibility Figure 1: The plausibility function ply1(Aθ0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S) defined in (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='7) as a function of θ0 with y1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 The conditional IMs As seen in the previous section for the simple one-point case, the basic IMs framework is similar to the frequentist pivotal method of constructing confidence intervals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' However, the way to make inference by predicting unobserved auxiliary variables can help develop efficient inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Two such methods, called conditional IMs and marginal IMs, are first explained here for efficient inference in the special cases of the two-point problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' A new method of taking the strengths of conditional IMs and marginal IMs is proposed for the general case of the two-point problem in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In the n = 2 case, we have two pairs of observations (t1, y1) and (t2, y2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The sampling model for this case can be written as � y1 = ϑ0(t1) + u⋆ 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' y2 = ϑ0(t2) + u⋆ 2, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='9) where u⋆ 1 and u⋆ 2 represent unobserved realizations of U1, U2 iid ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The IM framework for the current problem repeats the three steps introduced in the previous section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' However, we notice that in this two- point case, we have two auxiliary variables U1, U2, which would need to be predicted using a two-dimensional predictive random set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' As Martin and Liu (2015a) pointed out, a more efficient inference procedure can be obtained by reducing the dimension of the auxiliary variable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In particular, if some functions of the original auxiliary variable are fully observed, we can condition on the fully observed information to sharpen our prediction of the unobserved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Notice that from (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='9) we can obtain y2 − y1 = ϑ0(t2) − ϑ0(t1) + u⋆ 2 − u⋆ 1, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='10) which motivates us to introduce a new auxiliary variable V = U2 − U1 ∼ N(0, 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' By the H¨older condition (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2), we have |ϑ0(t1) − ϑ0(t2)| ≤ M|t1 − t2|γ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='11) Letting B := M|t1 − t2|γ, from (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='10) and (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='11) we obtain v⋆ = u⋆ 2 − u⋆ 1 = y2 − y1 − (ϑ0(t2) − ϑ0(t1)) ∈ [y2 − y1 − B, y2 − y1 + B].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='12) For each fixed v ∈ [y2 − y1 − B, y2 − y1 + B], the conditional distributions of a linear function of U1 and U2, for example, U2, given V = v can be derived as follows: (U2, V ) ∼ N2 �� 0 0 � , � 1 1 1 2 �� =⇒ U2|V = v ∼ N �v 2, 1 2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='13) 4 Thus, these conditional distributions, sharper than the corresponding marginal distributions, motivate us to construct more efficient predictive random sets for more efficient inference about (ϑ0(t1), ϑ0(t2)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For the special case of B = 0, the equation (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='10) simplifies to y2 − y1 = u⋆ 2 − u⋆ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Most important, u⋆ 2 − u⋆ 1 in this extreme case is fully observed and, therefore, can be easily used to predict u⋆ 1 and u⋆ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Formally, we can construct a conditional inferential model (CIM) as proposed by Martin and Liu (2015a).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Similar to basic IMs, conditional IMs also have their three steps: A-step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Under the original IM framework, the association step is achieved via the baseline association (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In the case of B = 0, this baseline association can be decomposed as � y2 − y1 = u⋆ 2 − u⋆ 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' y2 = ϑ(t2) + u⋆ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='14) This decomposition immediately suggests an alternative association.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Let PU2|y2−y1 denote the condi- tional distribution of U2, given U2 − U1 = y2 − y1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Since y2 − y1 does not provide information on the parameter ϑ, we can establish a new association Y2 = ϑ(t2) + ˜U2, ( ˜U2 ∼ PU2|y2−y1), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='15) which is referred to as the conditional association.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Using (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='15), we can associate the observed infor- mation Y2 and the parameter ϑ with the new auxiliary variable ˜U2 ∼ PU2|Y2−Y1 to get the collection of sets Θy2(˜u⋆ 2) = {ϑ ∈ Θ : y2 = ϑ(t2) + ˜u⋆ 2}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' P-step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Fixing the observed value Y2 − Y1 = y2 − y1, we predict the unobserved value ˜u⋆ 2 of ˜U2 with a conditionally admissible predictive random set S2 ∼ PS2|y2−y1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Notice that equation (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='13) implies that ˜U2 ∼ N � y2−y1 2 , 1 2 � , which gives rise to the default predictive random set given by S2 = � ˜u2 : ����˜u2 − y2 − y1 2 ���� ≤ ���� ˜U2 − y2 − y1 2 ���� � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' C-step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We combine the results of the association and prediction steps to get Θy2(S2) = � u2∈S2 Θy2(u2) = � ϑ ∈ Θ : ����ϑ(t2) − y1 + y2 2 ���� ≤ ���� ˜U2 − y2 − y1 2 ���� � , where ˜U2 ∼ N((y2−y1)/2, 1/2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For any assertion A ⊆ Θ, the corresponding conditional belief function is given by cbely2|y2−y1(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S2) = PS2|y2−y1{Θy2(S2) ⊆ A | Θy2(S2) ̸= ∅}, (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='16) and the conditional plausibility function is given by cply2|y2−y1(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S2) = 1 − cbely2|y2−y1(Ac;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S2) = PS2|y2−y1{Θy2(S2) ∩ A ̸= ∅ | Θy2(S2) ̸= ∅}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='17) Note that, for any singleton assertion A = {ϑ0}, ϑ0 ∈ Θ, the conditional belief function cbely2|y2−y1(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S2) = 0, while the conditional plausibility function takes the form of cply2(A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' S2) = PS2|y2−y1{ϑ0 ∈ Θy2(S2)} = 2 � 1 − Φ �√ 2 ����ϑ0(t2) − y1 + y2 2 ���� �� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 5 3 Partial Conditioning 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 The general B = M|t2 − t1| case and its challenges When B is large enough, neither observation can provide information on the location of the other observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In this case, we may simply choose to make inference based on associated single observation alone.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The idea is supported by the method of marginal IMs (Martin and Liu, 2015b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' However, for the general case with 0 < B < ∞, it remains challenging to construct efficient IMs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' A seemingly attractive approach is the fiducial-type approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In this case, inference is obtained using the conditional distribution of U2 given U2 − U1 in the constrained interval by taking the distribution of U2 − U1 as its original distribution restricted in the constrained interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Since this constrained interval is implicitly dependent on the unobserved realizations of U2 and U1, such a probability operation is questionable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In fact, it generates an uncertainty assessment without the guarantee of the desired frequency calibration (Liu and Martin, 2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' A conservative but simple approach to making valid inference is to weaken the conditional belief function (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='16) by taking its infimum and the conditional plausibility function (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='17) by taking its supremum over all possible values of the conditioning variable U2 − U1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' While valid inference is produced, this leads to inefficient inference, which can be even worse than that based on a single data point alone.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 A new approach based on partial conditioning We aim to construct nested predictive random sets by introducing the predictive distribution of the form N �λ 2 v⋆, 1 − λ + λ2 2 � , (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) where λ ∈ [0, 1] is some function of B and v⋆ is a realization of V = U2 − U1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Note that the conditional and marginal IMs are obtained as two extreme cases with λ = 1 for the fully conditional and λ = 0 for the completely marginal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The case of 0 < λ < 1 thus resembles an approach we refer to as partial conditioning or, more exactly, partial regression of, say, U2 on V = U2 − U1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The intuition and theoretical support for the use of (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) for IM-based inference are explained below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The key idea is to take into account the strengths of both conditional and marginal IMs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' That is, we consider conditional IMs but conservative for validity when B is small, and marginal IMs when conditioning- based conservative IMs are inefficient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Technically, this amounts to utilizing the partial regression U2 = λ 2 V + ε to predict U2 through prediction of ε in such a way that the prediction of ε is valid marginally, and thereby prediction of U2 is valid conditionally, given V in some interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Marginally, ε is normal with mean zero and variance Var(ε) = Var � U2 − λ 2 V � = � 1 − λ 2 �2 + �λ 2 �2 = 1 − λ + λ2 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This leads to the use of (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) as a valid inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To predict U2, we utilize the predictive random sets Sv∗ = � z : ����z − λ 2 v⋆ ���� ≤ ����Z − λ 2 v⋆ ���� � , Z ∼ N �λ 2 v⋆, 1 − λ + λ2 2 � when V is known to be v∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For the two-point problem, V is known to be some v∗ in the interval [y2 − y1 − B, y2 − y1 + B].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Thus, the use of the conservative prediction random set S = � v∗∈[y2−y1−B, y2−y1+B] Sv∗ 6 provides valid inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For constructing confidence intervals, for example, the plausibility region � v⋆∈[y2−y1−B, y2−y1+B] � λ 2 v⋆ − z1−α/2 � 1 − λ + λ2 2 , λ 2 v⋆ + z1−α/2 � 1 − λ + λ2 2 � covers u⋆ 2 with probability at least 100(1 − α)%, for all λ ∈ [0, 1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Note that the length of the plausibility interval is given by � λ 2 (y2 − y1 + B) − λ 2 (y2 − y1 − B) � + 2z1−α/2 � 1 − λ + λ2 2 = λB + 2z1−α/2 � 1 − λ + λ2 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) Using the usual measure of efficiency in terms of interval length, we minimize (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) over λ ∈ [0, 1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This leads to the choice of λ: ˆλB = � � � � � � � 1 − B � 2z2 1−α/2 − B2 , if 0 ≤ B < z1−α/2;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 0, if B ≥ z1−α/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The corresponding widths are equal to B + � 2z2 1−α/2 − B2 and 2z1−α/2, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 B Width G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G Marginal Mixture Conditional Figure 2: Comparison of the widths of the plausibility intervals constructed for the assertion A(t) = {ϑ0 : [0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 1] → R | ϑ0(t) = √ t} ⊆ Θ1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1/2 using the three different approaches: Marginal IMs,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Partial Conditioning (“Mixture”),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' and Conservative Conditional IMs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The optimal value of ˆλB suggests a full conditional IM when B is small relative to the width of the target confidence interval or the confidence level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Figure 2 shows the numerical results of a simulation with 100 trials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We observe that the plausibility intervals obtained using partial conditioning are the narrowest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 4 The general n case For better understanding, we provide detailed investigation on the three-point case in Subsection 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The general n case follows as a straightforward generalization of the three-point problem and is summarized in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 7 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 The n = 3 case For the case of n = 3, we have three pairs of observations (ti, yi), i = 1, 2, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Without loss of generality, assume that t1 ≤ t2 ≤ t3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' In this case, we have the data generation model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' � � � � � y1 = ϑ0(t1) + u⋆ 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' y2 = ϑ0(t2) + u⋆ 2;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' y3 = ϑ0(t3) + u⋆ 3 (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) where u⋆ 1, u⋆ 2, u⋆ 3 represent unobserved realizations of U1, U2, U3 iid ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The problem of interest is to make a (marginal) inference on ϑ0(ti), i = 1, 2, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To proceed, we first find the relevant conditional distribu- tions in Subsection 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1, and then provide the optimal solution under a partial conditioning framework in Subsection 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 Conditional Distributions The system of pairwise differences from (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1), that is, � � � � � y2 − y1 = ϑ0(t2) − ϑ0(t1) + u⋆ 2 − u⋆ 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' y3 − y1 = ϑ0(t3) − ϑ0(t1) + u⋆ 3 − u⋆ 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' y3 − y2 = ϑ0(t3) − ϑ0(t2) + u⋆ 3 − u⋆ 2 (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) motivates us to introduce the potential conditioning auxiliary variables V21 = U2 − U1, V31 = U3 − U1, V32 = U3 − U2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Clearly, we have that (V21, V31, V32) ∼ N3 � � � � 0 0 0 � � , � � 2 1 −1 1 2 1 −1 1 2 � � � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' By the H¨older condition (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2), we have � � � � � |ϑ0(t2) − ϑ0(t1)| ≤ M|t2 − t1|γ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' |ϑ0(t3) − ϑ0(t1)| ≤ M|t3 − t1|γ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' |ϑ0(t3) − ϑ0(t2)| ≤ M|t3 − t2|γ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='3) For notational convenience, define Bij := M|ti − tj|γ, 1 ≤ i, j ≤ 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' From (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2) and (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='3) we obtain the observed constraints on the pairwise differences v⋆ i,: v⋆ ij = u⋆ i − u⋆ j = yi − yj − (ϑ0(ti) − ϑ0(tj)) ∈ [yi − yj − Bij, yi − yj + Bij], 1 ≤ j < i ≤ 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4) A complete characterization of the conditional distributions of the original auxiliary variables U1, U2, U3 given the new auxiliary variables V21, V31, V32 is provided in Propositions 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 and 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2, with proofs given in the appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Direct applications of these propositions yield the corresponding results with respect to the parameter ϑ ∈ Θ, which are summarized in Corollaries 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 and 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 8 Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For any fixed vij = ui − uj with i > j, the conditional distributions of U1, U2, U3 given Vij = vij take the following forms: U1|v21 ∼ N �u1 − u2 2 , 1 2 � , U1|v31 ∼ N �u1 − u3 2 , 1 2 � , U1|v32 ∼ N(0, 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5) U2|v21 ∼ N �u2 − u1 2 , 1 2 � , U2|v32 ∼ N �u2 − u3 2 , 1 2 � , U2|v31 ∼ N(0, 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6) U3|v31 ∼ N �u3 − u1 2 , 1 2 � , U3|v32 ∼ N �u3 − u2 2 , 1 2 � , U3|v21 ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='7) Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 yields the following results regarding the parameter ϑ ∈ Θ: ϑ(t1) = Y1 − U1 ∈ N �Y1 + Y2 2 ± B21 2 , 1 2 � � N �Y1 + Y3 2 ± B31 2 , 1 2 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8) ϑ(t2) = Y2 − U2 ∈ N �Y1 + Y2 2 ± B21 2 , 1 2 � � N �Y2 + Y3 2 ± B32 2 , 1 2 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='9) ϑ(t3) = Y3 − U3 ∈ N �Y1 + Y3 2 ± B31 2 , 1 2 � � N �Y2 + Y3 2 ± B32 2 , 1 2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='10) Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For any fixed vi1j1 = ui1 − uj1 and vi2j2 = ui2 − uj2 with i1 > j1 and i2 > j2, the conditional distributions of U1, U2, U3 given Vi1j1 = vi1j1 and Vi2j2 = vi2j2 take the following forms: U1|v21, v31, U1|v21, v32, U1|v31, v32 ∼ N �2u1 − u2 − u3 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='11) U2|v21, v31, U2|v21, v32, U2|v31, v32 ∼ N �2u2 − u1 − u3 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='12) U3|v21, v31, U3|v21, v32, U3|v31, v32 ∼ N �2u3 − u1 − u2 3 , 1 3 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='13) Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 yields the following results with respect to parameter ϑ ∈ Θ: ϑ(t1) = Y1 − U1 ∈ N �Y1 + Y2 + Y3 3 ± B21 + B31 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' ϑ(t2) = Y2 − U2 ∈ N �Y1 + Y2 + Y3 3 ± B21 + B32 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' ϑ(t3) = Y3 − U3 ∈ N �Y1 + Y2 + Y3 3 ± B31 + B32 3 , 1 3 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' These results can be restated more compactly as ϑ(ti) ∈ N � ¯Y ± ¯Bi, 1 3 � , i = 1, 2, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' where ¯Y = (Y1 + Y2 + Y3)/3, ¯Bi = (Bi1 + Bi2 + Bi3)/3, and we define Bii = 0 for i = 1, 2, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 9 Remark 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For the sake of completeness, we remark that the covariance matrices in (U1, V21, V31, V32) ∼ � � � � � ��� 0 0 0 0 � ��� , � ��� 1 −1 −1 0 −1 2 1 −1 −1 1 2 1 0 −1 1 2 � ��� � � � � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U2, V21, V31, V32) ∼ � � � � � ��� 0 0 0 0 � ��� , � ��� 1 1 0 −1 1 2 1 −1 0 1 2 1 −1 −1 1 2 � ��� � � � � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U3, V21, V31, V32) ∼ � � � � � ��� 0 0 0 0 � ��� , � ��� 1 0 1 1 0 2 1 −1 1 1 2 1 1 −1 1 2 � ��� � � � � , are all degenerate, due to the fact that V21, V31, V32 are linearly dependent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 Prediction via partial conditioning Recall that our goal is to predict u⋆ 2 to make a valid probabilistic inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Without loss of generality, assume that t2 − t1 ≤ t3 − t2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Motivated by the idea of partial conditioning for the two-point problem, we construct nested predictive random sets by introducing the partial regression of u⋆ 2 on pairwise differences: N � λ1 �u⋆ 2 − u⋆ 1 2 � + λ2 �2u⋆ 2 − u⋆ 1 − u⋆ 3 3 � , �λ1 2 + λ2 3 �2 + � 1 − λ1 2 − 2 3λ2 �2 + �λ2 3 �2� , (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='14) where the mixing proportions λ1, λ2 are functions of Bij satisfying λ1, λ2 ≥ 0 and λ1 + λ2 ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Note that the variance expression arises naturally from the fact that Var � U2 − λ1 2 (U2 − U1) − λ2 3 (2U2 − U1 − U3) � = �λ1 2 + λ2 3 �2 + � 1 − λ1 2 − 2 3λ2 �2 + �λ2 3 �2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To predict U2, we take the predictive random sets Sv⋆ 21,v⋆ 32 = � z : ����z − � λ1 �v⋆ 21 2 � + λ2 �v⋆ 21 − v⋆ 32 3 ������ ≤ ����Z − � λ1 �v⋆ 21 2 � + λ2 �v⋆ 21 − v⋆ 32 3 ������ � , where we recall that v⋆ ij = u⋆ i − u⋆ j = yi − yj − (ϑ0(ti) − ϑ0(tj)) ∈ [yi − yj − Bij, yi − yj + Bij], 1 ≤ j < i ≤ 3, and Z follows the Gaussian distribution (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The predictive random sets are clearly marginally valid, and thus the corresponding IMs are valid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Therefore, the we utilize the conservative predictive random set S = � v⋆ 21∈[y2−y1±B21] v⋆ 32∈[y3−y2±B32] Sv⋆ 21,v⋆ 32 For constructing confidence intervals, this suggests the plausibility region � v⋆ 21∈[y2−y1±B21] v⋆ 32∈[y3−y2±B32] � � � ��λ1 2 + λ2 3 � v⋆ 21 − �λ2 3 � v⋆ 32 � ± z1−α/2 ��λ1 2 + λ2 3 �2 + � 1 − λ1 2 − 2 3λ2 �2 + �λ2 3 �2 � � � , 10 which covers u⋆ 2 with probability at least 100(1 − α)%.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' It can be easily verified that the width of this plausibility region is given by 2 ��λ1 2 + λ2 3 � B21 + �λ2 3 � B32 � + 2z1−α/2 ��λ1 2 + λ2 3 �2 + � 1 − λ1 2 − 2 3λ2 �2 + �λ2 3 �2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The optimal mixing proportions λ1, λ2 are the solutions to the constrained optimization problem min λ1, λ2 ��λ1 2 + λ2 3 � B21 + �λ2 3 � B32 � + z1−α/2 ��λ1 2 + λ2 3 �2 + � 1 − λ1 2 − 2 3λ2 �2 + �λ2 3 �2 s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' λ1, λ2 ≥ 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' λ1 + λ2 ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='15) Solving optimization problem (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='15) analytically requires finding the roots of a set of quadratic equations, a process which can become quite burdensome.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Instead, we adopt numerical optimization methods such as BFGS to solve problem (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='15) in order to obtain the optimal mixing proportions which minimize the length of the plausibility interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Figure 3 shows the numerical results of a simulation with 500 trials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Again, we find that the plausibility intervals obtained using optimal mixture distributions are the narrowest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='B21 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Width ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GGG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G GGG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Marginal ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Mixture ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content="Cond'l (1 pt) " metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content="Cond'l (2 pts) " metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='(a) Width vs B12 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='B32 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Width ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='GG ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='G ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Marginal ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Mixture ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content="Cond'l (1 pt) " metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content="Cond'l (2 pts) " metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='(b) Width vs B23 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='Figure 3: Comparison of the widths of the plausibility intervals constructed for the assertion ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='A(t) = {ϑ0 : [0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 1] → R | ϑ0(t) = √ t} ⊆ Θ1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1/2 using the four different approaches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 The general n case In this section, we extend our previous discussions to the general case of n observations, and provide a generic method to construct valid and efficient pointwise plausibility intervals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Denote our sequence of observations by (ti, yi), i = 1, · · · , n, where it is not necessary to assume that the ti’s are sorted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' With the assumption σ = 1 made in Section 1, the association (1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1) can be written as yi = ϑ0(ti) + u⋆ i , i = 1, · · · , n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='16) 11 In the rest of this section, we outline the process of conducting inference on Ui for an arbitrary 1 ≤ i ≤ n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' As a preprocessing step, we sort the n observations in ascending order of their distance from the i-th observation, such that 0 = |ti − t(i) 0 | ≤ |ti − t(i) 1 | ≤ · · · ≤ |ti − t(i) n−1|, correspondingly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Denote by U (i) 0 , U (i) 1 , · · · , U (i) n−1 the corresponding U variables, and by y(i) 0 , y(i) 1 , · · · , y(i) n−1 the corresponding y observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Notice that U (i) 0 = Ui holds trivially.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To predict Ui, we can utilize the marginal distribution N(0, 1), and the conditional distributions Ui|ui − u(i) 1 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Ui|ui − u(i) 1 , ui − u(i) 2 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' · · · ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Ui|ui − u(i) 1 , ui − u(i) 2 , · · · , ui − u(i) n−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This motivates us to consider the partial regression of Ui on the pairwise differences of the U variables: N � � � n−1 � k=1 λk k + 1 � �k ui − k � j=1 u(i) j � � , � 1 − n−1 � k=1 λk k k + 1 �2 + n−1 � j=1 � � n−1 � k=j λk k + 1 � � 2� � � , (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='17) where the mixing proportions λ1, λ2, · · · , λn−1 are functions of Bij satisfying λ1, λ2, · · · , λn−1 ≥ 0 and �n−1 k=1 λk ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The variance expression in Equation (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='17) is obtained by routine algebraic operations: Var � � �Ui − n−1 � k=1 λk k + 1 � �k Ui − k � j=1 U (i) j � � � � � = � 1 − n−1 � k=1 λk k k + 1 �2 + Var � � � n−1 � k=1 � � λk k + 1 k � j=1 U (i) j � � � � � , (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='18) To deal with the second term on the right-hand-side of the last equation, we notice that the summation signs can be exchanged in the following manner: n−1 � k=1 � � λk k + 1 k � j=1 U (i) j � � = n−1 � k=1 k � j=1 λk k + 1U (i) j = n−1 � j=1 n−1 � k=1 λk k + 1U (i) j = n−1 � j=1 � � n−1 � k=j λk k + 1 � � U (i) j .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Therefore, we have Var � � � n−1 � k=1 � � λk k + 1 k � j=1 U (i) j � � � � � = Var � � � n−1 � j=1 � � n−1 � k=j λk k + 1 � � U (i) j � � � = n−1 � j=1 � � n−1 � k=j λk k + 1 � � 2 , where the last step follows from the independence of U (i) j , j = 1, · · · , n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Substituting this result into equation (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='18) yields the variance expression shown in equation (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' To predict Ui, we take the predictive random sets S{v(i) j :j=1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=',k} = � � �z : ������ z − n−1 � k=1 λk k + 1 k � j=1 v(i) j ������ ≤ ������ Z − n−1 � k=1 λk k + 1 k � j=1 v(i) j ������ � � � , Z ∼ (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='17), where v(i) j := ui − u(i) j = yi − y(i) j − (ϑ0(ti) − ϑ0(t(i) j )) ∈ [yi − y(i) j − B(i) j , yi − y(i) j + B(i) j ], with B(i) j := |t(i) j − ti|, 1 ≤ i, j ≤ n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For valid inference, we take the (conservative) predictive random sets S = � {v(i) j ∈ � yi−y(i) j ±B(i) j � ,j=1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=',k} S{v(i) j :j=1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=',k}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This predictive random set is clearly marginally valid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Let ∆λ = � � � � � � 1 − n−1 � k=1 λk k k + 1 �2 + n−1 � j=1 � � n−1 � k=j λk k + 1 � � 2 , 12 the standard deviation of the partial regression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The plausibility region � v(i) j ∈ � yi−y(i) j ±B(i) j � � � � n−1 � k=1 λk k + 1 k � j=1 v(i) j ± z1−α/2∆λ � � � , or equivalently, � � n−1 � k=1 λk k + 1 k � j=1 � yi − y(i) j − B(i) j � − z1−α/2∆λ, n−1 � k=1 λk k + 1 k � j=1 � yi − y(i) j + B(i) j � + z1−α/2∆λ � � , covers u⋆ i with probability at least 100(1 − α)%.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The width of the plausibility region is given by 2 n−1 � k=1 λk k + 1 k � j=1 B(i) j + 2z1−α/2∆λ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The optimal set of mixing proportions λ1, λ2, · · · , λn−1 are the solutions to the constrained optimization problem min λ1,λ2,··· ,λn−1 n−1 � k=1 λk k + 1 k � j=1 B(i) j + z1−α/2∆λ subject to λ1, λ2, · · · , λn−1 ≥ 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' n−1 � k=1 λk ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='19) Solving optimization problem (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='19) analytically requires finding the roots of a set of quadratic equations, a process which can become quite burdensome.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Instead, we adopt numerical optimization methods such as BFGS to solve problem (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='19) in order to obtain the optimal mixing proportions which minimize the width of the plausibility interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' This is demonstrated using numerical results in Figures 4, 5, and 6 for different configurations of n, M, and γ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 −2 −1 0 1 2 3 t y (a) n = 20, σ = 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 −2 −1 0 1 2 3 t y (b) n = 20, σ = 1/2 Figure 4: Visualization of the plausibility intervals constructed using the mixture distribution for the assertion A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = √ t} ⊆ Θ1,1/2 under various settings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 13 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 −3 −2 −1 0 1 2 t y (a) n = 15, M = 1/2, γ = 1/5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 −3 −2 −1 0 1 2 t y (b) n = 15, M = 1/2, γ = 1 Figure 5: Visualization of the plausibility intervals constructed using the mixture distribution for the assertion A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = 0} ⊆ Θ1,1/2 under various settings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 −3 −2 −1 0 1 2 t y (a) n = 10 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='0 −3 −2 −1 0 1 2 t y (b) n = 100 Figure 6: Visualization of the plausibility intervals constructed using the mixture distribution for the assertion A(t) = {ϑ0 : [0, 1] → R | ϑ0(t) = 0} ⊆ Θ1,1/2 for various number of observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 5 Discussion We have developed a partial conditioning method to extend IMs for valid and efficient inference about many- normal-means when the means are subject to constraints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' While the basic idea and method are developed for many-normal-means, they can be extended to handle many-binomial-means and many-Poisson-means with similiar constraints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Nevertheless, a more general mathematical theory deserves future development.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 14 Appendix: Proofs Proof of Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' For the predictive random set S = [−|U1|, |U1|], U1 ∼ N(0, 1), we have PS{S ̸∋ u⋆} = P{|U1| < |u⋆ 1|} = 2Φ(|u⋆ 1|) − 1, where Φ(·) denotes the standard Normal CDF.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' It follows that for each α ∈ (0, 1), PU1{PS{S ̸∋ U1} ≥ 1 − α} = PU1{2Φ(|U1|) − 1 ≥ 1 − α} = PU1{Φ(|U1|) ≥ 1 − α/2} = α, which verifies that (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='5) holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Thus, we have shown that the predictive random set S is valid, and Theo- rem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 implies that the IM previously defined is valid as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Proof of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The conditional distributions for U1 can be derived as follows: � � � � � � � � � � � � � � � � � � � (U1, V21) ∼ N2 �� 0 0 � , � 1 −1 −1 2 �� ⇒ U1|V21 = v21 ∼ N � −v21 2 , 1 2 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U1, V31) ∼ N2 �� 0 0 � , � 1 −1 −1 2 �� ⇒ U1|V31 = v31 ∼ N � −v31 2 , 1 2 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U1, V32) ∼ N2 �� 0 0 � , � 1 0 0 2 �� ⇒ U1|V32 = v32 ∼ N(0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The conditional distributions for U2 can be derived as follows: � � � � � � � � � � � � � � � � � � � (U2, V21) ∼ N2 �� 0 0 � , � 1 1 1 2 �� ⇒ U2|V21 = v21 ∼ N �v21 2 , 1 2 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U2, V31) ∼ N2 �� 0 0 � , � 1 0 0 2 �� ⇒ U2|V31 = v31 ∼ N(0, 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U2, V32) ∼ N2 �� 0 0 � , � 1 −1 −1 2 �� ⇒ U2|V32 = v32 ∼ N � −v32 2 , 1 2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The conditional distributions for U3 can be derived as follows: � � � � � � � � � � � � � � � � � � � (U3, V21) ∼ N2 �� 0 0 � , � 1 0 0 2 �� ⇒ U3|V21 = v21 ∼ N(0, 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U3, V31) ∼ N2 �� 0 0 � , � 1 1 1 2 �� ⇒ U3|V31 = v31 ∼ N �v31 2 , 1 2 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U3, V32) ∼ N2 �� 0 0 � , � 1 1 1 2 �� ⇒ U3|V32 = v32 ∼ N �v32 2 , 1 2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The proof is complete by recalling that vij = ui − uj for 1 ≤ j < i ≤ 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 15 Proof of Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We notice that the results in (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4) imply the following: u1 − u2 = −v21 ∈ [Y1 − Y2 − B21, Y1 − Y2 + B21], u1 − u3 = −v31 ∈ [Y1 − Y3 − B31, Y1 − Y3 + B31];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' u2 − u1 = v21 ∈ [Y2 − Y1 − B21, Y2 − Y1 + B21], u2 − u3 = −v32 ∈ [Y2 − Y3 − B32, Y2 − Y3 + B32];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' u3 − u1 = v31 ∈ [Y3 − Y1 − B31, Y3 − Y1 + B31], u3 − u2 = v32 ∈ [Y3 − Y2 − B32, Y3 − Y2 + B32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Combining these results with Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='1 completes the proof of the corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Proof of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The conditional distributions for U1 can be derived as follows: � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � (U1, V21, V31) ∼ N3 � � � � 0 0 0 � � , � � 1 −1 −1 −1 2 1 −1 1 2 � � � � ⇒ U1|v21, v31 ∼ N � −v21 + v31 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U1, V21, V32) ∼ N3 � � � � 0 0 0 � � , � � 1 −1 0 −1 2 −1 0 −1 2 � � � � ⇒ U1|v21, v32 ∼ N � −2v21 + v32 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U1, V31, V32) ∼ N3 � � � � 0 0 0 � � , � � 1 −1 0 −1 2 1 0 1 2 � � � � ⇒ U1|v31, v32 ∼ N �v32 − 2v31 3 , 1 3 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The conditional distributions for U2 can be derived as follows: � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � (U2, V21, V31) ∼ N3 � � � � 0 0 0 � � , � � 1 1 0 1 2 1 0 1 2 � � � � ⇒ U2|v21, v31 ∼ N �2v21 − v31 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U2, V21, V32) ∼ N3 � � � � 0 0 0 � � , � � 1 1 −1 1 2 −1 −1 −1 2 � � � � ⇒ U2|v21, v32 ∼ N �v21 − v32 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U2, V31, V32) ∼ N3 � � � � 0 0 0 � � , � � 1 0 −1 0 2 1 −1 1 2 � � � � ⇒ U2|v31, v32 ∼ N �v31 − 2v32 3 , 1 3 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' The conditional distributions for U3 can be derived as follows: � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � (U3, V21, V31) ∼ N3 � � � � 0 0 0 � � , � � 1 0 1 0 2 1 1 1 2 � � � � ⇒ U3|v21, v31 ∼ N �2v31 − v21 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U3, V21, V32) ∼ N3 � � � � 0 0 0 � � , � � 1 0 1 0 2 −1 1 −1 2 � � � � ⇒ U3|v21, v32 ∼ N �v21 + 2v32 3 , 1 3 � ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' (U3, V31, V32) ∼ N3 � � � � 0 0 0 � � , � � 1 1 1 1 2 1 1 1 2 � � � � ⇒ U3|v31, v32 ∼ N �v31 + v32 3 , 1 3 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 16 It can be easily verified that the three conditional distributions in each group are actually equivalent, and that the results (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='11) – (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='13) hold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Proof of Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' We notice that the results in (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='4) implies the following: 2u1 − u2 − u3 = −(v21 + v31) ∈ [2Y1 − Y2 − Y3 − (B21 + B31), 2Y1 − Y2 − Y3 + (B21 + B31)];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 2u2 − u1 − u3 = v21 − v32 ∈ [2Y2 − Y1 − Y3 − (B21 + B32), 2Y2 − Y1 − Y3 + (B21 + B32)];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 2u3 − u1 − u2 = v31 + v32 ∈ [2Y3 − Y1 − Y2 − (B31 + B32), 2Y1 − Y2 − Y3 + (B31 + B32)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Combining these results with Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='2 completes the proof of the corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' References R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Gong and X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content='-L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Meng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Judicious judgment meets unsettling updating: Dilation, sure loss and Simpson’s paradox.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Statistical Science, 36(2):169–190, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Leaf and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Inference about constrained parameters using the elastic belief method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' International Journal of Approximate Reasoning, 53(5):709–727, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Lepskii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' On a problem of adaptive estimation in gaussian white noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Theory of Probability & Its Applications, 35(3):454–466, 1991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Martin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Frameworks for prior-free posterior probabilistic inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Wiley Interdisciplinary Reviews: Computational Statistics, 7(1):77–85, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Martin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Comment: Settle the unsettling: An inferential models perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Statistical Science, 36(2):196–200, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Martin and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Inferential models: A framework for prior-free posterior probabilistic inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Journal of the American Statistical Association, 108(501):301–313, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Martin and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Conditional inferential models: combining information for prior-free probabilistic inference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Journal of the Royal Statistical Society: Series B (Statistical Methodology), 77(1):195–217, 2015a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Martin and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Liu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Marginal inferential models: prior-free probabilistic inference on interest parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' Journal of the American Statistical Association, 110(512):1621–1631, 2015b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} +page_content=' 17' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VtE3T4oBgHgl3EQfbAqu/content/2301.04512v1.pdf'} diff --git a/W9E3T4oBgHgl3EQf1AvA/content/tmp_files/2301.04743v1.pdf.txt b/W9E3T4oBgHgl3EQf1AvA/content/tmp_files/2301.04743v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..198bf6833b229b3a38110c09e38e3e4a8214540f --- /dev/null +++ b/W9E3T4oBgHgl3EQf1AvA/content/tmp_files/2301.04743v1.pdf.txt @@ -0,0 +1,735 @@ +Analysis of Interior Rubble Void Spaces at Champlain Towers South +Collapse +Ananya Rao1, Robin Murphy2, David Merrick3, and Howie Choset1 +Abstract— The 2021 Champlain Towers South Condomini- +ums collapse in Surfside, Florida, resulted 98 deaths. Nine +people are thought to have survived the initial collapse, and +might have been rescued if rescue workers could have located +them. Perhaps, if rescue workers had been able to use robots to +search the interior of the rubble pile, outcomes might have been +better. An improved understanding of the environment in which +a robot would have to operate to be able to search the interior of +a rubble pile would help roboticists develop better suited robotic +platforms and control strategies. To this end, this work offers an +approach to characterize and visualize the interior of a rubble +pile and conduct a preliminary analysis of the occurrence of +voids. Specifically, the analysis makes opportunistic use of four +days of aerial imagery gathered from responders at Surfside +to create a 3D volumetric aggregated model of the collapse in +order to identify and characterize void spaces in the interior +of the rubble. The preliminary results confirm expectations of +small number and scale of these interior voids. The results +can inform better selection and control of existing robots for +disaster response, aid in determining the design specifications +(specifically scale and form factor), and improve control of +future robotic platforms developed for search operations in +rubble. +I. INTRODUCTION +Improving the response to structural collapse disasters is +a major focus of search and rescue robotics research. Due +to the size and accessibility of areas to search, searching +the interior of a rubble pile presents different challenges +than searching the standing or partially damaged portions +of a structure, and thus requires a different set of tools +than what is conventionally available today. We believe +that robots should be in such a toolbox, and searching +rubble piles necessitates different sizes and form factors of +robots. This work focuses on the collapse of the Champlain +Towers South Condominium in June 2021, which is the third +largest structural collapse in the United States [1]. Of the 98 +resulting deaths, nine were deemed to not be caused by crush +injuries, indicating that these nine victims might have been +rescued if they had been quickly found and extricated from +the rubble [2], [3]. +Rescue workers may be able to use uncrewed ground +vehicles (UGVs) to search the interior of a rubble pile at +a disaster site to help locate survivors, as robots can be +sent into areas that cannot be accessed by human or canine +*Portions of this work were supported by NSF Grant CMMI-2140528. +Any opinions, findings, and conclusions or recommendations expressed in +this material are those of the author(s) and do not necessarily reflect the +views of the National Science Foundation. +1A. Rao and H. Choset are with the Robotics Institute, Carnegie Mellon +University, Pittsburgh, PA, USA ananyara@andrew.cmu.edu +2R. Murphy is with Texas A&M University, College Station, TX, USA +3D. Merrick is with Florida State University, Tallahassee, FL, USA +rescue workers. UGVs have been successfully used at the +World Trade Center disaster [4], and the Berkman Plaza II +collapse [5], [6], as well as in post-earthquake relief efforts in +Mexico City [7] and Christchurch [5]. However, despite this, +UGVs were not used for searching for survivors at Surfside. +One possible reason for the lack of UGVs is that UGVs +are not designed to search the interior of rubble. Prior work +[5] has attempted to characterize the interior of rubble as +a work envelope, but that characterization has been based +on inferences from what robots to date have already en- +countered. Structural engineers have not been able to predict +the occurrence of voids in ways that help with robot design +specifications. Therefore, an improved understanding of the +number, locations, and characteristics of void spaces in +disaster sites is needed to inform the development of better +suited robotic search and rescue platforms. +This work provides a data-driven analysis of void spaces +in the Surfside rubble, with the goal of characterizing and +visualizing interior voids in order to inform future work. The +data used in this work was collected by flying an uncrewed +aerial system (UAS) over the disaster site at Surfside in a +grid-like flight pattern. Aerial RGB and thermal images, and +oblique imagery, were collected by responders for use in +strategic planning on-site. Multiple flights were conducted +each day for the duration of the response, with the first flight +occurring on June 25, 2021 at 16:30 local time. The images +collected were processed using Structure from Motion soft- +ware (specifically Agisoft Metashape Professional) to create +dense 3D point clouds of the disaster site (see Fig 1). Agisoft +Metashape was also used to generate orthophoto mosaics, +which are raster images made by merging orthophotos (aerial +or satellite photographs which have been transformed to +correct for perspective so that they appear to have been taken +from vertically above at an infinite distance) of the collapse. +This work focuses on data collected on four separate days +(June 27-30, 2021) at similar times of day. This limited +subset of the available data is used as a proof of concept +of using the approach presented in this work to locate and +characterize void spaces. +In this work, the point clouds generated from data col- +lected in four different UAS flights are used to identify +and characterize voids in interior rubble, with the aim of +better understanding the environments in which rescue UGVs +must operate. The constructed point clouds are manually +aligned, and the resulting point cloud stack is systematically +investigated to identify and locate potential void spaces. +The potential voids and related oblique imagery are then +evaluated to determine scale and cause of the voids. This +arXiv:2301.04743v1 [cs.RO] 11 Jan 2023 + +Fig. 1. +Aerial imagery collected to support the Surfside disaster response [8] was processed to create an aggregated 3D model of the disaster site in the +form of a point cloud stack (right), which was then analyzed to find and characterize voids. For contextualizing this 3D volumetric model, an aerial view +of the Surfside collapse site [9] is shown (left). +approach allows roboticists to see voids in rubble, and offers +evidence of their location, frequency of occurrence, and size. +The paper is organized as follows. Prior work in disaster +site analysis, the use of ground robots at other disaster +responses, and void detection in 3D point clouds is discussed +in Section II. Our approach to locating and characterizing +interior rubble voids is presented in Section III. Section IV +contains observations made, and analysis of identified voids. +Finally, conclusions and directions for future work are pre- +sented in Section V. +II. BACKGROUND AND PRIOR WORK +Prior work has attempted to characterize voids in rubble +suitable for robotics, either through sensing with ground +penetrating radar of simulated voids or from inference from +cumulative studies; this work attempts to directly detect +and characterize voids in an actual collapse using aerial +imagery. Prior work has also considered detecting voids in +3D point clouds, however those efforts have focused on more +structured point clouds of synthetic data with smaller number +of points. This work addresses large, complex and noisy 3D +point clouds generated from real-world imagery. +A. Analysis of Voids in Rubble +Prior works have attempted to infer void characteristics +from technical search tools, or to characterize in situ voids +with ground penetrating radar [10]. However, these efforts +have been conducted in low fidelity physical simulations, +and not with actual voids. Further, ground penetrating radar +suffers performance degradation in heterogeneous condi- +tions, due to signal scattering, thereby leading to limited +applicability in the mixed-material environment of a rubble +pile at a disaster site. +There is also analysis of building collapses from a civil +engineering perspective. For example, for the Surfside col- +lapse, the nature and location of the rubble pile was analyzed +in order to estimate the cause of the collapse, and some +associated structural information [11]. From this analysis, a +simulation of the collapse can also be created. However, such +analyses do not consider the interior properties of the rubble +pile itself, and do not provide information on what voids may +have existed in the rubble. +The most comprehensive formal definition of the charac- +teristics of rubble for robots appears in Disaster Robotics [5] +and related articles [20], [21]. The set of attributes presented +in Disaster Robotics helps characterize the interior rubble +as a robot work envelope, which is useful for selecting +and deploying robots at a disaster site, as well as for +developing new robotics rescue technologies. There are +eight attributes: 1.scale, 2.verticality, 3.tortuosity, 4.surface +properties, 5.severity of obstacles, 6.accessibility elements, +7.number of regions, and 8.non-navigational constraints, with +2-7 being collectively referred to as traversability. These +characteristics overlap with, and go beyond, two other sets +of attributes developed for ground robot navigation, see [18], +[19]. This work focuses on determining scale and number of +regions. +B. Void Detection in 3D Point Clouds +There exist several methods for detecting voids in point +clouds. One method has the user manually detect point +cloud voids, and then is able to fill these point cloud +voids in geometrically complex areas [12]. This approach +however cannot process both smooth and sharp (or low and +high frequency geometry) voids at the same time, as such +properties are encoded into the parameters that are selected +at run-time. Other methods create polygonal meshes from +input point clouds using a Delaunay Triangulation algorithm, +and then either perform a set of angle checks in order to +determine which vertices are on the boundary of a void [13], +or conduct a data search on the mesh to determine which +vertices are present only in a single polygon, and thus lie on + +6/27 +6/28 +6/29 +6/30 +51.2 +38.4 +(w) +N 25.6 +25.6 +12.8 +51.2 +76.8 +102.4 +x (m) +25.6 +51.2 +y (m)the void boundary [14]. Another work employs Ball Pivoting +mesh generation [15] to improve void detection. +While these methods have shown results in automatically +detecting voids in 3D point clouds, they do not perform +well on highly unstructured data, and point clouds that are +missing sections, such as that of a rubble pile. The approach +presented in this work uses systematic manual inspection, as +the focus is on characterizing the voids found, and not yet +on automating void characterization. +III. LOCATING AND CHARACTERIZING VOID SPACES +In this work, a systematic manual approach is used to +find, locate and characterize void spaces in the rubble pile of +the Surfside building collapse. The approach can be broadly +divided into two high-level steps: locating potential voids +spaces, and characterizing the identified voids. +A. Potential Void Space Identification +Finding potential void spaces involves processing the data +gathered at Surfside, and then manually looking for voids in +the following steps. +1) Aerial images of the disaster site were collected from +flying an uncrewed aerial system (UAS) to support +response efforts [8]. +2) Agisoft Metashape was used to process the collected +data into dense 3D point clouds of the disaster site. +3) The 3D point clouds were manually aligned and +“stacked“ to create an aggregated 3D volumetric rep- +resentation of the collapse (see Fig 1). +4) Cross sections were taken of the 3D volumetric model +of the rubble pile to be able to look for potential voids. +The cross sections were taken on a 4m x 4m grid, as +depicted in Fig 2. +The UAS flights over the Surfside site were conducted +at several times each day during the response and recovery +to map the disaster site [8]. This analysis opportunistically +uses the imagery from one mapping flight at roughly the +same time each day for four days: June 27, 2021 13:30, June +28, 2021 13:30, June 29, 2021 13:00, and June 30, 2021 +11:00 local time. These four days were chosen because it +was a manageable set of data and it spanned days where the +activity on the rubble pile was limited. Starting on July 1, +2021, operations increased and more people and equipment +began to cover the surface of the rubble pile distorting the +construction of point clouds. +B. Void Space Characterization +Once potential void spaces have been identified, each one +was characterized, specifically in terms of size and scale +Additionally, each potential void space was analyzed to +distinguish between naturally occurring voids and artifacts +of excavation. The steps for analyzing each potential void +space are as follows. +1) The centroid location of the potential void was esti- +mated, and then plotted on each of the four orthophoto +mosaics of the Surfside site being considered in this +work (June 27 - 30, 2021). The orthophoto mosaics +Fig. 2. +In order to systematically look for void spaces in the interior of the +rubble pile, “slices“ or cross sectional views are taken into the aggregated +3D volumetric model, or point cloud stack, to examine cross sections of the +rubble pile on a 4m x 4m grid. The slices are 1m x 64m in the XY plane, +and have the depth of the entire rubble pile. Here the slices are depicted +on the orthophoto mosaic constructed from images gathered during a single +UAS flight on June 27, 2021. +were generated using Agisoft Metashape, with one +orthophoto mosaic for each UAS flight. +2) Dimensions of the potential void were estimated by +inspecting the XZ plane and YZ plane cross sections +of the void space. Here, the Z axis points up. +3) The estimated dimensions were used to approximate +the size and volume of the potential void space. +4) The orthophoto mosaics, and oblique imagery of the +potential void, were inspected for excavation activity, +and compared to void dimensions and shape seen in +the cross sectional views in order to determine if the +potential void occurred naturally or was created by +excavation. +IV. RESULTS AND DISCUSSION +To frame our observations, the effort described in this +paper focuses on characterizing voids, which will help estab- +lish the necessary form factor and perceptual constraints for +search and rescue robots operating in the interior of rubble. +This work represents the first step: finding and confirming +naturally occurring voids in the rubble pile. +Consistent +with +hypotheses +formed +from +Disaster +Robotics, Very few potential voids are seen in the rubble pile. +“Slicing“ the rubble pile model (i.e. inspecting cross sections +of the 3D volumetric model) on a 4m x 4m grid produced +only six candidate voids (depicted in Fig 3) out of 225 +possible locations. The candidate voids average a maximum +height of 1.46m, a minimum height of 0.26m, and a cross +sectional width of 4.89m. A summary of their characteristics +is presented in Table I. While the voids themselves are not +large enough to be survivable, they could offer breathing +space, and show the types of void spaces that a robot would +need to reach, inspect, and transit through. +The layers of the aggregated volumetric model constructed +did appear to match the oblique imagery, implying that the +layers are reasonably correct, and capture large objects and +the “stair steps“ of the layers of a pancake collapse. An +example of this is seen in Fig 4. + +10 +20 +30 +40 +(w) ^ +09 +60 +70 +- +80 +06 +100 +20 +40 +60 +80 +100 +120 +140 +160 +180 +x (m)Void +Approximate +Maximum +Minimum +XZ Plane Cross +YZ Plane Cross +Cause +Volume (m3) +Height (m) +Height (m) +Sectional Width (m) +Sectional Width (m) +Yellow +29.10 +2.28 +0.37 +4.23 +2.87 +Excavation +Cyan +34.53 +1.88 +0.33 +4.25 +6.25 +Excavation +Orange +41.71 +1.45 +0.15 +5.03 +6.91 +Excavation +Purple +28.71 +1.19 +0.20 +6.37 +5.06 +Excavation +Pink +10.94 +1.14 +0.26 +3.41 +5.73 +Naturally Formed +Green +10.33 +0.82 +0.23 +2.69 +5.91 +Naturally Formed +TABLE I +SUMMARY OF CHARACTERISTICS OF IDENTIFIED VOID SPACES IN THE INTERIOR OF THE SURFSIDE RUBBLE PILE. +Fig. 3. +The locations and estimated XY plane bounding boxes of the six identified voids are overlaid on the orthophoto mosaic from June 27, 2021. +Our observations also support the hypothesis that robots +deployed to similar disaster sites would need to be able +to burrow into the rubble in order to be effective (this is +consistent with Disaster Robotics). The voids seen in the +Surfside collapse do not appear to form a connected set of +tunnels. The voids also did not appear to have entrances from +the surface of the rubble. A UGV would thus need to be able +to burrow into the rubble to both access the voids, and also +to traverse between them. +Further, it is difficult to determine which voids are nat- +urally occurring and which are artifacts of excavation. By +examining oblique imagery from throughout the four days, +four out of six of the candidate voids appeared to be the +result of excavation (see Fig 5). Oblique imagery from June +27, 2021 and June 28, 2021 show high excavation activity in +the regions of the yellow, cyan, orange, and purple potential +voids. This leads us to infer that the void space seen between +the June 27 and June 28 layers is the result of excavation. +Two of the six voids, specifically the pink and green voids +in Fig 3, were in the pancake section of the collapse, and the +associated oblique imagery suggests naturally forming voids +were revealed by removing slabs. +A layer of material is visibly removed between the June +27th and June 28th (see Fig 6), however, the void seen +between the corresponding layers of the digital surface model +is larger than would be expected from excavation alone, +based on the estimated thickness of the slabs removed. This +implied that a naturally formed void space was underneath +the layer of slabs that were removed. +Similarly, slabs were removed from the pink void region +between June 27th and June 28th, with a void space that is +larger than the estimated thickness of the slabs (see Fig 7). +In both the green and pink voids, slabs act as a “ceiling“ of +sorts for the void space underneath. +Analysis of the voids shows that they may require robots +with smaller form factors and more flexibility, which is +consistent with predictions made in Disaster Robotics 2014. +The six voids varied in maximum height from 0.82m to +2.28m, minimum height from 0.15m to 0.37m, and cross +sectional width from 2.69m to 6.91m. The two voids in +the pancake section (pink and green) were on the order of +0.25m - 1.14m in height. While not survivable, these voids do +indicate the size constraints for a robot to be able to penetrate + +10 +20 +30 +40 +60 +70 +80 +06 +100 +20 +40 +60 +80 +100 +120 +140 +160 +180 +x (m)Fig. 4. +By inspecting oblique imagery of the purple void (right), and relating this to a cross sectional view of the purple void (left), it is seen that large +objects and features are reasonably recognizable and consistent between the imagery and the layers of the constructed 3D volumetric model. +Fig. 5. +High levels of activity are seen in the oblique imagery from June +27, 2021 (left), and June 28, 2021 (right) in the region of the yellow, cyan, +orange, and purple voids. It can be inferred from this that these potential +voids are really artifacts of excavation. +Fig. 6. +Some material was removed in the green void region between June +27, 2021 (top left) and June 28, 2021 (top right). There appears to be some +slabs removed, which is consistent with the shape of the top of the void +space seen in the XZ plane cross sectional view of the green void (bottom). +This implies that the green void was naturally formed during the collapse. +Fig. 7. +Some material was removed in the pink void region between June +27, 2021 (top left) and June 28, 2021 (top right). There appears to be some +slabs removed, which is consistent with the shape of the top of the void +space seen in the XZ plane cross sectional view of the pink void (bottom). +This implies that the pink void was naturally formed during the collapse. +and navigate into the void, and search for bigger voids. They +also provide insight into the form factors best suited for +exploring these voids - smaller, more flexible robots, like +snake robots, are best suited. +Candidate voids at Surfside did not show the verticality +encountered in void spaces seen at the World Trade Center, +parking garages, and mine disasters, however the depth of the +rubble pile reinforces the notion that robots have to penetrate +more than 6m into the rubble to reach areas where there +might be survivable voids. +V. CONCLUSIONS +This preliminary work establishes a promising approach to +characterizing and visualising the interior of rubble that not +only allows for formally characterizing voids as robot work +envelopes, which can aid in the design of effective robots, but + +6/27 +6/28 +10.24 +6/29 +6/30 +8.96, +(w) z +7.68 +6.4 +40.96 +42.24 +43.52 +44.8 +46.08 +47.36 +48.64 +49.92 +51.2 +y (m)10.24 +6/27 +6/28 +6/29 +6/30 +8.96 +N +6.4 +5.12 +61.44 +62.72 +64 +65.28 +66.56 +67.84 +69.12 +70.4 +x (m)6/27 +6/28 +6/29 +6/30 +8.96 +w) +N +58.88 +61.44 +64 +66.56 +69.12 +x (m)also offers data that could assist civil engineers in predicting +where voids are likely to form. Further, this work could help +rescue workers focus search efforts on areas where voids +are likely. The approach enables void spaces in rubble to +be visualized for the first time by using a 3D volumetric +aggregated model of the entire collapse that can be examined +as cross-sections or “fly throughs.” +However, further work is needed. The results showed an +average translation error in point cloud alignment of about +80 cm, even though it is possible to detect interesting void +spaces from point clouds, improvements would need to be +made in order to more accurately characterize the dimensions +of these voids. Development of a tool to automatically +register and align point clouds would help reduce the trans- +lation error. Automatic point cloud alignment on this data +is challenging due to it’s complex and dynamic nature, and +an analysis of existing methods for alignment supports the +need for additional work in this space [16]. This work also +only considers four days of data (27 June, 2021 - 30 June, +2021), which correspond to the “top“ layers of the collapse. +More work is needed to cover all of the days of data, but +activity significantly increased on the site after 30 June 2021, +so while voids may appear deeper in the rubble pile, it may +be harder to distinguish between naturally occurring voids +and artifacts of excavation. +Further, this is an opportunistic post hoc study; the aerial +imagery was collected for strategic planning for the on-the- +ground response and has notable gaps in frequency of data +collection and viewpoints that would help ascertain voids. +The data was also restricted to RGB images. This approach +can be improved both with dedicated data collection at higher +frequency of UAS flights and from different angles, and also +by using additional sensors, such as Lidar. More frequent +mapping imagery and obliques, or the use of Lidar, would +likely support automated registration of point cloud layers. +However, such data collection may interfere with the search +and rescue priorities at a disaster site, and therefore, there +is no guarantee that better data can be obtained, and most +certainly, we do not want to interfere with the work flow of +the rescue workers, as saving lives is the highest priority. +Despite these limitations, the preliminary results presented +in this work confirm expectations of the small number +and small scale of void spaces in rubble. These results +can be used for better selection and control of existing +robotic platforms for a disaster response. The purpose of +this paper was to show that identifying the existence of +voids is possible. The next step is then to provide in-depth +analysis of the these voids, to both better characterize the +rubble pile for possible locations of survivors, as well as +understand how real rubble piles impact robot design. Such +design considerations include size, number of degrees of +freedom, payload, sensory needs (especially for confined +spaces) and compliance in the mechanism. Compliance is +important because the rubble pile is highly unstructured, and +thus will have an abundance of unmodeled features. While +this paper does not provide a detailed analysis of identified +voids, further analysis will explore small size and actuation, +structure or robot and payload, compliance due to lack of +structure, and perception required to explore these voids. The +complexity of the rubble with the different types of collapse +modes may favor a modular robotics approach in order to +adapt to the different types of confined space environments +in different collapse areas. +ACKNOWLEDGMENT +The authors thank Ananya Bal and Dr. Robert Ladig for +their help with the data analysis and thoughtful discussions. +REFERENCES +[1] K. +Barnett. +Surfside +condo +collapse +is +third +largest +building +failure +in +country’s +history. +CBS +Miami, +2021. +https://miami.cbslocal.com/2021/06/29/surfside-condo-collapse- +third-largest-building-failure-us-history/. +[2] W. +Rhodes +and +R. +Ruiz-Goiriena. +At +least +9 +victims +of +98 +dead +may +have +initially +survived +surfside +condo +collapse, +but +were +not +found +by +rescue +teams, +investigation +shows. +USA +Today, +Aug. +24, +2021; +Mar +14, +2022. +https://www.usatoday.com/story/news/nation/2021/08/24/surfside- +condo-collapse-there-may-have-been-more-survivors- +florida/8152538002/. +[3] S. +Thaler. +Nine +of +the +98 +surfside +victims +may +have +survived +the +initial +condo +collapse: +Fire +rescue +logs +show +woman was alive in rubble for ten hours. +UK Daily Mail, +2021. +https://www.dailymail.co.uk/news/article-9924083/At-9-98- +Surfside-victims-SURVIVED-initial-condo-collapse.html. +[4] R. R. Murphy. Trial by fire. IEEE Robotics and Automation Magazine, +11(3):50–61, 2004. +[5] R. R. Murphy. Disaster robotics. MIT Press, 2014. +[6] S. Tadokoro, R. Murphy, S. Stover, W. Brack, M. Konyo, T. Nishimura, +and O. Tanimoto. +Application of active scope camera to forensic +investigation of construction accident. +In 2009 IEEE Workshop on +Advanced Robotics and its Social Impacts, pages 47–50, 2009. +[7] J. Whitman, N. Zevallos, M. Travers, and H. Choset. Snake robot +urban search after the 2017 mexico city earthquake. In 2018 IEEE +International Symposium on Safety, Security, and Rescue Robotics +(SSRR), pages 1–6, 2018. +[8] R. Murphy, D. Merrick, J. Adams, J. Broder, A. Bush, L. Hart, and +R. Hawkins. How robots helped out after the surfside condo collapse: +Responders flew drones night and day to survey the collapse and search +for survivors. IEEE Spectrum, 2021. +[9] E. +C. +McLaughlin. +What +we +know +about +the +building +that +partially +collapsed +in +surfside, +florida. +CNN, +2021. +https://www.cnn.com/2021/06/24/us/what-to-know-about-the-surfside- +building-collapse/index.html. +[10] D. Hu, S. Li, J. Chen, and V. R. Kamat. +Detecting, locating, and +characterizing voids in disaster rubble for search and rescue. In ” +Advanced Engineering Informatics, volume 111, pages 109–124, 2019. +[11] X. Lu, H. Guan, H. Sun, Y. Li, Z. Zheng, Y. Fei, and L. Zuo. +A preliminary analysis and discussion of the condominium building +collapse in surfside, florida, us, june 24, 2021. Frontiers of Structural +and Civil Engineering, pages 1097–1110, 2021. +[12] L. Yang, Q. Yan, and C. Xiao. Shape-controllable geometry comple- +tion for point cloud models. The Visual Computer, 33(3):385–398, +2017. +[13] T. K. Dey and J. Giesen. +Detecting undersampling in surface +reconstruction. Discrete and Computational Geometry, 15:329–345, +2003. +[14] Y. Jun. +A piecewise hole filling algorithm in reverse engineering. +Computer-Aided Design, 37(2):263–270, 2005. +[15] B. Bird, B. Lennox, S. Watson, and T. Wright. +Autonomous void +detection and characterisation in point clouds and triangular meshes. +International Journal of Computational Vision and Robotics, 9(4):368 +– 386, August 2019. +[16] A. Bal, R. Ladig, P. Goyal, J. Galeotti, H. Choset, D. Merrick, and +R. Murphy. A comparison of point cloud registration techniques for +on-site disaster data from the surfside structural collapse. 2022. + diff --git a/W9E3T4oBgHgl3EQf1AvA/content/tmp_files/load_file.txt b/W9E3T4oBgHgl3EQf1AvA/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..2716db3afe733a04a65baeaaab2af0f7353ad7d0 --- /dev/null +++ b/W9E3T4oBgHgl3EQf1AvA/content/tmp_files/load_file.txt @@ -0,0 +1,395 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf,len=394 +page_content='Analysis of Interior Rubble Void Spaces at Champlain Towers South Collapse Ananya Rao1, Robin Murphy2, David Merrick3, and Howie Choset1 Abstract— The 2021 Champlain Towers South Condomini- ums collapse in Surfside, Florida, resulted 98 deaths.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Nine people are thought to have survived the initial collapse, and might have been rescued if rescue workers could have located them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Perhaps, if rescue workers had been able to use robots to search the interior of the rubble pile, outcomes might have been better.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' An improved understanding of the environment in which a robot would have to operate to be able to search the interior of a rubble pile would help roboticists develop better suited robotic platforms and control strategies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' To this end, this work offers an approach to characterize and visualize the interior of a rubble pile and conduct a preliminary analysis of the occurrence of voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Specifically, the analysis makes opportunistic use of four days of aerial imagery gathered from responders at Surfside to create a 3D volumetric aggregated model of the collapse in order to identify and characterize void spaces in the interior of the rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The preliminary results confirm expectations of small number and scale of these interior voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The results can inform better selection and control of existing robots for disaster response, aid in determining the design specifications (specifically scale and form factor), and improve control of future robotic platforms developed for search operations in rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' INTRODUCTION Improving the response to structural collapse disasters is a major focus of search and rescue robotics research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Due to the size and accessibility of areas to search, searching the interior of a rubble pile presents different challenges than searching the standing or partially damaged portions of a structure, and thus requires a different set of tools than what is conventionally available today.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' We believe that robots should be in such a toolbox, and searching rubble piles necessitates different sizes and form factors of robots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work focuses on the collapse of the Champlain Towers South Condominium in June 2021, which is the third largest structural collapse in the United States [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Of the 98 resulting deaths, nine were deemed to not be caused by crush injuries, indicating that these nine victims might have been rescued if they had been quickly found and extricated from the rubble [2], [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Rescue workers may be able to use uncrewed ground vehicles (UGVs) to search the interior of a rubble pile at a disaster site to help locate survivors, as robots can be sent into areas that cannot be accessed by human or canine Portions of this work were supported by NSF Grant CMMI-2140528.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 1A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Rao and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Choset are with the Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA ananyara@andrew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='cmu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='edu 2R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Murphy is with Texas A&M University, College Station, TX, USA 3D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Merrick is with Florida State University, Tallahassee, FL, USA rescue workers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' UGVs have been successfully used at the World Trade Center disaster [4], and the Berkman Plaza II collapse [5], [6], as well as in post-earthquake relief efforts in Mexico City [7] and Christchurch [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' However, despite this, UGVs were not used for searching for survivors at Surfside.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' One possible reason for the lack of UGVs is that UGVs are not designed to search the interior of rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Prior work [5] has attempted to characterize the interior of rubble as a work envelope, but that characterization has been based on inferences from what robots to date have already en- countered.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Structural engineers have not been able to predict the occurrence of voids in ways that help with robot design specifications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Therefore, an improved understanding of the number, locations, and characteristics of void spaces in disaster sites is needed to inform the development of better suited robotic search and rescue platforms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work provides a data-driven analysis of void spaces in the Surfside rubble, with the goal of characterizing and visualizing interior voids in order to inform future work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The data used in this work was collected by flying an uncrewed aerial system (UAS) over the disaster site at Surfside in a grid-like flight pattern.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Aerial RGB and thermal images, and oblique imagery, were collected by responders for use in strategic planning on-site.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Multiple flights were conducted each day for the duration of the response, with the first flight occurring on June 25, 2021 at 16:30 local time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The images collected were processed using Structure from Motion soft- ware (specifically Agisoft Metashape Professional) to create dense 3D point clouds of the disaster site (see Fig 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Agisoft Metashape was also used to generate orthophoto mosaics, which are raster images made by merging orthophotos (aerial or satellite photographs which have been transformed to correct for perspective so that they appear to have been taken from vertically above at an infinite distance) of the collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work focuses on data collected on four separate days (June 27-30, 2021) at similar times of day.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This limited subset of the available data is used as a proof of concept of using the approach presented in this work to locate and characterize void spaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' In this work, the point clouds generated from data col- lected in four different UAS flights are used to identify and characterize voids in interior rubble, with the aim of better understanding the environments in which rescue UGVs must operate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The constructed point clouds are manually aligned, and the resulting point cloud stack is systematically investigated to identify and locate potential void spaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The potential voids and related oblique imagery are then evaluated to determine scale and cause of the voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='04743v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='RO] 11 Jan 2023 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Aerial imagery collected to support the Surfside disaster response [8] was processed to create an aggregated 3D model of the disaster site in the form of a point cloud stack (right), which was then analyzed to find and characterize voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' For contextualizing this 3D volumetric model, an aerial view of the Surfside collapse site [9] is shown (left).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' approach allows roboticists to see voids in rubble, and offers evidence of their location, frequency of occurrence, and size.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The paper is organized as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Prior work in disaster site analysis, the use of ground robots at other disaster responses, and void detection in 3D point clouds is discussed in Section II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Our approach to locating and characterizing interior rubble voids is presented in Section III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Section IV contains observations made, and analysis of identified voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Finally, conclusions and directions for future work are pre- sented in Section V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' BACKGROUND AND PRIOR WORK Prior work has attempted to characterize voids in rubble suitable for robotics, either through sensing with ground penetrating radar of simulated voids or from inference from cumulative studies;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' this work attempts to directly detect and characterize voids in an actual collapse using aerial imagery.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Prior work has also considered detecting voids in 3D point clouds, however those efforts have focused on more structured point clouds of synthetic data with smaller number of points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work addresses large, complex and noisy 3D point clouds generated from real-world imagery.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Analysis of Voids in Rubble Prior works have attempted to infer void characteristics from technical search tools, or to characterize in situ voids with ground penetrating radar [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' However, these efforts have been conducted in low fidelity physical simulations, and not with actual voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Further, ground penetrating radar suffers performance degradation in heterogeneous condi- tions, due to signal scattering, thereby leading to limited applicability in the mixed-material environment of a rubble pile at a disaster site.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' There is also analysis of building collapses from a civil engineering perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' For example, for the Surfside col- lapse, the nature and location of the rubble pile was analyzed in order to estimate the cause of the collapse, and some associated structural information [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' From this analysis, a simulation of the collapse can also be created.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' However, such analyses do not consider the interior properties of the rubble pile itself, and do not provide information on what voids may have existed in the rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The most comprehensive formal definition of the charac- teristics of rubble for robots appears in Disaster Robotics [5] and related articles [20], [21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The set of attributes presented in Disaster Robotics helps characterize the interior rubble as a robot work envelope, which is useful for selecting and deploying robots at a disaster site, as well as for developing new robotics rescue technologies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' There are eight attributes: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='scale, 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='verticality, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='tortuosity, 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='surface properties, 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='severity of obstacles, 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='accessibility elements, 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='number of regions, and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='non-navigational constraints, with 2-7 being collectively referred to as traversability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' These characteristics overlap with, and go beyond, two other sets of attributes developed for ground robot navigation, see [18], [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work focuses on determining scale and number of regions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Void Detection in 3D Point Clouds There exist several methods for detecting voids in point clouds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' One method has the user manually detect point cloud voids, and then is able to fill these point cloud voids in geometrically complex areas [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This approach however cannot process both smooth and sharp (or low and high frequency geometry) voids at the same time, as such properties are encoded into the parameters that are selected at run-time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Other methods create polygonal meshes from input point clouds using a Delaunay Triangulation algorithm, and then either perform a set of angle checks in order to determine which vertices are on the boundary of a void [13], or conduct a data search on the mesh to determine which vertices are present only in a single polygon, and thus lie on 6/27 6/28 6/29 6/30 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='2 38.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='4 (w) N 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='6 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='6 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='8 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='2 76.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='8 102.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='4 x (m) 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='6 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='2 y (m)the void boundary [14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Another work employs Ball Pivoting mesh generation [15] to improve void detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' While these methods have shown results in automatically detecting voids in 3D point clouds, they do not perform well on highly unstructured data, and point clouds that are missing sections, such as that of a rubble pile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The approach presented in this work uses systematic manual inspection, as the focus is on characterizing the voids found, and not yet on automating void characterization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' LOCATING AND CHARACTERIZING VOID SPACES In this work, a systematic manual approach is used to find, locate and characterize void spaces in the rubble pile of the Surfside building collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The approach can be broadly divided into two high-level steps: locating potential voids spaces, and characterizing the identified voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Potential Void Space Identification Finding potential void spaces involves processing the data gathered at Surfside, and then manually looking for voids in the following steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 1) Aerial images of the disaster site were collected from flying an uncrewed aerial system (UAS) to support response efforts [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 2) Agisoft Metashape was used to process the collected data into dense 3D point clouds of the disaster site.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 3) The 3D point clouds were manually aligned and “stacked“ to create an aggregated 3D volumetric rep- resentation of the collapse (see Fig 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 4) Cross sections were taken of the 3D volumetric model of the rubble pile to be able to look for potential voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The cross sections were taken on a 4m x 4m grid, as depicted in Fig 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The UAS flights over the Surfside site were conducted at several times each day during the response and recovery to map the disaster site [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This analysis opportunistically uses the imagery from one mapping flight at roughly the same time each day for four days: June 27, 2021 13:30, June 28, 2021 13:30, June 29, 2021 13:00, and June 30, 2021 11:00 local time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' These four days were chosen because it was a manageable set of data and it spanned days where the activity on the rubble pile was limited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Starting on July 1, 2021, operations increased and more people and equipment began to cover the surface of the rubble pile distorting the construction of point clouds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Void Space Characterization Once potential void spaces have been identified, each one was characterized, specifically in terms of size and scale Additionally, each potential void space was analyzed to distinguish between naturally occurring voids and artifacts of excavation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The steps for analyzing each potential void space are as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 1) The centroid location of the potential void was esti- mated, and then plotted on each of the four orthophoto mosaics of the Surfside site being considered in this work (June 27 - 30, 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The orthophoto mosaics Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' In order to systematically look for void spaces in the interior of the rubble pile, “slices“ or cross sectional views are taken into the aggregated 3D volumetric model, or point cloud stack, to examine cross sections of the rubble pile on a 4m x 4m grid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The slices are 1m x 64m in the XY plane, and have the depth of the entire rubble pile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Here the slices are depicted on the orthophoto mosaic constructed from images gathered during a single UAS flight on June 27, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' were generated using Agisoft Metashape, with one orthophoto mosaic for each UAS flight.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 2) Dimensions of the potential void were estimated by inspecting the XZ plane and YZ plane cross sections of the void space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Here, the Z axis points up.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 3) The estimated dimensions were used to approximate the size and volume of the potential void space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 4) The orthophoto mosaics, and oblique imagery of the potential void, were inspected for excavation activity, and compared to void dimensions and shape seen in the cross sectional views in order to determine if the potential void occurred naturally or was created by excavation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' RESULTS AND DISCUSSION To frame our observations, the effort described in this paper focuses on characterizing voids, which will help estab- lish the necessary form factor and perceptual constraints for search and rescue robots operating in the interior of rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work represents the first step: finding and confirming naturally occurring voids in the rubble pile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Consistent with hypotheses formed from Disaster Robotics, Very few potential voids are seen in the rubble pile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' “Slicing“ the rubble pile model (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' inspecting cross sections of the 3D volumetric model) on a 4m x 4m grid produced only six candidate voids (depicted in Fig 3) out of 225 possible locations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The candidate voids average a maximum height of 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='46m, a minimum height of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='26m, and a cross sectional width of 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='89m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A summary of their characteristics is presented in Table I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' While the voids themselves are not large enough to be survivable, they could offer breathing space, and show the types of void spaces that a robot would need to reach, inspect, and transit through.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The layers of the aggregated volumetric model constructed did appear to match the oblique imagery, implying that the layers are reasonably correct, and capture large objects and the “stair steps“ of the layers of a pancake collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' An example of this is seen in Fig 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 10 20 30 40 (w) ^ 09 60 70 80 06 100 20 40 60 80 100 120 140 160 180 x (m)Void Approximate Maximum Minimum XZ Plane Cross YZ Plane Cross Cause Volume (m3) Height (m) Height (m) Sectional Width (m) Sectional Width (m) Yellow 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='10 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='28 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='37 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='23 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='87 Excavation Cyan 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='53 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='88 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='33 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='25 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='25 Excavation Orange 41.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='71 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='45 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='15 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='03 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='91 Excavation Purple 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='71 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='19 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='20 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='37 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='06 Excavation Pink 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='94 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='14 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='26 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='41 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='73 Naturally Formed Green 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='33 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='82 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='23 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='69 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='91 Naturally Formed TABLE I SUMMARY OF CHARACTERISTICS OF IDENTIFIED VOID SPACES IN THE INTERIOR OF THE SURFSIDE RUBBLE PILE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The locations and estimated XY plane bounding boxes of the six identified voids are overlaid on the orthophoto mosaic from June 27, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Our observations also support the hypothesis that robots deployed to similar disaster sites would need to be able to burrow into the rubble in order to be effective (this is consistent with Disaster Robotics).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The voids seen in the Surfside collapse do not appear to form a connected set of tunnels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The voids also did not appear to have entrances from the surface of the rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A UGV would thus need to be able to burrow into the rubble to both access the voids, and also to traverse between them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Further, it is difficult to determine which voids are nat- urally occurring and which are artifacts of excavation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' By examining oblique imagery from throughout the four days, four out of six of the candidate voids appeared to be the result of excavation (see Fig 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Oblique imagery from June 27, 2021 and June 28, 2021 show high excavation activity in the regions of the yellow, cyan, orange, and purple potential voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This leads us to infer that the void space seen between the June 27 and June 28 layers is the result of excavation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Two of the six voids, specifically the pink and green voids in Fig 3, were in the pancake section of the collapse, and the associated oblique imagery suggests naturally forming voids were revealed by removing slabs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A layer of material is visibly removed between the June 27th and June 28th (see Fig 6), however, the void seen between the corresponding layers of the digital surface model is larger than would be expected from excavation alone, based on the estimated thickness of the slabs removed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This implied that a naturally formed void space was underneath the layer of slabs that were removed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Similarly, slabs were removed from the pink void region between June 27th and June 28th, with a void space that is larger than the estimated thickness of the slabs (see Fig 7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' In both the green and pink voids, slabs act as a “ceiling“ of sorts for the void space underneath.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Analysis of the voids shows that they may require robots with smaller form factors and more flexibility, which is consistent with predictions made in Disaster Robotics 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The six voids varied in maximum height from 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='82m to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='28m, minimum height from 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='15m to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='37m, and cross sectional width from 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='69m to 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='91m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The two voids in the pancake section (pink and green) were on the order of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='25m - 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='14m in height.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' While not survivable, these voids do indicate the size constraints for a robot to be able to penetrate 10 20 30 40 60 70 80 06 100 20 40 60 80 100 120 140 160 180 x (m)Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' By inspecting oblique imagery of the purple void (right), and relating this to a cross sectional view of the purple void (left), it is seen that large objects and features are reasonably recognizable and consistent between the imagery and the layers of the constructed 3D volumetric model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' High levels of activity are seen in the oblique imagery from June 27, 2021 (left), and June 28, 2021 (right) in the region of the yellow, cyan, orange, and purple voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' It can be inferred from this that these potential voids are really artifacts of excavation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Some material was removed in the green void region between June 27, 2021 (top left) and June 28, 2021 (top right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' There appears to be some slabs removed, which is consistent with the shape of the top of the void space seen in the XZ plane cross sectional view of the green void (bottom).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This implies that the green void was naturally formed during the collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Some material was removed in the pink void region between June 27, 2021 (top left) and June 28, 2021 (top right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' There appears to be some slabs removed, which is consistent with the shape of the top of the void space seen in the XZ plane cross sectional view of the pink void (bottom).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This implies that the pink void was naturally formed during the collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' and navigate into the void, and search for bigger voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' They also provide insight into the form factors best suited for exploring these voids - smaller, more flexible robots, like snake robots, are best suited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Candidate voids at Surfside did not show the verticality encountered in void spaces seen at the World Trade Center, parking garages, and mine disasters, however the depth of the rubble pile reinforces the notion that robots have to penetrate more than 6m into the rubble to reach areas where there might be survivable voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' CONCLUSIONS This preliminary work establishes a promising approach to characterizing and visualising the interior of rubble that not only allows for formally characterizing voids as robot work envelopes, which can aid in the design of effective robots, but 6/27 6/28 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='24 6/29 6/30 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='96, (w) z 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='68 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='4 40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='96 42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='24 43.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='52 44.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='8 46.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='08 47.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='36 48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='64 49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='92 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='2 y (m)10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='24 6/27 6/28 6/29 6/30 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='96 N 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='4 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='12 61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='44 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='72 64 65.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='28 66.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='56 67.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='84 69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='12 70.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='4 x (m)6/27 6/28 6/29 6/30 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='96 w) N 58.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='88 61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='44 64 66.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='56 69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='12 x (m)also offers data that could assist civil engineers in predicting where voids are likely to form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Further, this work could help rescue workers focus search efforts on areas where voids are likely.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The approach enables void spaces in rubble to be visualized for the first time by using a 3D volumetric aggregated model of the entire collapse that can be examined as cross-sections or “fly throughs.” However, further work is needed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The results showed an average translation error in point cloud alignment of about 80 cm, even though it is possible to detect interesting void spaces from point clouds, improvements would need to be made in order to more accurately characterize the dimensions of these voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Development of a tool to automatically register and align point clouds would help reduce the trans- lation error.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Automatic point cloud alignment on this data is challenging due to it’s complex and dynamic nature, and an analysis of existing methods for alignment supports the need for additional work in this space [16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This work also only considers four days of data (27 June, 2021 - 30 June, 2021), which correspond to the “top“ layers of the collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' More work is needed to cover all of the days of data, but activity significantly increased on the site after 30 June 2021, so while voids may appear deeper in the rubble pile, it may be harder to distinguish between naturally occurring voids and artifacts of excavation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Further, this is an opportunistic post hoc study;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' the aerial imagery was collected for strategic planning for the on-the- ground response and has notable gaps in frequency of data collection and viewpoints that would help ascertain voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The data was also restricted to RGB images.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' This approach can be improved both with dedicated data collection at higher frequency of UAS flights and from different angles, and also by using additional sensors, such as Lidar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' More frequent mapping imagery and obliques, or the use of Lidar, would likely support automated registration of point cloud layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' However, such data collection may interfere with the search and rescue priorities at a disaster site, and therefore, there is no guarantee that better data can be obtained, and most certainly, we do not want to interfere with the work flow of the rescue workers, as saving lives is the highest priority.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Despite these limitations, the preliminary results presented in this work confirm expectations of the small number and small scale of void spaces in rubble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' These results can be used for better selection and control of existing robotic platforms for a disaster response.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The purpose of this paper was to show that identifying the existence of voids is possible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The next step is then to provide in-depth analysis of the these voids, to both better characterize the rubble pile for possible locations of survivors, as well as understand how real rubble piles impact robot design.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Such design considerations include size, number of degrees of freedom, payload, sensory needs (especially for confined spaces) and compliance in the mechanism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Compliance is important because the rubble pile is highly unstructured, and thus will have an abundance of unmodeled features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' While this paper does not provide a detailed analysis of identified voids, further analysis will explore small size and actuation, structure or robot and payload, compliance due to lack of structure, and perception required to explore these voids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The complexity of the rubble with the different types of collapse modes may favor a modular robotics approach in order to adapt to the different types of confined space environments in different collapse areas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' ACKNOWLEDGMENT The authors thank Ananya Bal and Dr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Robert Ladig for their help with the data analysis and thoughtful discussions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' REFERENCES [1] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Barnett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Surfside condo collapse is third largest building failure in country’s history.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' CBS Miami, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' https://miami.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='cbslocal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='com/2021/06/29/surfside-condo-collapse- third-largest-building-failure-us-history/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [2] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Rhodes and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Ruiz-Goiriena.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' At least 9 victims of 98 dead may have initially survived surfside condo collapse, but were not found by rescue teams, investigation shows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' USA Today, Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 24, 2021;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Mar 14, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='usatoday.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='com/story/news/nation/2021/08/24/surfside- condo-collapse-there-may-have-been-more-survivors- florida/8152538002/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [3] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Thaler.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Nine of the 98 surfside victims may have survived the initial condo collapse: Fire rescue logs show woman was alive in rubble for ten hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' UK Daily Mail, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='dailymail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='co.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='uk/news/article-9924083/At-9-98- Surfside-victims-SURVIVED-initial-condo-collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='html.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [4] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Murphy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Trial by fire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' IEEE Robotics and Automation Magazine, 11(3):50–61, 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [5] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Murphy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Disaster robotics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' MIT Press, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [6] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Tadokoro, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Murphy, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Stover, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Brack, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Konyo, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Nishimura, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Tanimoto.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Application of active scope camera to forensic investigation of construction accident.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' In 2009 IEEE Workshop on Advanced Robotics and its Social Impacts, pages 47–50, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [7] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Whitman, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Zevallos, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Travers, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Choset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Snake robot urban search after the 2017 mexico city earthquake.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' In 2018 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR), pages 1–6, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [8] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Murphy, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Merrick, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Adams, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Broder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Bush, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Hart, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Hawkins.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' How robots helped out after the surfside condo collapse: Responders flew drones night and day to survey the collapse and search for survivors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' IEEE Spectrum, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [9] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' McLaughlin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' What we know about the building that partially collapsed in surfside, florida.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' CNN, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='cnn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='com/2021/06/24/us/what-to-know-about-the-surfside- building-collapse/index.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content='html.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [10] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Hu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Li, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Chen, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Kamat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Detecting, locating, and characterizing voids in disaster rubble for search and rescue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' In ” Advanced Engineering Informatics, volume 111, pages 109–124, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [11] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Lu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Guan, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Li, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Zheng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Fei, and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Zuo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A preliminary analysis and discussion of the condominium building collapse in surfside, florida, us, june 24, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Frontiers of Structural and Civil Engineering, pages 1097–1110, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [12] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Yang, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Yan, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Xiao.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Shape-controllable geometry comple- tion for point cloud models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' The Visual Computer, 33(3):385–398, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [13] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Dey and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Giesen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Detecting undersampling in surface reconstruction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Discrete and Computational Geometry, 15:329–345, 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [14] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A piecewise hole filling algorithm in reverse engineering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Computer-Aided Design, 37(2):263–270, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [15] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Bird, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Lennox, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Watson, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Wright.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Autonomous void detection and characterisation in point clouds and triangular meshes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' International Journal of Computational Vision and Robotics, 9(4):368 – 386, August 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' [16] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Bal, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Ladig, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Goyal, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Galeotti, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Choset, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Merrick, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' Murphy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' A comparison of point cloud registration techniques for on-site disaster data from the surfside structural collapse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/W9E3T4oBgHgl3EQf1AvA/content/2301.04743v1.pdf'} diff --git a/WNFLT4oBgHgl3EQfSy8N/content/tmp_files/2301.12042v1.pdf.txt b/WNFLT4oBgHgl3EQfSy8N/content/tmp_files/2301.12042v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..0498c4d2a1b0c90c63331b3e86047b891ef922dd --- /dev/null +++ b/WNFLT4oBgHgl3EQfSy8N/content/tmp_files/2301.12042v1.pdf.txt @@ -0,0 +1,682 @@ +arXiv:2301.12042v1 [gr-qc] 28 Jan 2023 +The gauge-invariant formulation of the local +expansion rate driven by the local average density +in an inhomogeneous universe +Masanori Tomonaga ∗1, Masumi Kasai †2, and Toshifumi Futamase +‡1 +1Faculty of Science, Kyoto Sangyo University, Motoyama, +Kamigamo, Kita-ku, Kyoto, 603-8555 Japan +2Graduate school of Science and Technology, Hirosaki +University,Hirosaki 036-8561, Japan +Abstract +The Hubble tension casts a blight on the standard cosmology. +As +a possible solution to the problem, the local variation of the expansion +rate has been proposed where the spatial averaging over a finite domain +was introduced in order to restore the local Friedmannian behavior in an +inhomogeneous cosmology. So far, however, the approaches are limited to +the particular choices of the gauges, and it has been unclear whether the +results are gauge-invariant. In this paper, we present the gauge-invariant +formulation of the local expansion rate which is driven by the spatial +average of the gauge-invariant inhomogeneous density. We show that the +local cosmological parameters in the finite domain may change from the +global parameters, and the relations between them are expressed by the +gauge-invariant averaged density. +1 +Introduction +The Hubble constant H0 is one of the most important cosmological parameter +since it characterizes the global properties of our universe. The standard cos- +mology is based on the assumption of the homogeneity and isotropy. Thus, the +Hubble parameter H0 is regarded as a constant over at least the horizon scale +which is also the prediction of the inflationary scenario. However recent obser- +vations suggest a non negligible difference between local and global (or recent +∗email:i2185088@cc.kyoto-su.ac.jp +†email:kasai@hirosaki-u.ac.jp +‡email:tof@cc.kyoto-su.ac.jp +1 + +and old) Hubble parameter [1, 2]. +There has been a large number of studies which try to resolve the discrepancy +[3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]. We regard that the dif- +ference of the local and global Hubble parameter is real and be explained by +the inhomogeneous distribution of the matter. In fact, the observation of the +K-band luminosity density seems to suggest that region with several hundred +Mpc around us has low density with density contrast δK ∼ −0.5 compared with +the globally averaged density [19]. Furthermore, there is some indication that +the voids are actually low density by weak lensing observation. Thus, it will be +meaningful to pursue the indication of the cosmological inhomogeneity. +The homogeneous and isotropic universe (here we call Friedmann Universe) +appears as the result of some kind of averaging procedure since the universe +is actually very inhomogeneous. There are various ways to averaging inhomo- +geneous universe (such as the light-corn averaging that is directly related with +observational quantities). In this paper, we only consider the scalar perturba- +tions in the linear order and the spatial averaging [20, 21, 22, 23, 24, 25, 26, +27, 28, 29, 30, 31, 32, 33, 34, 35]. Our purpose is not studying the averaging +itself in general inhomogeneous spacetimes, but rather the gauge dependence +of the relationship between locally averaged and globally averaged spacetime +in the linearly perturbed universe using the spatial averaging. +By adopting +the spatial averaging defined below, we were able to derive a locally averaged +Friedmann universe and have obtained the following relation between the locally +average Hubble parameter and the globally averaged Hubble parameter within +the framework of the general relativistic perturbation theory[36, 37] +HD0 = H0 +� +1 − 1 +3f(t0)⟨∆⟩Dt0 +� +, +(1) +where HD0 is the averaged Hubble parameter at the present time t0 over a +finite domain D, and H0 is the global, or the horizon scale Hubble parameter, +and ⟨∆⟩Dt0 is the present density contrast average over the domain D. f(t) = +d log ∆/d log a is the growth function of the density contrast. +However, the +treatment is carried out in the comoving synchronous and Newtonian gauge, +and there is some question if the averaging and the result are gauge invariant +or not. +In order to answer the question, we study the spatial averaging in the frame- +work of the gauge invariant cosmological perturbation theory, and find that local +Hubble parameter (and local cosmological parameter) can be describe used by +gauge-invariant physical quantities averaged in the local region D. +2 +Gauge-invariant linear perturbation theory +In this section, we briefly summarize the gauge-invariant perturbation theory +[38, 39]. We assume the flat background with dust fluid. Then the background +metric is +ds2 = −dt2 + a2(t)δijdxidxj +(2) +2 + +and the energy momentum tensor is +T µ = ρb(t) uµuν, +uµ = (1, 0, 0, 0) . +(3) +From the Einstein equations, we obtain the following Friedmann equation +� ˙a +a +�2 += 8πG +3 +ρb + Λ +3 +(4) +and from T µν +;ν = 0, we obtain the energy conservation equation +˙ρb + 3 ˙a +aρb = 0 . +(5) +Next, we write the metric and the energy-momentum tensor in the perturbed +universe as follows: +ds2 = gµνdxµdxν +(6) +g00 = −(1 + 2A) +(7) +g0i = −B,i +(8) +gij = a2(t) (δij + 2E,ij + 2Fδij) +(9) +T µν = ρ uµuν +(10) +ρ = ρb (1 + δ) +(11) +uµ = (u0, ui) = +� +1 − A, a−2δijv,j +� +, +(12) +where we consider only scalar perturbations and the scalar perturbation vari- +ables A, B, E, F, δ and v are arbitrary functions of t and xi, and assumed to be +small quantities. +Now consider the scalar type infinitesimal gauge transformation +¯t = t + α , +(13) +¯xi = xi + δijβ,j , +(14) +where α and β are arbitrary functions of t and xi, which are regarded as small as +the perturbation variables. The gauge dependence of the perturbed quantities +are +¯A = A − ˙α , +(15) +¯B = B − α + a2 ˙β , +(16) +¯E = E − β , +(17) +¯F = F − ˙a +aα , +(18) +¯δ = δ + 3 ˙a +aα , +(19) +¯v = v + a2 ˙β . +(20) +3 + +Then, the following gauge invariant quantities are defined in the usual manner. +Φ ≡ A − +� +B + a2 ˙E +�· +, +(21) +Ψ ≡ F − ˙a +a +� +B + a2 ˙E +� +, +(22) +∆ ≡ δ − 3 ˙a +a (v − B) , +(23) +V ≡ v + a2 ˙E . +(24) +Using these quantities, we can obtain the first-order equations in terms of +the gauge invariant quantities of linearized Einstein equation as follows: +− 1 +a2 ∇2Ψ = 4πGρb∆ +(25) +˙a +aΦ − ˙Ψ = −4πGρbV +(26) +Ψ + Φ = 0. +(27) +Using (27), the equations (25) and (26) are re-written as +1 +a2 ∇2Φ = 4πGρb∆, +(28) +˙Φ + ˙a +aΦ = −4πGρbV. +(29) +From T µν +;ν = 0, we obtain +˙∆ + 1 +a2 ∇2V = 0, +(30) +˙V + Φ = 0. +(31) +Differentiating (30) with respect to t and using (28) and (31), we obtain +¨ +∆ + 2 ˙a +a +˙∆ − 4πGρb∆ = 0. +(32) +The solution of the second-order differential equation (32) generally has two +independent modes as follows: +∆(t, xi) = D+(t)Q+(xi) + D−(t)Q−(xi) , +(33) +where +D+(t) = H +� t +dt′ +(aH)2 , +(34) +D−(t) = H = ˙a +a , +(35) +4 + +Q+(x) and Q−(x) represent the spatially dependent part of the growing and +decaying mode of the density contrast, respectively. +In summary, from (4) and (28) multiplied by 2/3, we obtain the following +equation +� ˙a +a +�2 ++ 2 +3 +1 +a2 ∇2Φ = 8πG +3 +(ρb + ρb∆) + Λ +3 +(36) +as the perturbed version of the Friedmann equation, and from (5) including +ρb∆, +∂ +∂t (ρb + ρb∆) + 3 ˙a +a (ρb + ρb∆) − ρb ˙∆ = 0 +(37) +as the perturbed version of the energy conservation equation. +3 +Spatial averaging over a local domain in the +perturbed universe +In the previous section, we have employed the standard assumption that the in- +homogeneous matter density ρ can be decomposed into the homogeneous back- +ground part ρb(t) and the small perturbed part δ. In the actual inhomogeneous +universe, however, we need to extract the homogeneous part through the aver- +aging procedure. +We define the spatial volume VD of a finite small domain D in the t = const. +hypersurface Σt as +VD ≡ +� +D +� +det(gij) d3x . +(38) +D is sufficiently smaller than the horizon scale but more than the scale at which +the picture of the Hubble expansion is valid, e.g. more than several 10 Mpc. +Using the metric described in the previous section, Σt is specified by the normal +vector +nµ = +� +1 − A, 1 +a2 δijB,j +� +. +(39) +Contrary to those in [36] and [37], no gauge-fixing is made in (38) in order to +specify the t = const. hypersurface Σt. +Fixing the gauge A = B = v = 0 +reproduces the results in [36], and another gauge B = E = 0 leads to those in +[37]. The spatial average of a scalar quantity Q(t, xi) over the domain D is in +general +⟨Q⟩ ≡ +1 +VD +� +D +Q +� +det(gij) d3x . +(40) +Therefore, the average density in this domain is +⟨ρ⟩ ≡ +1 +VD +� +D +ρ +� +det(gij) d3x . +(41) +Since we can observe only a finite nearby region of the entire space, it is likely +that the average density ⟨ρ⟩ in the nearby region does not always coincides with +the background density ρb. +5 + +Spatially averaging (37) over a local domain D, we obtain +�∂̺ +∂t +� ++ 3 ˙a +a⟨̺⟩ − ρb⟨ ˙∆⟩ = 0 , +(42) +where we have defined the gauge-invariant inhomogeneous density +̺ ≡ ρb + ρb∆ +(43) +in order to distinguish it from ρ = ρb + ρbδ. +Note that the time derivative does not commute with the spatial averaging +in general. In fact, for a physical quantity Q we have +�∂Q +∂t +� +− d +dt ⟨Q⟩ = +�1 +2gij ˙gij +� +⟨Q⟩ − +�1 +2gij ˙gijQ +� +. +(44) +However, if we consider the case Q → ρ up to the linear order, we obtain +�∂̺ +∂t +� +− d +dt ⟨̺⟩ = +�1 +2gij ˙gij +� +⟨ρb + ρb∆⟩ − +�1 +2gij ˙gij(ρb + ρb∆) +� += +� +3 ˙a +a +� +⟨ρb∆⟩ − +� +3 ˙a +aρb∆ +� += 0. +(45) +Therefore, using the relation (45), it is straightforward to show from (42) the +following equation holds up to the linear order of the perturbations: +d +dt ⟨̺⟩ + 3 ˙aD +aD +⟨̺⟩ = 0 , +(46) +where +˙aD +aD +≡ ˙a +a − 1 +3⟨ ˙∆⟩ +(47) +can be regarded as the local expansion rate driven by the local average density +⟨̺⟩. +In order to express (36) in terms of aD, we rewrite as +� ˙a +a +�2 +− 2 +3 +˙a +a +˙∆ + 2 +3 +1 +a2 +� +∇2Φ + a2 ˙a +a +˙∆ +� += 8πG +3 +(ρb + ρb∆) + Λ +3 . +(48) +Spatially averaging (48) and substituting (47), we obtain +� ˙aD +aD +�2 ++ Keff +a2 +D += 8πG +3 +⟨̺⟩ + Λ +3 , +(49) +6 + +where +Keff ≡ 2 +3 +� +∇2Φ + a2 ˙a +a +˙∆ +� +(50) += 2 +3 +� +−a2 ˙H∆ + a2H ˙∆ +� +(51) += 2 +3a2H2 +�D+ +H +�· +⟨Q+(xi)⟩ +(52) += 2 +3⟨Q+(xi)⟩ +(53) +is a constant which can be regarded as the effective curvature constant on the +local domain in the averaged sense. +Although (48) looks similar to that obtained in [36], we emphasize the fol- +lowing advantages of our analysis in this paper over that in [36]. +(1) [36]’s result is heavily dependent on the solution of δ ∝ a in the Einstein-de +Sitter background. In particular all of the averaged quantities are defined +and calculated directly using the growing mode solution in the Einstein-de +Sitter background described in eqs.(14-17) in [36]. So it is unclear whether +it holds in any other background. In this paper, we explicitly showed that +this averaged picture holds backgrounds other than the Einstein-de Sitter +background, especially even if Λ ̸= 0 background. +(2) If we don’t ignore the decaying mode of δ, [36] does not work. However, +our discussion has no problem even if we consider the decaying mode. +(3) It was unclear that [36]’s result is valid gauges other than comoving syn- +chronous gauge. Therefore, we explicitly showed that we can describe us- +ing the spatial average of gauge-invariant variables all the averaged density, +expansion rate, and (effective) curvature constant in an inhomogeneous +universe. +4 +The cosmological parameters in the nearby re- +gions expressed by the gauge-invariant vari- +ables +We define the global Hubble parameter as +H0 ≡ ˙a +a +��� +t0 +(54) +and the global density parameters as +Ωm ≡ 8πGρb(t0) +3H2 +0 +(55) +7 + +and +ΩΛ ≡ +Λ +3H2 +0 +, +(56) +where Ωm + ΩΛ = 1 since we have assumed the flat background. These global +parameters are supposed to be determined by the very large-scale and distant +observations such as the cosmic microwave background. +On the other hand, the cosmological parameters which are obtained from the +observations in the local nearby regions are certainly determined by the local +average density ⟨̺⟩, rather than by the background density ρb. We define the +local Hubble parameter as +˜H0 ≡ ˙aD +aD +��� +t0 = H0 +� +1 − 1 +3f(t0)⟨∆⟩t0 +� +, +(57) +where +f(t) ≡ d ln D+ +d ln a +(58) +is the growth function of the gauge-invariant density perturbation ∆, and the +local density parameters as +˜Ωm ≡ 8πG⟨̺⟩ +3 ˜H2 +0 += Ωm +� +1 + +� +1 + 2 +3f(t0) +� +⟨∆⟩t0 +� +(59) +and +˜ΩΛ ≡ +Λ +3 ˜H2 +0 += ΩΛ +� +1 + 2 +3f(t0)⟨∆⟩t0 +� +, +(60) +which are valid up to the linear order in the gauge-invariant variable ∆. +The local cosmological parameters coincide with the global ones if and only +if ⟨∆⟩ = 0. Otherwise, the local parameters may change. Let us show a simple +estimation in the case Λ = 0, where f(t) = 1. If the local nearby region is, say, +30% under dense, namely ⟨∆⟩t0 = −0.3, the local Hubble parameter ˜H0 can be +10% larger than the global H0. +5 +Conclusion and Discussion +Motivated by the Hubble tension, there have been many studies on the possible +resolutions. One of them is the local variation of the cosmological parameters +due to inhomogeneous matter distribution. We have also studied the inhomo- +geneous universe by spatial averaging and obtained an interesting result on the +relation between the local and global Hubble parameters which might explain +the Hubble tension. +However, the question of the gauge invariance of the result is not fully under- +stood. In this paper we address this question. We employ the gauge-invariant +linear cosmological perturbation theory to show that the relationship between +local and global cosmological parameter can be describe used by the gauge- +invariant physical quantities that averaged in the local region. +8 + +It is of some interest to develop this treatment to the second order since the +density contrast report by the observation of the K-band luminosity density is +of the order −0.5. Although we gave an argument based on the order of magni- +tude discussion of the cosmological Poisson equation, it is clearly not sufficient. +Another direction of this study is to consider the possible interpretation by the +inhomogeneity of the observation of m-z relation of Type Ia supernovae and +CMB Power spectrum. We hope to study this possibility in future. +Acknowledgment +This work is supported by a Grant-in-Aid for Scientific Research from JSPS(No. +20K03937) for TF. +References +[1] Nabila Aghanim, Yashar Akrami, Mark Ashdown, J Aumont, C Bacci- +galupi, M Ballardini, AJ Banday, RB Barreiro, N Bartolo, S Basak, et al. +Planck 2018 results-vi. cosmological parameters. +Astronomy & Astro- +physics, 641:A6, 2020. +[2] Adam G Riess, Lucas M Macri, Samantha L Hoffmann, Dan Scolnic, Ste- +fano Casertano, Alexei V Filippenko, Brad E Tucker, Mark J Reid, David O +Jones, Jeffrey M Silverman, et al. A 2.4% determination of the local value +of the hubble constant. The Astrophysical Journal, 826(1):56, 2016. +[3] Kiyotomo Ichiki, Chul-Moon Yoo, and Masamune Oguri. Relationship be- +tween the cmb, sunyaev-zel’dovich cluster counts, and local hubble parame- +ter measurements in a simple void model. Physical Review D, 93(2):023529, +2016. +[4] Krzysztof Bolejko. Emerging spatial curvature can resolve the tension be- +tween high-redshift cmb and low-redshift distance ladder measurements of +the hubble constant. Physical Review D, 97(10):103529, 2018. +[5] Kenji Tomita. +Cosmological models with the energy density of random +fluctuations and the hubble-constant problem. Progress of Theoretical and +Experimental Physics, 2017(8), 2017. +[6] Kenji Tomita. Super-horizon second-order perturbations for cosmological +random fluctuations and the hubble-constant problem. Progress of Theo- +retical and Experimental Physics, 2018(2):021E01, 2018. +[7] Micol Benetti, Welber Miranda, Humberto A Borges, Cassio Pigozzo, Saulo +Carneiro, and Jailson S Alcaniz. +Looking for interactions in the cos- +mological dark sector. Journal of Cosmology and Astroparticle Physics, +2019(12):023, 2019. +9 + +[8] ¨Ozg¨ur Akarsu, Suresh Kumar, Shivani Sharma, and Luigi Tedesco. Con- +straints on a bianchi type i spacetime extension of the standard λ cdm +model. Physical Review D, 100(2):023532, 2019. +[9] Paolo Cea. The ellipsoidal universe and the hubble tension. arXiv preprint +arXiv:2201.04548, 2022. +[10] Nikita Blinov, Celeste Keith, and Dan Hooper. Warm decaying dark matter +and the hubble tension. Journal of Cosmology and Astroparticle Physics, +2020(06):005, 2020. +[11] MC Bento, Orfeu Bertolami, and Anjan A Sen. +Generalized chaplygin +gas, accelerated expansion, and dark-energy-matter unification. Physical +Review D, 66(4):043507, 2002. +[12] Eleonora Di Valentino, Olga Mena, Supriya Pan, Luca Visinelli, Weiqiang +Yang, Alessandro Melchiorri, David F Mota, Adam G Riess, and Joseph +Silk. In the realm of the hubble tension—a review of solutions. Classical +and Quantum Gravity, 38(15):153001, 2021. +[13] Micol Benetti, HA Borges, Cassio Pigozzo, Saulo Carneiro, and JS Al- +caniz. Dark sector interactions and the curvature of the universe in light +of planck’s 2018 data. Journal of Cosmology and Astroparticle Physics, +2021(08):014, 2021. +[14] Kim V Berghaus and Tanvi Karwal. Thermal friction as a solution to the +hubble tension. Physical Review D, 101(8):083537, 2020. +[15] Karsten Jedamzik and Levon Pogosian. Relieving the hubble tension with +primordial magnetic fields. Physical Review Letters, 125(18):181302, 2020. +[16] Hayley J Macpherson, Paul D Lasky, and Daniel J Price. The trouble with +hubble: Local versus global expansion rates in inhomogeneous cosmological +simulations with numerical relativity. The Astrophysical Journal Letters, +865(1):L4, 2018. +[17] Rong-Gen Cai, Zong-Kuan Guo, Li Li, Shao-Jiang Wang, and Wang-Wei +Yu. Chameleon dark energy can resolve the hubble tension. Physical Review +D, 103(12):L121302, 2021. +[18] Marco San Mart´ın and Carlos Rubio. Hubble tension and matter inho- +mogeneities: a theoretical perspective. arXiv preprint arXiv:2107.14377, +2021. +[19] Ryan C Keenan, Amy J Barger, and Lennox L Cowie. Evidence for a 300 +megaparsec scale under-density in the local galaxy distribution. The As- +trophysical Journal, 775(1):62, 2013. +[20] Thomas Buchert and Mauro Carfora. Regional averaging and scaling in +relativistic cosmology. Classical and Quantum Gravity, 19(23):6109, 2002. +10 + +[21] David L Wiltshire. Cosmic clocks, cosmic variance and cosmic averages. +New Journal of Physics, 9(10):377, 2007. +[22] Thomas Buchert and J¨urgen Ehlers. Averaging inhomogeneous newtonian +cosmologies. arXiv preprint astro-ph/9510056, 1995. +[23] Ido Ben-Dayan, Maurizio Gasperini, Giovanni Marozzi, Fabien Nugier, and +Gabriele Veneziano. Backreaction on the luminosity-redshift relation from +gauge invariant light-cone averaging. Journal of Cosmology and Astropar- +ticle Physics, 2012(04):036, 2012. +[24] Maurizio Gasperini, G Marozzi, F Nugier, and Gabriele Veneziano. Light- +cone averaging in cosmology: Formalism and applications. Journal of Cos- +mology and Astroparticle Physics, 2011(07):008, 2011. +[25] Giuseppe Fanizza, Maurizio Gasperini, Giovanni Marozzi, and Gabriele +Veneziano. Generalized covariant prescriptions for averaging cosmological +observables. Journal of Cosmology and Astroparticle Physics, 2020(02):017, +2020. +[26] Jaiyul Yoo and Ruth Durrer. Gauge-transformation properties of cosmo- +logical observables and its application to the light-cone average. Journal of +Cosmology and Astroparticle Physics, 2017(09):016, 2017. +[27] Miko�laj Korzy´nski. Covariant coarse graining of inhomogeneous dust flow +in general relativity. Classical and quantum gravity, 27(10):105015, 2010. +[28] R Zalaletdinov. +Space-time averages of classical physical fields, annals. +eur. acad. sci.(2003) 344 [arxiv: gr-qc/0411004] aa coley, n. pelavas and +rm zalaletdinov, cosmological solutions in macroscopic gravity. Phys. Rev. +Lett, 95:151102, 2005. +[29] Chris Clarkson, Timothy Clifton, Alan Coley, and Rockhee Sung. +Ob- +servational constraints on the averaged universe. +Physical Review D, +85(4):043506, 2012. +[30] Alan A Coley and N Pelavas. Averaging spherically symmetric spacetimes +in general relativity. Physical Review D, 74(8):087301, 2006. +[31] Asta Heinesen, Pierre Mourier, and Thomas Buchert. On the covariance of +scalar averaging and backreaction in relativistic inhomogeneous cosmology. +Classical and Quantum Gravity, 36(7):075001, 2019. +[32] Thomas Buchert. On average properties of inhomogeneous fluids in general +relativity: perfect fluid cosmologies. General Relativity and Gravitation, +33(8):1381–1405, 2001. +[33] Thomas Buchert. On average properties of inhomogeneous fluids in general +relativity: dust cosmologies. General Relativity and Gravitation, 32(1):105– +125, 2000. +11 + +[34] Jelle P Boersma. Averaging in cosmology. Physical Review D, 57(2):798, +1998. +[35] Toyokazu Sekiguchi and Tomo Takahashi. Early recombination as a solution +to the H0 tension. Phys. Rev. D, 103:083507, Apr 2021. +[36] Masumi Kasai and Toshifumi Futamase. A possible solution to the hub- +ble constant discrepancy: Cosmology where the local volume expansion is +driven by the domain average density. Progress of Theoretical and Experi- +mental Physics, 2019(7):073E01, 2019. +[37] Masanori Tomonaga and Toshifumi Futamase. A comment on the averaging +in an inhomogeneous cosmology and the hubble constant problem. Progress +of Theoretical and Experimental Physics, 2021(10):103E02, 2021. +[38] James M Bardeen. Gauge-invariant cosmological perturbations. Physical +Review D, 22(8):1882, 1980. +[39] H Kodama and M Sasaki. +Multiple fluid density perturbations. +Prog. +Theor. Phys. Suppl, 78, 1984. +12 + diff --git a/WNFLT4oBgHgl3EQfSy8N/content/tmp_files/load_file.txt b/WNFLT4oBgHgl3EQfSy8N/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..ced749b2e23454b625a7b4dd26aa570e95e81eaf --- /dev/null +++ b/WNFLT4oBgHgl3EQfSy8N/content/tmp_files/load_file.txt @@ -0,0 +1,243 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf,len=242 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='12042v1 [gr-qc] 28 Jan 2023 The gauge-invariant formulation of the local expansion rate driven by the local average density in an inhomogeneous universe Masanori Tomonaga ∗1, Masumi Kasai †2, and Toshifumi Futamase ‡1 1Faculty of Science, Kyoto Sangyo University, Motoyama, Kamigamo, Kita-ku, Kyoto, 603-8555 Japan 2Graduate school of Science and Technology, Hirosaki University,Hirosaki 036-8561, Japan Abstract The Hubble tension casts a blight on the standard cosmology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' As a possible solution to the problem, the local variation of the expansion rate has been proposed where the spatial averaging over a finite domain was introduced in order to restore the local Friedmannian behavior in an inhomogeneous cosmology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' So far, however, the approaches are limited to the particular choices of the gauges, and it has been unclear whether the results are gauge-invariant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In this paper, we present the gauge-invariant formulation of the local expansion rate which is driven by the spatial average of the gauge-invariant inhomogeneous density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We show that the local cosmological parameters in the finite domain may change from the global parameters, and the relations between them are expressed by the gauge-invariant averaged density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 1 Introduction The Hubble constant H0 is one of the most important cosmological parameter since it characterizes the global properties of our universe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The standard cos- mology is based on the assumption of the homogeneity and isotropy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Thus, the Hubble parameter H0 is regarded as a constant over at least the horizon scale which is also the prediction of the inflationary scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' However recent obser- vations suggest a non negligible difference between local and global (or recent ∗email:i2185088@cc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='kyoto-su.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='jp †email:kasai@hirosaki-u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='jp ‡email:tof@cc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='kyoto-su.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='jp 1 and old) Hubble parameter [1, 2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' There has been a large number of studies which try to resolve the discrepancy [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We regard that the dif- ference of the local and global Hubble parameter is real and be explained by the inhomogeneous distribution of the matter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In fact, the observation of the K-band luminosity density seems to suggest that region with several hundred Mpc around us has low density with density contrast δK ∼ −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='5 compared with the globally averaged density [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Furthermore, there is some indication that the voids are actually low density by weak lensing observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Thus, it will be meaningful to pursue the indication of the cosmological inhomogeneity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The homogeneous and isotropic universe (here we call Friedmann Universe) appears as the result of some kind of averaging procedure since the universe is actually very inhomogeneous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' There are various ways to averaging inhomo- geneous universe (such as the light-corn averaging that is directly related with observational quantities).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In this paper, we only consider the scalar perturba- tions in the linear order and the spatial averaging [20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Our purpose is not studying the averaging itself in general inhomogeneous spacetimes, but rather the gauge dependence of the relationship between locally averaged and globally averaged spacetime in the linearly perturbed universe using the spatial averaging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' By adopting the spatial averaging defined below,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' we were able to derive a locally averaged Friedmann universe and have obtained the following relation between the locally average Hubble parameter and the globally averaged Hubble parameter within the framework of the general relativistic perturbation theory[36,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 37] HD0 = H0 � 1 − 1 3f(t0)⟨∆⟩Dt0 � ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (1) where HD0 is the averaged Hubble parameter at the present time t0 over a finite domain D,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' and H0 is the global,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' or the horizon scale Hubble parameter,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' and ⟨∆⟩Dt0 is the present density contrast average over the domain D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' f(t) = d log ∆/d log a is the growth function of the density contrast.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' However, the treatment is carried out in the comoving synchronous and Newtonian gauge, and there is some question if the averaging and the result are gauge invariant or not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In order to answer the question, we study the spatial averaging in the frame- work of the gauge invariant cosmological perturbation theory, and find that local Hubble parameter (and local cosmological parameter) can be describe used by gauge-invariant physical quantities averaged in the local region D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 2 Gauge-invariant linear perturbation theory In this section, we briefly summarize the gauge-invariant perturbation theory [38, 39].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We assume the flat background with dust fluid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Then the background metric is ds2 = −dt2 + a2(t)δijdxidxj (2) 2 and the energy momentum tensor is T µ = ρb(t) uµuν, uµ = (1, 0, 0, 0) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (3) From the Einstein equations, we obtain the following Friedmann equation � ˙a a �2 = 8πG 3 ρb + Λ 3 (4) and from T µν ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='ν = 0, we obtain the energy conservation equation ˙ρb + 3 ˙a aρb = 0 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (5) Next, we write the metric and the energy-momentum tensor in the perturbed universe as follows: ds2 = gµνdxµdxν (6) g00 = −(1 + 2A) (7) g0i = −B,i (8) gij = a2(t) (δij + 2E,ij + 2Fδij) (9) T µν = ρ uµuν (10) ρ = ρb (1 + δ) (11) uµ = (u0, ui) = � 1 − A, a−2δijv,j � , (12) where we consider only scalar perturbations and the scalar perturbation vari- ables A, B, E, F, δ and v are arbitrary functions of t and xi, and assumed to be small quantities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Now consider the scalar type infinitesimal gauge transformation ¯t = t + α , (13) ¯xi = xi + δijβ,j , (14) where α and β are arbitrary functions of t and xi, which are regarded as small as the perturbation variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The gauge dependence of the perturbed quantities are ¯A = A − ˙α , (15) ¯B = B − α + a2 ˙β , (16) ¯E = E − β , (17) ¯F = F − ˙a aα , (18) ¯δ = δ + 3 ˙a aα , (19) ¯v = v + a2 ˙β .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (20) 3 Then, the following gauge invariant quantities are defined in the usual manner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Φ ≡ A − � B + a2 ˙E �· , (21) Ψ ≡ F − ˙a a � B + a2 ˙E � , (22) ∆ ≡ δ − 3 ˙a a (v − B) , (23) V ≡ v + a2 ˙E .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (24) Using these quantities, we can obtain the first-order equations in terms of the gauge invariant quantities of linearized Einstein equation as follows: − 1 a2 ∇2Ψ = 4πGρb∆ (25) ˙a aΦ − ˙Ψ = −4πGρbV (26) Ψ + Φ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (27) Using (27), the equations (25) and (26) are re-written as 1 a2 ∇2Φ = 4πGρb∆, (28) ˙Φ + ˙a aΦ = −4πGρbV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (29) From T µν ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='ν = 0, we obtain ˙∆ + 1 a2 ∇2V = 0, (30) ˙V + Φ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (31) Differentiating (30) with respect to t and using (28) and (31), we obtain ¨ ∆ + 2 ˙a a ˙∆ − 4πGρb∆ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (32) The solution of the second-order differential equation (32) generally has two independent modes as follows: ∆(t, xi) = D+(t)Q+(xi) + D−(t)Q−(xi) , (33) where D+(t) = H � t dt′ (aH)2 , (34) D−(t) = H = ˙a a , (35) 4 Q+(x) and Q−(x) represent the spatially dependent part of the growing and decaying mode of the density contrast, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In summary, from (4) and (28) multiplied by 2/3, we obtain the following equation � ˙a a �2 + 2 3 1 a2 ∇2Φ = 8πG 3 (ρb + ρb∆) + Λ 3 (36) as the perturbed version of the Friedmann equation, and from (5) including ρb∆, ∂ ∂t (ρb + ρb∆) + 3 ˙a a (ρb + ρb∆) − ρb ˙∆ = 0 (37) as the perturbed version of the energy conservation equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 3 Spatial averaging over a local domain in the perturbed universe In the previous section, we have employed the standard assumption that the in- homogeneous matter density ρ can be decomposed into the homogeneous back- ground part ρb(t) and the small perturbed part δ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In the actual inhomogeneous universe, however, we need to extract the homogeneous part through the aver- aging procedure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We define the spatial volume VD of a finite small domain D in the t = const.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' hypersurface Σt as VD ≡ � D � det(gij) d3x .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (38) D is sufficiently smaller than the horizon scale but more than the scale at which the picture of the Hubble expansion is valid, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' more than several 10 Mpc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Using the metric described in the previous section, Σt is specified by the normal vector nµ = � 1 − A, 1 a2 δijB,j � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (39) Contrary to those in [36] and [37], no gauge-fixing is made in (38) in order to specify the t = const.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' hypersurface Σt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Fixing the gauge A = B = v = 0 reproduces the results in [36], and another gauge B = E = 0 leads to those in [37].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The spatial average of a scalar quantity Q(t, xi) over the domain D is in general ⟨Q⟩ ≡ 1 VD � D Q � det(gij) d3x .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (40) Therefore, the average density in this domain is ⟨ρ⟩ ≡ 1 VD � D ρ � det(gij) d3x .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (41) Since we can observe only a finite nearby region of the entire space, it is likely that the average density ⟨ρ⟩ in the nearby region does not always coincides with the background density ρb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 5 Spatially averaging (37) over a local domain D, we obtain �∂̺ ∂t � + 3 ˙a a⟨̺⟩ − ρb⟨ ˙∆⟩ = 0 , (42) where we have defined the gauge-invariant inhomogeneous density ̺ ≡ ρb + ρb∆ (43) in order to distinguish it from ρ = ρb + ρbδ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Note that the time derivative does not commute with the spatial averaging in general.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In fact, for a physical quantity Q we have �∂Q ∂t � − d dt ⟨Q⟩ = �1 2gij ˙gij � ⟨Q⟩ − �1 2gij ˙gijQ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (44) However, if we consider the case Q → ρ up to the linear order, we obtain �∂̺ ∂t � − d dt ⟨̺⟩ = �1 2gij ˙gij � ⟨ρb + ρb∆⟩ − �1 2gij ˙gij(ρb + ρb∆) � = � 3 ˙a a � ⟨ρb∆⟩ − � 3 ˙a aρb∆ � = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (45) Therefore, using the relation (45), it is straightforward to show from (42) the following equation holds up to the linear order of the perturbations: d dt ⟨̺⟩ + 3 ˙aD aD ⟨̺⟩ = 0 , (46) where ˙aD aD ≡ ˙a a − 1 3⟨ ˙∆⟩ (47) can be regarded as the local expansion rate driven by the local average density ⟨̺⟩.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In order to express (36) in terms of aD, we rewrite as � ˙a a �2 − 2 3 ˙a a ˙∆ + 2 3 1 a2 � ∇2Φ + a2 ˙a a ˙∆ � = 8πG 3 (ρb + ρb∆) + Λ 3 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (48) Spatially averaging (48) and substituting (47), we obtain � ˙aD aD �2 + Keff a2 D = 8πG 3 ⟨̺⟩ + Λ 3 , (49) 6 where Keff ≡ 2 3 � ∇2Φ + a2 ˙a a ˙∆ � (50) = 2 3 � −a2 ˙H∆ + a2H ˙∆ � (51) = 2 3a2H2 �D+ H �· ⟨Q+(xi)⟩ (52) = 2 3⟨Q+(xi)⟩ (53) is a constant which can be regarded as the effective curvature constant on the local domain in the averaged sense.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Although (48) looks similar to that obtained in [36], we emphasize the fol- lowing advantages of our analysis in this paper over that in [36].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (1) [36]’s result is heavily dependent on the solution of δ ∝ a in the Einstein-de Sitter background.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In particular all of the averaged quantities are defined and calculated directly using the growing mode solution in the Einstein-de Sitter background described in eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (14-17) in [36].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' So it is unclear whether it holds in any other background.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In this paper, we explicitly showed that this averaged picture holds backgrounds other than the Einstein-de Sitter background, especially even if Λ ̸= 0 background.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (2) If we don’t ignore the decaying mode of δ, [36] does not work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' However, our discussion has no problem even if we consider the decaying mode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (3) It was unclear that [36]’s result is valid gauges other than comoving syn- chronous gauge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Therefore, we explicitly showed that we can describe us- ing the spatial average of gauge-invariant variables all the averaged density, expansion rate, and (effective) curvature constant in an inhomogeneous universe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 4 The cosmological parameters in the nearby re- gions expressed by the gauge-invariant vari- ables We define the global Hubble parameter as H0 ≡ ˙a a ��� t0 (54) and the global density parameters as Ωm ≡ 8πGρb(t0) 3H2 0 (55) 7 and ΩΛ ≡ Λ 3H2 0 , (56) where Ωm + ΩΛ = 1 since we have assumed the flat background.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' These global parameters are supposed to be determined by the very large-scale and distant observations such as the cosmic microwave background.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' On the other hand, the cosmological parameters which are obtained from the observations in the local nearby regions are certainly determined by the local average density ⟨̺⟩, rather than by the background density ρb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We define the local Hubble parameter as ˜H0 ≡ ˙aD aD ��� t0 = H0 � 1 − 1 3f(t0)⟨∆⟩t0 � , (57) where f(t) ≡ d ln D+ d ln a (58) is the growth function of the gauge-invariant density perturbation ∆, and the local density parameters as ˜Ωm ≡ 8πG⟨̺⟩ 3 ˜H2 0 = Ωm � 1 + � 1 + 2 3f(t0) � ⟨∆⟩t0 � (59) and ˜ΩΛ ≡ Λ 3 ˜H2 0 = ΩΛ � 1 + 2 3f(t0)⟨∆⟩t0 � , (60) which are valid up to the linear order in the gauge-invariant variable ∆.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The local cosmological parameters coincide with the global ones if and only if ⟨∆⟩ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Otherwise, the local parameters may change.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Let us show a simple estimation in the case Λ = 0, where f(t) = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' If the local nearby region is, say, 30% under dense, namely ⟨∆⟩t0 = −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='3, the local Hubble parameter ˜H0 can be 10% larger than the global H0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 5 Conclusion and Discussion Motivated by the Hubble tension, there have been many studies on the possible resolutions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' One of them is the local variation of the cosmological parameters due to inhomogeneous matter distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We have also studied the inhomo- geneous universe by spatial averaging and obtained an interesting result on the relation between the local and global Hubble parameters which might explain the Hubble tension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' However, the question of the gauge invariance of the result is not fully under- stood.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In this paper we address this question.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We employ the gauge-invariant linear cosmological perturbation theory to show that the relationship between local and global cosmological parameter can be describe used by the gauge- invariant physical quantities that averaged in the local region.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 8 It is of some interest to develop this treatment to the second order since the density contrast report by the observation of the K-band luminosity density is of the order −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Although we gave an argument based on the order of magni- tude discussion of the cosmological Poisson equation, it is clearly not sufficient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Another direction of this study is to consider the possible interpretation by the inhomogeneity of the observation of m-z relation of Type Ia supernovae and CMB Power spectrum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' We hope to study this possibility in future.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Acknowledgment This work is supported by a Grant-in-Aid for Scientific Research from JSPS(No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 20K03937) for TF.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' References [1] Nabila Aghanim, Yashar Akrami, Mark Ashdown, J Aumont, C Bacci- galupi, M Ballardini, AJ Banday, RB Barreiro, N Bartolo, S Basak, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Planck 2018 results-vi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' cosmological parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Astronomy & Astro- physics, 641:A6, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [2] Adam G Riess, Lucas M Macri, Samantha L Hoffmann, Dan Scolnic, Ste- fano Casertano, Alexei V Filippenko, Brad E Tucker, Mark J Reid, David O Jones, Jeffrey M Silverman, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' A 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='4% determination of the local value of the hubble constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The Astrophysical Journal, 826(1):56, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [3] Kiyotomo Ichiki, Chul-Moon Yoo, and Masamune Oguri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Relationship be- tween the cmb, sunyaev-zel’dovich cluster counts, and local hubble parame- ter measurements in a simple void model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 93(2):023529, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [4] Krzysztof Bolejko.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Emerging spatial curvature can resolve the tension be- tween high-redshift cmb and low-redshift distance ladder measurements of the hubble constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 97(10):103529, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [5] Kenji Tomita.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Cosmological models with the energy density of random fluctuations and the hubble-constant problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Progress of Theoretical and Experimental Physics, 2017(8), 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [6] Kenji Tomita.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Super-horizon second-order perturbations for cosmological random fluctuations and the hubble-constant problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Progress of Theo- retical and Experimental Physics, 2018(2):021E01, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [7] Micol Benetti, Welber Miranda, Humberto A Borges, Cassio Pigozzo, Saulo Carneiro, and Jailson S Alcaniz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Looking for interactions in the cos- mological dark sector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cosmology and Astroparticle Physics, 2019(12):023, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 9 [8] ¨Ozg¨ur Akarsu, Suresh Kumar, Shivani Sharma, and Luigi Tedesco.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Con- straints on a bianchi type i spacetime extension of the standard λ cdm model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 100(2):023532, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [9] Paolo Cea.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The ellipsoidal universe and the hubble tension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' arXiv preprint arXiv:2201.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='04548, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [10] Nikita Blinov, Celeste Keith, and Dan Hooper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Warm decaying dark matter and the hubble tension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cosmology and Astroparticle Physics, 2020(06):005, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [11] MC Bento, Orfeu Bertolami, and Anjan A Sen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Generalized chaplygin gas, accelerated expansion, and dark-energy-matter unification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 66(4):043507, 2002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [12] Eleonora Di Valentino, Olga Mena, Supriya Pan, Luca Visinelli, Weiqiang Yang, Alessandro Melchiorri, David F Mota, Adam G Riess, and Joseph Silk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' In the realm of the hubble tension—a review of solutions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Classical and Quantum Gravity, 38(15):153001, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [13] Micol Benetti, HA Borges, Cassio Pigozzo, Saulo Carneiro, and JS Al- caniz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Dark sector interactions and the curvature of the universe in light of planck’s 2018 data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cosmology and Astroparticle Physics, 2021(08):014, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [14] Kim V Berghaus and Tanvi Karwal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Thermal friction as a solution to the hubble tension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 101(8):083537, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [15] Karsten Jedamzik and Levon Pogosian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Relieving the hubble tension with primordial magnetic fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review Letters, 125(18):181302, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [16] Hayley J Macpherson, Paul D Lasky, and Daniel J Price.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The trouble with hubble: Local versus global expansion rates in inhomogeneous cosmological simulations with numerical relativity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The Astrophysical Journal Letters, 865(1):L4, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [17] Rong-Gen Cai, Zong-Kuan Guo, Li Li, Shao-Jiang Wang, and Wang-Wei Yu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Chameleon dark energy can resolve the hubble tension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 103(12):L121302, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [18] Marco San Mart´ın and Carlos Rubio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Hubble tension and matter inho- mogeneities: a theoretical perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' arXiv preprint arXiv:2107.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content='14377, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [19] Ryan C Keenan, Amy J Barger, and Lennox L Cowie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Evidence for a 300 megaparsec scale under-density in the local galaxy distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' The As- trophysical Journal, 775(1):62, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [20] Thomas Buchert and Mauro Carfora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Regional averaging and scaling in relativistic cosmology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Classical and Quantum Gravity, 19(23):6109, 2002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 10 [21] David L Wiltshire.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Cosmic clocks, cosmic variance and cosmic averages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' New Journal of Physics, 9(10):377, 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [22] Thomas Buchert and J¨urgen Ehlers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Averaging inhomogeneous newtonian cosmologies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' arXiv preprint astro-ph/9510056, 1995.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [23] Ido Ben-Dayan, Maurizio Gasperini, Giovanni Marozzi, Fabien Nugier, and Gabriele Veneziano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Backreaction on the luminosity-redshift relation from gauge invariant light-cone averaging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cosmology and Astropar- ticle Physics, 2012(04):036, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [24] Maurizio Gasperini, G Marozzi, F Nugier, and Gabriele Veneziano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Light- cone averaging in cosmology: Formalism and applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cos- mology and Astroparticle Physics, 2011(07):008, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [25] Giuseppe Fanizza, Maurizio Gasperini, Giovanni Marozzi, and Gabriele Veneziano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Generalized covariant prescriptions for averaging cosmological observables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cosmology and Astroparticle Physics, 2020(02):017, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [26] Jaiyul Yoo and Ruth Durrer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Gauge-transformation properties of cosmo- logical observables and its application to the light-cone average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Journal of Cosmology and Astroparticle Physics, 2017(09):016, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [27] Miko�laj Korzy´nski.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Covariant coarse graining of inhomogeneous dust flow in general relativity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Classical and quantum gravity, 27(10):105015, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [28] R Zalaletdinov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Space-time averages of classical physical fields, annals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' (2003) 344 [arxiv: gr-qc/0411004] aa coley, n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' pelavas and rm zalaletdinov, cosmological solutions in macroscopic gravity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Lett, 95:151102, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [29] Chris Clarkson, Timothy Clifton, Alan Coley, and Rockhee Sung.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Ob- servational constraints on the averaged universe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 85(4):043506, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [30] Alan A Coley and N Pelavas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Averaging spherically symmetric spacetimes in general relativity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 74(8):087301, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [31] Asta Heinesen, Pierre Mourier, and Thomas Buchert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' On the covariance of scalar averaging and backreaction in relativistic inhomogeneous cosmology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Classical and Quantum Gravity, 36(7):075001, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [32] Thomas Buchert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' On average properties of inhomogeneous fluids in general relativity: perfect fluid cosmologies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' General Relativity and Gravitation, 33(8):1381–1405, 2001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [33] Thomas Buchert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' On average properties of inhomogeneous fluids in general relativity: dust cosmologies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' General Relativity and Gravitation, 32(1):105– 125, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 11 [34] Jelle P Boersma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Averaging in cosmology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 57(2):798, 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [35] Toyokazu Sekiguchi and Tomo Takahashi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Early recombination as a solution to the H0 tension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' D, 103:083507, Apr 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [36] Masumi Kasai and Toshifumi Futamase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' A possible solution to the hub- ble constant discrepancy: Cosmology where the local volume expansion is driven by the domain average density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Progress of Theoretical and Experi- mental Physics, 2019(7):073E01, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [37] Masanori Tomonaga and Toshifumi Futamase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' A comment on the averaging in an inhomogeneous cosmology and the hubble constant problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Progress of Theoretical and Experimental Physics, 2021(10):103E02, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [38] James M Bardeen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Gauge-invariant cosmological perturbations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Physical Review D, 22(8):1882, 1980.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' [39] H Kodama and M Sasaki.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Multiple fluid density perturbations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Prog.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Theor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' Suppl, 78, 1984.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} +page_content=' 12' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/WNFLT4oBgHgl3EQfSy8N/content/2301.12042v1.pdf'} diff --git a/WdFJT4oBgHgl3EQf4i1g/content/2301.11666v1.pdf b/WdFJT4oBgHgl3EQf4i1g/content/2301.11666v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c226934453a26831b7a714a79d822ed3f0a29315 --- /dev/null +++ b/WdFJT4oBgHgl3EQf4i1g/content/2301.11666v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60cf854c33bbef7e1572e24a8260163c65fa1d42ad00d0d6ad9dd03cdbc52227 +size 502245 diff --git a/WdFJT4oBgHgl3EQf4i1g/vector_store/index.pkl b/WdFJT4oBgHgl3EQf4i1g/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..f7e5d667104d7468e13d7c4a07f2760e2ff938f3 --- /dev/null +++ b/WdFJT4oBgHgl3EQf4i1g/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:456d178aed18b020582e0aaf81201a4ddd19ff1624e6ae6a5cd300989e83bf68 +size 134587 diff --git a/X9FRT4oBgHgl3EQf_DhA/content/2301.13693v1.pdf b/X9FRT4oBgHgl3EQf_DhA/content/2301.13693v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..da52d44c5ba7e8c519d7f97ede450bdc8052ea6c --- /dev/null +++ b/X9FRT4oBgHgl3EQf_DhA/content/2301.13693v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c1c8c3cce51e6bb5526a4f81a8459dec31a83fecc2a6e059cad8b320d36a284e +size 263244 diff --git a/X9FRT4oBgHgl3EQf_DhA/vector_store/index.pkl b/X9FRT4oBgHgl3EQf_DhA/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..b07cceb07c89e37828ca200311d9c3eccb7e860b --- /dev/null +++ b/X9FRT4oBgHgl3EQf_DhA/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d67f452324b1794afd3bdac9db77a31b5bad110920b32cc7f1d0b1f6186ef880 +size 73701 diff --git a/YNAyT4oBgHgl3EQf9fo9/vector_store/index.pkl b/YNAyT4oBgHgl3EQf9fo9/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..13677e162a8aca55c6ebf1bad87c7a00648af8b7 --- /dev/null +++ b/YNAyT4oBgHgl3EQf9fo9/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8a113e79082cc7efb2e97c8011215fc22b5def9789bf9f989af9c73fbb32041 +size 156978 diff --git a/YtE1T4oBgHgl3EQfwQVD/content/tmp_files/2301.03408v1.pdf.txt b/YtE1T4oBgHgl3EQfwQVD/content/tmp_files/2301.03408v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..54b0accfa030686bfbd731a0c67e5d7321d40c32 --- /dev/null +++ b/YtE1T4oBgHgl3EQfwQVD/content/tmp_files/2301.03408v1.pdf.txt @@ -0,0 +1,2304 @@ +1 +Dynamic Functional Connectivity +Christine Ahrends1 and Diego Vidaurre1,2 +1 Center of Functionally Integrative Neuroscience, Department of Clinical Medicine, Aarhus University, Denmark +2 Department of Psychiatry, University of Oxford, United Kingdom + +Abstract +Most generally, dynamic functional connectivity (FC) refers to the non-instantaneous +couplings across timeseries from a set of brain areas, here as measured by fMRI. This is in +contrast to static FC, which is defined as purely instantaneous relations. In this chapter, we +provide a hands-on description of a non-exhaustive selection of different methods used to +estimate dynamic FC (such as sliding windows, clustering approaches, Hidden Markov +Models, and multivariate autoregressive models), and we explain, using practical examples, +how data should be prepared for dynamic FC analyses and how models of dynamic FC can +be evaluated. We also discuss current developments in the dynamic FC research field, +including challenges of reliability and reproducibility, and perspectives of using dynamic FC +for prediction. +Keywords Functional connectivity, dynamics, sliding windows, clustering, Hidden Markov +Model +1 Background & motivation +The most traditional use of fMRI in research has been to detect changes in the signal +according to some task design, for example in response to a visual stimulus (Filippi, 2016; +Frahm et al., 1992; Friston et al., 2006; Kwong et al., 1992; Ogawa et al., 1992). This is a +first-order statistic of the signal —i.e. the average value of the signal for a given region. +Since the scale of the BOLD signal is not necessarily interpretable, it makes sense only + +2 +relative to a baseline. For example, is the signal in a given region higher on average after a +stimulus presentation than before? +In the last decade, however, there has been an explosion of studies aiming to characterise +individual differences in fMRI activity, not only in task but also in rest —that is, in the +absence of an explicit, constrained task (Lee et al., 2013; Lurie et al., 2019; Smith, Vidaurre, +et al., 2013). In rest, first order statistics are less useful because there is no baseline to +compare to, and there are no observable conditions to build a contrast. This is one of the +reasons why we look at second-order statistics, such as functional connectivity (FC), which +can be interpreted above and beyond the (arbitrary) scale of the signal (Power et al., 2011). +Another reason is that we expect FC to capture aspects of neural function and structure that +are not directly linked to, or visible in, first-order statistics. One example is neural +communication, which however might be reflected in FC in ways that are not necessarily +direct or linear. FC can also reveal aspects of topographical organisation at large brain +scales that cannot be retrieved from first-order statistics (Margulies et al., 2016). +FC is defined as a statistical relationship between two brain areas. The most common +measures of FC are linear, namely Pearson’s correlation or unnormalised covariance, or the +inverse of these, which serves to partialise the information; that is, in the inverse of a +covariance matrix (also called a precision matrix), the coefficient relating area i and area j +defines their (linear) relation after removing the influence of all the other areas (Joseph F. +Hair, 2009; Joseph F Hair, 2009). There are other measures of FC that aim at capturing +nonlinear information, most remarkably based in information-theory, such as mutual +information. Although these measures are more general and can potentially capture other +aspects of connectivity that escape a covariance or correlation matrix, they either require +more data to obtain a clean estimation or impose their own assumptions. In this chapter, we +will focus our discussion on linear FC, since it is by far the most common choice, not just in +general but particularly when the interest is in characterising the dynamic aspects of FC. + +3 +Another useful distinction is between FC and effective connectivity (EC) (Friston, 2011). As +opposed to FC, which is a mere statistical description of the data, EC is defined as the +direct, causal influence that an area has over another in the context of the network to which +they belong, and under specific models of biophysical constraints. Whereas EC can capture +aspects of the data that FC cannot, the estimation of EC is also more dependent on specific +assumptions and the inference of the parameters is more difficult. This chapter only focuses +on FC. +But what is dynamic FC precisely? Most broadly, we can define it as any second-order, +cross-region information that is not captured by the so-called static FC. Static FC is a usual +term to refer to the correlation between voxels or regions as computed within the entire +duration of the scanning session. Note that, since the standard practice is to standardise the +signal per scanning session to have a mean zero and standard deviation one, the +covariance and correlation matrices are mathematically equal when we speak about static +FC. Critically, static FC, as per this definition, only captures instantaneous relationships; that +is, within a scanning session, if we permute the time points within a session such that the +permutation is the same for all voxels or areas, the static FC estimate would not change. But +there is other information that would be lost by permuting; this information exists above and +beyond the static FC, and that is what we refer to as dynamic FC here. +One of the problems in the study of dynamic FC is that the literature uses the same term for +different types of measures, which has considerably muddled the discussion about the topic. +By basing our discussion on a broad definition of dynamic FC, we intend to clarify the +different aspects of dynamic FC that the literature has covered. Under the umbrella of this +definition, there are two different aspects of dynamic FC: time-varying instantaneous FC, +and FC in the context of a linear dynamical system that models non-instantaneous aspects +of FC. For now, we will discuss them conceptually. Later, when we cover the different +approaches to model dynamic FC, we will specify which type of information each approach +is aiming to model. + +4 +Time-varying instantaneous FC, commonly referred to as just time-varying FC, reflects +within-session modulations of the covariance matrix across regions, which captures +between-area linear couplings. We note that, after standardising the signal, the mean is zero +and the variance is one across the entire session, but this is not necessarily the case for +shorter periods of the signal, where the signal, for example, could transiently have a higher +variance. This means that covariance and correlation are not exactly equivalent anymore, +opening different possibilities in how we model and interpret the data. These practical +aspects and what they mean conceptually will also be discussed below. +Apart from instantaneous FC, the signal has other temporal information at different time +scales, including the effect imposed by the hemodynamic response function as well as other +factors of neural and non-neural origins. These non-instantaneous couplings between areas, +which are also dynamic FC, can be modelled as a linear dynamical system, like the +autoregressive model. In short, an autoregressive model models the multivariate signal at +time point ������������ (namely ������������������������) as a linear function of previous time points plus (Gaussian) noise. +This linear function is embodied by a set of autoregressive coefficients A, which, however +linear, contain rich information about the data. We will cover autoregressive models in some +detail in the Practical Approaches to dynamic FC section. For now, it suffices to say that A, +with a single set of parameters, can capture non-instantaneous phenomena such as +travelling waves, chaotic behaviour, and oscillations. +Different approaches of analysis can capture one or the other aspect, or both. But +importantly, they are not independent from each other. For example, if the signal contains +both types of information, but we use an analysis approach that is focussed on only one, +information of the other aspect will necessarily leak into the estimation. For this reason, they +are not trivial to separate, and arguments about which one is a better-grounded description +of the data are not easy to make. More practically, different approaches to dynamic FC allow +us to address different research questions in complementary ways; see Calhoun et al. + +5 +(2014); Hutchison et al. (2013); Lurie et al. (2019); Preti et al. (2017) for some examples of +applications. +This chapter is mainly devoted to practical aspects of dynamic FC analysis. This includes: +data preparation prior to the analysis, a non-exhaustive description of existing approaches to +characterise dynamic FC (with an emphasis on implementation), and some discussion on +how to validate these models after they have been estimated, followed by conclusions and +perspectives. Where applicable, we will demonstrate examples with Matlab code or refer to +existing software packages that can be used for the implementation. We chose Matlab +because it may still be the most common language in neuroimaging, but a translation of the +code examples to other languages like Python or Julia is straightforward. All code examples +can also be found at https://github.com/ahrends/DynamicFC_examples. +2 Preparing data for dynamic FC analysis +When preparing fMRI data for dynamic FC analyses, a good starting point is to follow +preprocessing guidelines for resting state fMRI, such as the Human Connectome Project’s +(HCP) resting state preprocessing pipeline (Glasser et al., 2013; Smith, Beckmann, et al., +2013). Resting state preprocessing guidelines may be more suitable than task-specific +preprocessing recommendations since traditional analyses of task data typically take +advantage of averaging over trials. There are a few considerations specific to dynamic FC +approaches, which we will outline here. For further reference, the issue of preprocessing for +dynamic FC analyses has also been tested and discussed in Lydon-Staley et al. (2019) and +Vergara et al. (2017). +2.1 Temporal preprocessing +Most importantly, dynamic approaches are more heavily affected by temporal noise than +time-averaged types of analyses. With temporal noise, we here refer to any type of artefact + +6 +that varies over time, e.g., head motion, other physiological artefacts such as cardiac or +respiratory artefacts, etc. While these artefacts may almost disappear when averaging over +timepoints, such as in time-averaged or trial-averaged approaches, they can drastically +influence the estimation of dynamic FC (Nalci et al., 2019). For instance, if the dominant +temporal fluctuations in an fMRI timeseries are due to head motion, a dynamic FC model +may use its explanatory power to describe these movement-related variations rather than +the more subtle signal fluctuations stemming from neural activity. In a state-based model, +this may result in one or more states being actually motion states and not “brain” states, +while in a continuously-varying FC estimation, each FC estimate may to some extent be +biased by motion. +On the other hand, dynamic FC analyses also suffer from too aggressive clean-up in the +time domain, as some of the meaningful temporal variability can be removed along with the +temporal artefacts. This may be the case, for instance, by applying preprocessing +approaches that average over or censor time points, such as motion scrubbing (Power et al., +2012), or when regressing out temporal noise components using a full variance clean-up +approach. Also global signal regression can affect temporal variability, both positively and +negatively, and so its use in preprocessing data for dynamic FC analysis is controversial +(Murphy & Fox, 2017). +The goal in terms of temporal variability when preprocessing fMRI data for dynamic FC +analyses should therefore be to remove temporal artefacts while retaining non-artefactual +temporal variability as much as possible. This may be achieved by non-aggressive temporal +preprocessing strategies such as independent component analysis (ICA) in combination with +unique variance clean-up of noise-related components (Griffanti et al., 2014). An additional +consideration as regards temporal variability is temporal filtering. While a relatively lenient +high-pass filter can be useful in removing ultra-slow fluctuations, such as scanner drifts, a +very narrow filter will restrict the timescale on which dynamic changes in FC can be detected +by a model. In general, since temporal noise and the meaningful aspect of the signal are not + +7 +perfectly separable, achieving the right balance can be a difficult task. A sensible approach +may be to start with a relatively lenient temporal clean-up and to test post-hoc whether the +dynamic FC model was affected by known temporal artefacts, such as head motion, and +decide whether more aggressive temporal clean-up is necessary (Ciric et al., 2017; Parkes +et al., 2018). +An additional factor when doing dynamic FC analyses, and in particular time-varying FC +analyses, is the variability of time-averaged FC between subjects; i.e. how different subjects +are in terms of their average between-region correlations. Whether this variability between +subjects is artefactual (e.g., due to poor registration) or non-artefactual (i.e. meaningful +individual differences), it may mask the more subtle temporal variations in FC, making +dynamic FC difficult to detect in a group-level timeseries (Lehmann et al., 2017). While non- +artefactual between-subject variability may be of interest in the analyses, preprocessing +should strive to minimise artefactual differences between subjects when planning a group- +level dynamic FC analysis. In this case, it is recommended to test how representative FC +patterns are of the group of subjects. A poor balance of low temporal variability and high +between-subject variability can lead a dynamic FC model to converge to a static solution, +which only describes differences between subjects (Ahrends et al., 2022). The question of +between-subject variability in FC is discussed in more detail in Bijsterbosch et al. (2017). +2.2 Parcellations and timecourse extraction +Another important consideration when preparing fMRI data for dynamic FC analyses is the +choice of parcellation and method for timecourse extraction. One may argue that, at least to +some extent, the choice of parcellation depends on basic beliefs about brain organisation +that have little to do with the analysis at hand. However, the way in which we divide the brain +into parcels and how we extract timecourses in practice determines important aspects of the +data, such as the amount of temporal and between-subject variability. Consequently, the +choice of parcellation and method of timecourse extraction also greatly affect the estimation + +8 +of dynamic FC (Ahrends et al., 2022; Iraji et al., 2020; Pervaiz et al., 2020). For simplicity, +we will here only demonstrate the effects of two types of parcellations on temporal variability +and dynamic FC: a priori (structural or functional) binary parcellations and data-driven +functional weighted parcellations. +In a binary parcellation, the region of interest is outlined with a discrete boundary and each +voxel can either belong (1) or not belong (0) to the parcel. In a data-driven functional +weighted parcellation, each voxel has a specific (continuous) weight associated with each +parcel. These weights are estimated from the data based on functional activity. It should also +be noted that the interpretation of FC itself changes depending on how parcels are defined. +For instance, if parcels are binary (meaning that each voxel can either belong to a given +parcel or not) and non-overlapping (meaning that each voxel can only belong to one parcel), +FC can be interpreted as connectivity between a pair of distinct regions. If, on the other +hand, parcels are weighted (meaning that each voxel has a certain value in each parcel) and +overlapping (meaning that each voxel belongs to some degree to several or even all +parcels), FC may be interpreted as connectivity between distributed pairs of networks. +There are also several methods for extracting timecourses from voxels in these +parcellations, some of which we will demonstrate below. In a binary parcellation, parcel +timecourses can be extracted as the mean over voxels or the first principal component (PC). +In a weighted parcellation, timecourses can be extracted using a dual regression approach +to obtain individual spatial maps and individual timecourses of parcels (Beckmann et al., +2009). To understand the effects of the parcellation and timecourse extraction method on +temporal variability and dynamic FC estimation, we will simulate a few different scenarios. +Let’s first consider the optimal case, in which, within each parcel, voxel activity is relatively +well homogenous. This is an implicit assumption of binary parcellations, but it is not usually +tested explicitly. This case is illustrated in Figure 1. We simulate two functional clusters, A +and B, each with a consistent timecourse whose peak is located at the centre of the cluster. + +9 +In this scenario, both clusters’ timecourses are simply sine-waves. The only difference +between the clusters is the overall signal amplitude. +A = zeros(10,10); +for i = 1:5 + A(i, i:(end-(i-1))) = 0.2*i; + A(i:(end-(i-1)), i) = 0.2*i; + A(end-(i-1), i+1:end-(i-1)) = 0.2*i; + A(i+1:end-(i-1), end-(i-1)) = 0.2*i; +end +B = A; +T = 1:100; +yA = A(:)*sin(T); +yB = B(:)*sin(T)*0.7; +Here, yA are the timecourses for the voxels in cluster A and yB are the timecourses in +cluster B. The voxel weights for cluster A and cluster B as well as the empirical timecourses +yA and yB are shown in Figure 1, top row. + + +10 + +Figure 1 Parcellation and timecourse extraction for scenario 1, where voxel clusters in parcel have +homogeneous timecourses. +We will consider the case where both clusters are contained within the same parcel. In +binary, non-overlapping parcellations, timecourses are often extracted by simply computing +the mean over all voxels belonging to each parcel at each timepoint: +for t = T + parcel_mean(t) = mean([yA(:,t); yB(:,t)]); +end +The resulting timecourses are shown in Figure 1, second row. In this (optimal) case, the +average activity across voxels captures the overall pattern in the region adequately. +However, while this method is fast and easy to compute, it can have substantial +shortcomings depending on the actual temporal variance of voxel clusters in the parcel. + +Empirical voxel timecourses +Empirical voxel weights +Cluster A +5 +Cluster A +0.8 +10 +0.6 +Cluster B +Cluster B +15 +0.4 +20 +0.2 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +6 +10 +Binary parcellation, mean extracted +Voxel weights binary parcel +5 +Parcel 1 +10 +(binary) +0 +15 +20 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +6 +10 +Binary parcellation, 1st PC extracted +Voxel weights binary parcel +5 +Parcel 1 +10 +(binary) +0 +15 +20 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +6 +10 +Weighted parcellation, regression coefficients extracted +Voxel weights weighted parcel +5 +Parcel 1 +10 +(weighted) +15 +20 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +610 +Time11 +Consider the case where the two functional clusters’ activity contained within the parcel is +heterogeneous, e.g., they are negatively correlated (see Figure 2, top row): +yA = A(:)*sin(T); +yB = B(:)*-sin(T); +If we here extract the parcel timecourse as mean over all voxels, we will completely flatten +the timecourse so that the original temporal variance will be lost (see Figure 2, second row). +An alternative to extracting the parcel timecourse as the mean over all voxels is to use the +first principal component (PC) of the voxel timecourses within the given parcel: +parcel_pc_tmp = pca([yA; yB]); +parcel_pc = parcel_pc_tmp(:,1); +The resulting timecourses are shown in Figure 1 and Figure 2, third rows. Both in the case +of homogeneous activity within the parcel (Figure 1) and in the case of heterogeneous +activity within the parcel (Figure 2), the first PC captures the dominant pattern of voxel +activity over time. In this way, we can avoid the issue of the parcel timecourse flattening out +and instead preserve the most dominant temporal variance. However, using this approach, +some information about the temporal variance within the region of interest would still be lost, +namely that of cluster B in the example. + +12 + +Figure 2 Parcellations and timecourse extraction, scenario 2, where voxel clusters in parcel have +heterogeneous timecourses. +Using a weighted parcellation gives a more nuanced view of the activity within a parcel. We +can create a weighted parcel from the cluster timecourses by decomposing the data, for +instance, using independent component analysis (ICA): +ica_mdl = rica([yA;yB],1); +parcel_ic = ica_mdl.TransformWeights; +Here, parcel_ic is the timecourse of a single IC extracted from the voxel-level timecourses of +clusters A and B. We can access each voxel’s individual weight within the parcel: + +Empirical voxel timecourses +Empirical voxel weights +Cluster A +5 +Cluster A +0.8 +10 +0.6 +Cluster B +Cluster B +15 +0.4 +20 +0.2 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +6 +10 +Binary parcellation, mean extracted +Voxel weights binary parcel +5 +Parcel 1 +10 +(binary) +0 +15 +20 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +Binary parcellation, 1st PC extracted +Voxel weights binary parcel +5 +Parcel 1 +10 +(binary) +0 +15 +20 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +6 +10 +Weighted parcellation, regression coefficients extracted +Voxel weights weighted parce +5 +Parcel 1 +10 +(weighted) +15 +20 +0 +10 +20 +30 +40 +50 +60 +70 +80 +90 +100 +2 +610 +Time13 +z = transform(ica_mdl,[yA;yB]); +weights_ICA_A = reshape(z(1:100),[10,10]); +weights_ICA_B = reshape(z(101:end),[10,10]); +In scenario 1 (Figure 1, fourth row), the IC parcel timecourse corresponds closely to the +timecourse of cluster A. The voxel weights capture the original spatial distribution of the +clusters and are higher for the voxels belonging to cluster A than the voxels belonging to +cluster B. In this way, the parcel timecourse in combination with the voxel weights capture +not only the most dominant temporal pattern, but also the small amplitude difference +between clusters A and B. In scenario 2 (Figure 2, fourth row), the IC parcel timecourse +again corresponds closely to the timecourse of cluster A. However, while the estimated voxel +weights for the voxels from cluster A are positive, the voxels from cluster B have negative +weights assigned, which captures how cluster B’s voxels’ activity is negatively correlated +with the activity of cluster A voxels. This method thus retains almost all spatiotemporal +details of the original data, so that by multiplying the voxel weights with the parcel +timecourses, we could almost perfectly recreate the empirical voxel timecourses. +An additional advantage of estimating the parcellation based on the functional activity from +the dataset at hand is that we can avoid the issue of suboptimal parcel boundaries. This is +especially important in the context of dynamic FC. Consider the case of two functional +clusters, which are positively correlated at the beginning of a timeseries, then fall out of +synchrony in the middle of the timeseries, and are correlated again at the end of the +timeseries: +yA = A(:)*sin(T); +yB(:,1:30) = B(:)*sin(T(1:30)); +yB(:,31:70) = B(:)*sin(31:0.5:50.5); +yB(:,71:100) = B(:)*sin(T(71:100)); +We can calculate the average correlation between the two clusters, using e.g., a sliding +window approach (see section 3.1.1 for details). This is shown in Figure 3, top row. Binary +parcellations that are defined a priori on a different dataset can have the problem of +suboptimal parcel boundaries. In this case, rather than dividing the data into the “true” + +14 +clusters A and B, an a priori defined parcellation may split both clusters in the middle so that +half of cluster A and half of cluster B belong to parcel 1 and the other half of each cluster +belongs to parcel 2. In this case, whether timecourses are extracted using the mean or the +first PC, the two clusters’ activity will be contained equally in both parcels, making it +impossible to disambiguate the original functional clusters. +for t = T + parcel_mean(1,t) = mean([yA(1:50,t); yB(1:50,t)]); + parcel_mean(2,t) = mean([yA(51:100,t); yB(51:100,t)]); +end +parcel_pc_tmp_1 = pca([yA(1:50,:); yB(1:50,:)]); +parcel_pc(:,1) = parcel_pc_tmp_1(:,1); +parcel_pc_tmp_2 = pca([yA(51:100,:); yB(51:100,:)]); +parcel_pc(:,2) = parcel_pc_tmp_2(:,1); +This is shown in Figure 3, second and third row for the mean timecourse extraction and first +PC timecourse extraction methods, respectively. If we estimate dynamic FC between the two +parcels using these timecourses, FC would be high throughout the timecourse and not +capture the change in FC in the middle of the timeseries, since we are essentially computing +dynamic FC of both functional clusters with themselves. +The data-driven functional parcellation (ICA) is able to capture the true parcel boundaries +more accurately. We estimate ICA with two components: +ica_mdl = rica([yA;yB],2); +parcel_ic = ica_mdl.TransformWeights; +We then have a separate set of voxel weights for each component (parcel): +z = transform(ica_mdl,[yA;yB]); +weights_ICA_A_1 = reshape(z(1:100,1),[10,10]); +weights_ICA_B_1 = reshape(z(101:end,1),[10,10]); +weights_ICA_A_2 = reshape(z(1:100,2), [10,10]); +weights_ICA_B_2 = reshape(z(101:end,2),[10,10]); +These components capture the separate timecourses for clusters A and B. Estimating +dynamic FC between these two components more closely resembles the empirical dynamic +FC between the functional clusters (see Figure 3, fourth row). + +15 + + + +Figure 3 Dynamic FC between two cluster timecourses (empirical: row 1) extracted from binary +parcels with suboptimal parcel boundaries (row 2 and 3) vs. extracted from weighted, data-driven +parcels (row 4) +The examples showed how a suboptimal parcellation or timecourse extraction method can +lead to loss of temporal information and wrong estimation of the temporal relationships +between parcels. When preparing data for dynamic FC analysis, it is therefore crucial to +choose a suitable parcellation and timecourse extraction method. Extracting timecourses as +the first PC rather than the mean can preserve temporal variability in a binary parcellation. +For dynamic FC analyses, several studies found that data-driven functional parcellations, +such as group ICA approaches, are preferred over binary parcellations (Ahrends et al., 2022; +Iraji et al., 2020; Pervaiz et al., 2020), as data-driven parcellations not only preserve +temporal variability, but also avoid the issue of suboptimal parcel boundaries that can occur + +Empiricalvoxel timecourses +Empirical voxel weights +Empirical dFC +A +Cluster +5 +0.8 +Cluster A +10 +0.6 +0 +B +Cluster B +luster +15 +0.4 +0.2 -1 +20 +0 +20 +40 +60 +80 +100 +0 +2 +20 +40 +60 +80 +100 +Binary parcellation, mean extracted +Voxel weights binary parcels +dFc binary parcels, mean +5 +Parcel 1 +1 +10 +0 +■0 +Parcel 2 +15 +20 +-1F +0 +20 +40 +60 +80 +100 +2610 2610 +0 +20 +40 +60 +80 +100 +Parcel 1 +Parcel 2 +Binary parcellation, 1st Pc extracted +Voxel weights binary parcels +dFC binary parcels, 1st PC +5 +Parcel 1 +1 +10 +0 +0 +Parcel 2 +15 +20 +-1F +0 +20 +40 +60 +80 +100 +2610 2610 +0 +20 +40 +60 +80 +100 +Parcel 1 +Parcel 2 +Weighted parcellation, regression coefficient extracted +Voxel weights weighted parcels +dFC weighted parcels, regression coefficient +5 +Parcel 1 +10 +0 +15 +Parcel 2 +6 +20 +0 +20 +40 +60 +80 +100 +2610 2 +610 +0 +20 +40 +60 +80 +100 +Time +Parcel 1 +Parcel 2 +Time16 +with a priori binary parcellations. It should be mentioned that alternative approaches exist +that compute time-varying FC on the voxel-level or define a parcellation based on time- +varying FC (Preti & Van De Ville, 2017), but they are less common. +Beyond the type of parcellation, its coarseness needs to be taken into account. While it may +be tempting to choose a parcellation to be as fine-grained as possible to maximise spatial +resolution, a too fine-grained parcellation increases the number of parameters in a dynamic +FC model, which can lead to noisy or even ill-posed estimations. Broadly speaking, the +number of parameters to estimate in the model should be in balance with the number of +observations, which are often limited in common fMRI datasets. An alternative to using a +coarser parcellation is to use a dimensionality reduction technique, such as Principal +Component Analysis (PCA), on the extracted timecourses of a fine-grained parcellation. +However, working with transformed data in PCA-space may also introduce a bias for certain +dynamic FC models (Vidaurre, 2021). +3 Practical Approaches to dynamic FC +There is a plethora of approaches to evaluate dynamic FC. Here, we focus on methods that +fit our definition of dynamic FC, i.e., that are based on second-order statistics. This excludes, +for instance, popular methods based on co-activation maps (CAPs) (X. Liu et al., 2018), +which are based on first-order information. Also, we prioritise a practical understanding of +the methods at the expense of breadth, so only a subset of representative methods will be +discussed and analysed. For each approach there often are several software packages +available. +We present methods and software to estimate single-subject, single-session dynamic FC. +However, all the principles are applicable to compute the estimates over a range of sessions +and subjects. This would be done by concatenating all the time series and taking into +account when a session finishes and the next starts. We will assume that the data for a + +17 +given scanning session have been formatted into a matrix X, with as many rows as fMRI +volumes (i.e., time points) and as many columns as voxels, regions, or components. +3.1 Continuously-varying estimators +We refer as continuously-varying estimators to methods that produce an estimate of FC per +time point or window, such that the parameters that define such estimations vary smoothly in +the time axis. Because the dynamics are encoded by changes in these parameters, these +approaches are time-varying estimates according to the above classification. We can +separate two kinds of continuously-varying estimators: those based on sliding windows, +which divide the data set in contiguous pieces and perform the estimation separately per +piece; and all-data estimators, which use the data set all at once. Sliding windows are +simpler and much more common. +3.1.1 Sliding windows +The most basic approach to evaluate dynamic FC is sliding windows. Here, given a certain +(typically predefined) choice of window length in terms of number of frames, we perform the +estimation within this window and then slide the window by one or more time points and +repeat the estimation. This is done across the entire time series. For example, given p +number of brain regions, a basic estimation would be +window_length = 100; +C = zeros(p,p,total_session_duration - window_length + 1); +for t = 1:total_session_duration - window_length + 1 +C(:,:,t) = corr(X(t:t+window_length-1,:)); +end +In this code, we slide the window by one time point; for computational efficiency, the window +can be slid by more than one time point. A family of variations of this scheme is related to +the shape of the window: here a squared window was used, but there are alternatives that +can improve the estimation in practice such as a tapered or an exponentially decaying +window. In contrast to a static FC estimation that uses the entire timeseries (Figure 4A), + +18 +Figure 4B illustrates how FC is estimated by sliding windows (Figure 4C refers to a state- +based estimation, which will be described later). + +Figure 4 Static and time-varying FC. We refer to second-order statistics of fMRI timeseries, such as correlations +between regions of interest (ROIs), as functional connectivity (FC). A) Time-averaged (static) FC can be +computed, for instance, by correlating all pairs of ROIs over the entire timeseries. The resulting FC can be +illustrated as (ROI x ROI) FC matrix or as FC map in the brain. B) Continuously varying dynamic FC, such as +sliding window approaches, estimate FC separately on portions (windows) of the timeseries. Each window has an +associated FC matrix that can also be projected into the brain. C) State-based FC models estimate recurring +patterns of FC over the timeseries. Each state has an associated FC matrix that can also be projected into the +brain. (Abbreviations: ROI - region of interest; FC - functional connectivity; TR - repetition time). + +A +Time-averaged +(static) +fMRI timeseries +FC matrix +FC brain map +ROI 1 +ROI 1 +w +ROI2 +RO12 +ww +wiy +Mmw +Mr +ROI n +ROI1 +ROIn +ROIn +ROI2 +Time (TR) +ROI 1 +ROI n +ROI x ROI correlation +B +Time-varying FC: continuously-varying estimation +FC matrix +ROI 1 +ROI2 +ROI n +ROlxROI correlation +ROI 1 +ROI2 +Myny +W +ROIn +W +Time (TR) +C +Time-varying FC: state-based estimation +State 1 +State 2 +State 3 +State 4 +State k +ROI 1 +ROI2 +ROIn +FC matrix19 + +Here, we used correlations, but unnormalised covariance is also possible by using cov() +instead of corr(); in this case, we will also be including information about the variance of +the signal. Now, if we wish to compute partial correlation (or covariance), where the linear +influence of all the other regions is removed from the coefficient between region i and j, we +can consider the following code: +for t = 1:total_session_duration - window_length + 1 + +inv_mat = inv(C(:,:,t) + lambda * eye(p)); + +inv_mat = - (inv_mat ./ ... + +repmat(sqrt(abs(diag(inv_mat))),1,p)) ./ ... + + +repmat(sqrt(abs(diag(inv_mat)))',p,1); +inv_mat(inv_mat(p)>0)=0; +iC(:,:,t) = inv_mat; +end +Inverse covariance matrices (here, inv_mat) are also referred to as precision matrices. If the +number of regions is comparatively large, we need to regularise the matrix inversion to avoid +badly scaled results. This is what we did here by adding the identity matrix multiplied by the +regularisation constant lambda. Other types of regularisation that impose sparsity, by +driving some partial correlations to exactly zero, are also possible. In this case, the resulting +precision matrices can be mapped to a graph where non-zero coefficients relate to edges +between two nodes (voxels, areas, or components), and zero coefficients relate to +conditional independence between node i and j, given all the rest of the nodes (Cai et al., +2018; Friedman et al., 2008). +Sliding-window analyses can be very dependent on the choice of the window length, in the +sense that too short windows will render very unstable estimates and too long windows will +over-smooth the estimation and miss relatively fast changes. Although there are data-driven +approaches to adaptively optimise the window length, the resulting window length is itself +based on assumptions and subject to estimation noise. +A critical weakness of the sliding window approach is that, except perhaps for very long +windows, a large part of the variability observed across windows will inevitably be due to + +20 +estimation noise. This is because the number of time points within a window is not large +enough to yield a stable measure; this gets exacerbated by the autocorrelations in the +signal, which further reduce the effective degrees of freedom within the window (Afyouni et +al., 2019). +This can be easily verified empirically with the following code +total_session_duration = 2000; p = 10; +X = mvnrnd(zeros(total_session_duration,p),true_C); +for j = 1:p + X(:,j) = smooth(X(:,j),10); +end +X = X(101:end-100,:); +Here, true_C was a static covariance matrix obtained from real fMRI data, which we used to +sample Gaussian noise. We then applied some smoothing to generate some autocorrelation +in the data (imitating what we would observe in real fMRI data at a very basic level). If we +then apply a sliding window analysis, we will observe broadly varying estimations of time- +varying FC with either correlation or partial correlation, even though the underlying +covariance matrix is not time-varying. This is illustrated in the left panel of Figure 5A, where +we show, for a pair of signals, the results of performing a sliding window analysis on data +generated as per the above code; the dotted and solid grey lines correspond, respectively, to +the ground-truth correlation (i.e. true_C(1,2)) and empirical static correlation (i.e. +corr(X(:,1),X(:,2))) —note the large difference between the two due to smoothing, +even when using the entire length of the signals. For further discussion and examples on +model validation, see Section 4 below. +For comparison, we generated a second data set where there are two different covariance +matrices underlying the generation of the data; the middle section of the time series was +generated using one, and the beginning and end of the time series was generated using the +other: + +21 +total_session_duration = 2000; p = 10; +X1a = mvnrnd(zeros(total_session_duration/4,p),true_C1); +X1b = mvnrnd(zeros(total_session_duration/4,p),true_C1); +X2 = mvnrnd(zeros(total_session_duration/2,p),true_C2); +X = [X1a; X2; X1b]; +for j = 1:p + X(:,j) = smooth(X(:,j),10); +end +X = X(101:end-100,:); +Figure 5B shows that, while the real modulation in FC is apparent, the fluctuations around it +are also quite large. + +Figure 5 Result of performing a sliding-window (AB) or all-data, flexible time squares (CD) analysis on +synthetic data whose ground truth generative process either does not contain a time-varying +component (AC), or does contain it (CD). +This problem has put dynamic FC on the spot of criticism. Different approaches to generate +surrogate data and statistical tests have been proposed (see below), which tend to agree +that sliding-window approaches struggle to find genuine time-varying FC in most data, as the +variability of interest may drown in estimation noise. + +A +Amount of regularisation +0.8 +0.8 +Correlation +0.5 +0.6 +0.6 +Correlation/Partial Correlation +0.4 +0.4 +0.2 +0.5 +0.2 +0.2 +Volumes +Volumes +Empiricalcorrelation +Staticcorrelation +0.8 +Amount of regularisation +0.6 +Correlation +0.8 +0.4 +0.5 +0.2 +0.6 +0 +0.4 +-0.2 +-0.5 +0.2 +-0.4 +0.6 +Volumes +Volumes +B +D22 +3.1.2 All-data continuous estimators +One alternative to sliding windows within the paradigm of continuous estimators are methods +that estimate one correlation coefficient per time point by considering all the data set at +once. These methods do not use a predefined window, but the estimate at a given time point +still depends on neighbouring time points in an adaptive fashion. This is achieved by +imposing a regularisation constraint consisting of penalising the coefficient differences +between contiguous time points. One such method is flexible least squares (Kalaba & +Tesfatsion, 1989), which is implemented for example in the Dynamic BC toolbox (Liao et al., +2014). Multivariate approaches that estimate continuously-varying partial correlation +matrices also exist based on the same idea (Monti et al., 2014). Focussing on just two +signals, Figure 5C shows the behaviour of the flexible least squares method on the synthetic +data generated above, where there was no actual fluctuation of FC in the underlying +generative model of the data. As observed, choices with a low amount of regularisation +(redder lines) result also in largely varying estimates (even above the 1.0 boundary), while +higher amounts of regularisation converge to the empirical, non-varying correlation. When +there is actual variability in FC, Figure 5D shows that for some choices of the regularisation +parameter, the method is able to capture the modulation in the middle section of the time +series above and beyond the noise. +As we can see, while continuous approaches of this kind do not necessitate the specification +of a window length parameter, they are still quite dependent on the choice of the +regularisation parameter. However, the fact that, unlike sliding windows, these models have +a well-defined loss function lends to the use of quantitative metrics for selecting this +parameter, for instance based on cross-validation. Further, as the above simulation shows, +they can improve over the sliding-window estimator. On the other hand, they are more +complicated to implement and they are less readily available in standard software packages. + +23 +3.2 State-based estimators +As opposed to the continuous estimators discussed above, state-based estimators assume +that the data can be reasonably described using a discrete set of states (which can still be +large). This does not mean that by using these methods we assume any one-to-one, state- +to-mechanism mapping, or that there is any specific biophysical significance in the number +of states; this assumption is merely descriptive. Another important assumption of these +methods is exclusivity: only one state can be active at a given time. +A very important difference between continuous and state-based estimators is that, while for +a continuous estimator the FC parameters are essentially all we have to estimate, for a +state-based estimator we also have to estimate when the states occur (namely, the state +time courses), together with each state’s FC information. Another critical difference is that, +because the occurrences of these states can happen anytime in the time series, a state can +(and will probably) be assigned to several non-contiguous segments of the signal. +Consequently, while continuous estimators (like sliding windows) do not pool information +across subjects in any way, state-based estimators can have states that are shared across +subjects. State-based estimators are illustrated in Figure 4C. Here, each of the five states’ +FC is estimated based on several occurrences of varying length (dwell times) throughout the +timeseries. +These differences have an important practical implication: that the estimation of a state’s FC +is based on much more data than the estimation of a sliding window’s FC. For example, let +us suppose that we have 100 subjects with 20min of data per subject, and we run some +state-based estimator with 20 states; each state would have on average 100 minutes of +available data for the estimation of its FC parameters (e.g., a covariance or precision matrix), +as opposed to a typical sliding window of 1-2min. Now, if we scale up to larger data sets like +the HCP or the UK Biobank, states will have many hours of data available, effectively +making the estimation of FC extremely precise except for residual states. But, as mentioned, + +24 +in these methods we also have to estimate the state time courses, which is of course also +subject to estimation noise. We will briefly discuss below how to characterise and deal with +estimation noise at different levels. +We divide the state-based estimators into two classes: clustering methods (Allen et al., +2014) and generative models. Within the generative models, we focus on the hidden Markov +models (HMM) (Vidaurre et al., 2017). Conceptually, the main difference is that the former +has fewer assumptions and does not set up a model from which we can sample data. We +will compare them empirically later on. There are other state-based estimators, such as the +Kalman filter, that do not belong to any of these categories, but they are somewhat less +common in the field of neuroimaging and will not be discussed here. +3.2.1 Clustering approaches + +The most common clustering method builds upon sliding window estimates. Assuming we +have a pool of FC matrices (one per window), the simplest procedure would be to vectorise +the off-diagonal elements of the FC matrices, constructing a (no. of windows by pairs of +regions) matrix, on which we will then apply a clustering technique such as k-means (Allen et +al., 2014) or hierarchical clustering (Yang et al., 2014). By doing this, we would assign each +window to a different state, which constitutes a categorical state time course. Probabilistic +alternatives to k-means, such as a mixture of distributions, are also possible, in which case +the state time courses will be made of probabilities (that is, per window and state). Together +with the estimated state time courses, the state FC estimations would correspond to the +centre of the clusters. The hope of this method is that, even though the individual sliding +window estimates are quite noisy, the state FC matrices are made of averaging across many +windows and therefore will be less noisy. +A mathematically more principled version of this approach is based on the use of +Riemannian distances between the window FC matrices (instead of Euclidean, as results +from vectorising the FC matrices) (Pervaiz et al., 2020). + +25 +For simplicity, we will here illustrate only the simplest (and most common) approach with +code. Assuming we have already computed sliding-window estimates on data generated +using the code above, the following code implements the most basic clustering approach: +N_windows = total_session_duration - window_length + 1; +C_unwrapped = zeros(N_windows,p*(p-1)/2); +for j = 1:N_windows +Cj = C(:,:,j); Cj = Cj(triu(true(p),1)); +C_unwrapped(j,:) = Cj(:); +end +K = 4; +idx = kmeans(C_unwrapped,K); +state_time_courses = zeros(length(T),K); +for k = 1:K + state_time_courses(idx==k,k) = 1; +end +We can now get the fractional occupancies (i.e. the proportion of time taken by each state), +and display the state time courses as +mean(state_time_courses) +area(state_time_courses) +Another clustering approach that does not necessitate the specification of a window length +hyperparameter is Leading Eigenvector Dynamics Analysis (LEiDA) (Cabral et al., 2017). In +short, LEiDA computes functional connectivity estimates at each time point by first +computing the signals’ instantaneous phase (by means of the Hilbert transform) and then +calculating the cosine similarity between each pair of signals. Note that even though phase +is instantaneous, it still needs information from neighbouring time points. This yields a +relatively noisy estimation of functional connectivity per time point. To reduce the amount of +noise, the first eigenvector of each FC matrix is extracted using a singular value +decomposition, producing a (no. of time points by no. of regions) matrix containing FC +information. We can then apply a k-means algorithm as usual. For illustration, this procedure +is implemented in the code below: + +26 +Phase = zeros(total_session_duration,p); +for j=1:p +Phase(:,j) = angle(hilbert(X(:,j))); +end +Eigenvectors = zeros(total_session_duration,p); +for t = 1:total_session_duration + +C = zeros(p); +for j1 = 1:p-1 + + +for j2 = j1+1:p + + + +d = abs(Phase(t,j1) - Phase(t,j2)); + + + +if d > pi, d = 2*pi-d; end + + + +C(j1,j2) = cos(d); + + + +C(j2,j1) = C(j1,j2); + + +end +end +[v,d] = eig(C); +[~,i] = max(diag(d)); +Eigenvectors(t,:) = v(:,i); +end +[idx,eigen_centroids] = kmeans(Eigenvectors,K); +Here, eigen_centroids define the states. Each state essentially projects the brain areas +into a one-dimensional axis, such that areas close to each other in one extreme of the axis +(i.e., having the same sign in eigen_centroids) are in-phase, and areas that belong to +opposite extremes of the axis (i.e., having opposite signs) are anti-phase. Note that the sign +in eigen_centroids is arbitrary, and only the sign relationships between areas are +meaningful. +Overall, clustering approaches are generally simple to implement but they are not generative +models, i.e., we do not have a compact set of parameters to sample data or perform +statistical testing. The hidden Markov model (HMM), discussed in the next section, is a +generative model. +3.2.2 Generative models: Hidden Markov Models (HMMs) +Generative models are those that specify a full mathematical structure from which we can +sample new data sets (Bishop, 2006). As such, they define a probability distribution over the +data, and have a well-defined set of parameters over which we can do statistical inference + +27 +and testing. Here, we discuss the HMM, a generative model specifically designed to deal +with temporal (and other sequential) data (Vidaurre et al., 2018; Vidaurre et al., 2017). +The HMM is, rather than a single model, a family of probability distributions where each state +is in itself a probability distribution, defined by a set of state parameters. By choosing one or +another state distribution (also called observation models), the HMM can adapt to different +kinds of data. For example, state distributions can be Gaussian, multinomial, Wishart, +Poisson, etc. Here, we will focus on the Gaussian distribution because it is the common +choice for fMRI data. A Gaussian distribution is parametrised by a mean vector and a +covariance matrix containing FC information. This allows three main variants of the +Gaussian HMM. First, having a shared covariance matrix for all states and one mean vector +per state; this configuration however does not model changes in FC and will not be +discussed here. Second, we can have one mean vector and covariance matrix per state, +which is the most common approach. Third, in order to focus on FC, we can pin the state +means to zero and have only a covariance matrix per state (which is equivalent to having a +Wishart state distribution) (Vidaurre et al., 2021). +The remaining parameters of the HMM are the initial probabilities, i.e., the probability of the +trials to commence with a given state; and the transition probability matrix, with elements +(j,k) encoding the probability of transitioning from state j to state k. The initial probabilities, as +well as each of the rows of the transition probability matrix, are modelled as Dirichlet +distributions. +Apart from state exclusivity, and the fact that we use a discrete number of states, the use of +a transition probability matrix implies that the HMM builds upon a third assumption: +Markovianity. In the temporal domain, this means that which state is active in the present +time point depends on which state was active in the previous time point; or in more rigorous +terms, that the state at time point t is conditional independent to all the rest of the time points +given t-1 and t+1. In practice, this typically leads to having smoother, and not too abrupt, + +28 +state transitions. Note that even though the HMM does not model longer term dependencies, +the model does not forbid them either, and the resulting state time courses might still exhibit +long-term dependencies (Vidaurre et al., 2017). Versions of the HMM that explicitly model +the state visit durations, therefore dealing with longer term dependencies, are referred to as +semi-Markovian (HsMM). These have also been applied to fMRI data, in practice yielding +similar estimations to the standard HMM (Shappell et al., 2019). +Now, the inference of an HMM from data implies not only the estimation of the HMM +parameters just described but also the state time courses. While the HMM model can be +ported to different data sets, the state time courses are specific to the data set on which they +were estimated. To illustrate the use of the HMM and its inference, we will use the HMM- +MAR1 toolbox in Matlab. Assuming that we have added the toolbox to the Matlab path, we +can estimate a model by +options = struct(); +options.K = 8; +options.covtype = 'full'; +[hmm,Gamma,~,vpath] = hmmmar(X,T,options); +With this code, we are estimating an HMM model with K=8 states, and state time courses +Gamma, on data X, which might be composed of different concatenated fMRI sessions or +subjects. The length of each session is indicated in vector T (in number of time points). If we +wish to pin the mean of the state Gaussian distributions to zero such that the states are +defined only as FC matrices, we would use options.zeromean = 1. + +1 The name of the toolbox has a historical reason: the first observation model available was the +multivariate regressive model. But it now contains other distributions, such as Gaussian, Poisson, +Wishart, probabilistic PCA, or the multivariate regression model. All documentation is available at +https://github.com/OHBA-analysis/HMM-MAR/wiki + +29 +Using the same variables for T and options that we used for training, we can query the +fractional occupancies (the proportion of activation per state and subject) and plot the state +time courses as +FO = getFractionalOccupancy(Gamma,T,options) +for k = 1:8 + state_time_courses(vpath==k,k) = 1; +end +figure; imagesc(FO) +figure; area(state_time_courses(1:T(1),:)) +For interpretation purposes, it is important to note that, unlike the sliding window estimates +(which are typically correlations matrices), here a FC matrix is a covariance matrix and +therefore also conveys information about the variance of the signal. As mentioned above, +although the variance of an entire scanning session is one because of standardisation, the +variance for sub-periods of the signal might not be exactly one. As a consequence, the HMM +could be capturing changes in amplitude or variance in certain cases. +A final remark is about the estimation of the number of states. This could be done in different +ways, for example using the free energy. In real data this question is not very relevant +because the ground-truth number of states does not exist; more states will simply result in a +finer-grained estimation, and replicability is often a more sensible criterion (Vidaurre et al., +2017). +3.3 Multivariate autoregressive models +So far, we have been considering dynamic functional connectivity in terms of time-varying +instantaneous FC. In this section, we will describe the multivariate autoregressive model +(MAR), a linear dynamical system that models non-instantaneous aspects of FC with a +single set of parameters —i.e., without the use of states or windows (Harrison et al., 2003; +Rogers et al., 2010). +Mathematically, we can define the MAR model as the following generative model: + +30 +������������������������ = � ������������������������−������������ ������������������������ + ������������������������ +������������ +������������=1 + +Where ������������������������ is Gaussian noise, ������������������������ are matrices of autoregressive parameters and ������������ is the order +of the MAR model, which defines the complexity of the model and which we have to choose +beforehand. Order selection can make use of a pre-specified criterion, such Akaike’s +(Akaike, 1974), or use cross-validation; we will however not discuss this here since, in +practice, a precise estimation of the optimal order is not that relevant in neuroimaging. The +MAR model can be expanded with non-linear terms (e.g., area interactions), but we will not +discuss this possibility here either. We can reexpress the above equation in a standard +regression form as +������������������������ = ������������������������ ������������∗ + ������������������������ +where ������������������������ is defined as [������������������������−������������ . . . ������������������������] and ������������∗ is a (������������ ������������ by ������������) matrix that reunites all the ������������������������ +matrices across the ������������ channels. This allows us to straightforwardly apply any standard +regression estimation method to find the autoregressive parameters. For example, assuming +that there is only one subject in X, we can compute the autoregressive coefficients using +least squares: +Y = zeros(size(X,1) - L , L * p); +for j = 1:L +Y(:,(1:p)+(j-1)*p) = X(L-j+1:end-j,:); +end +A_star = pinv(Y) * X(L+1:end,:); +Although very simple to estimate, the MAR model contains rich information about the signal +on the frequency domain. Power, coherence, partial directed coherence, etc. can be readily +computed by Fourier transforming ������������∗ (in the code, A_star), and the richness of these +spectral estimates depends on the order of the model. This falls out of the scope of this +chapter; but see Faes and Nollo (2011) for a comprehensive account. + +31 +According to these equations, the diagonal elements of the ������������������������ matrices (and the +corresponding elements in ������������∗) refer to “self-connections”, and the off-diagonal elements refer +to cross-area connections, which we can interpret as dynamic FC. Note that, similarly to the +elements in a covariance matrix, the autoregressive coefficients are second-order statistics, +but are not instantaneous. +Conveniently, the autoregressive coefficients share with the precision matrix the property +that, for a very large amount of data, the coefficient(s) between variable i and variable j will +be exactly zero if they do not hold a direct linear (in this case lagged) relationship, i.e., after +we regress out the other variables. This has opened the window for the MAR model to be +used as a tool to infer effective connectivity. Critically, this is under the assumption that there +are no unobserved variables in the system. For example, we could get a non-zero coefficient +between two variables if these were connected through an unobserved third variable but not +directly. Of course, this assumption does not often hold in real neuroimaging data, but this +property could still be useful because, at least, if a coefficient is zero then we can assert that +the variables are not linearly connected. +Let us show the most basic scenario with an example, where we assume the following linear +dynamical system of order 1 with variables (brain areas) ������������(1),..., ������������(5): +������������(1)������������ = 0.5 ������������(1)������������−1 + 0.25 ������������(2)������������−1 + ������������(1)������������ +������������(2)������������ = 0.5 ������������(2)������������−1 + 0.25 ������������(1)������������−1 + 0.25 ������������(3)������������−1 + ������������(2)������������ +������������(3)������������ = 0.5 ������������(3)������������−1 + 0.25 ������������(2)������������−1 + 0.25 ������������(4)������������−1 + ������������(3)������������ +������������(4)������������ = 0.5 ������������(4)������������−1 + 0.25 ������������(3)������������−1 + 0.25 ������������(5)������������−1 + ������������(4)������������ +������������(5)������������ = 0.5 ������������(5)������������−1 + 0.25 ������������(4)������������−1 + ������������(5)������������ +And let us assume we sample data from this system with the following code: + +32 +p = 5; +total_session_duration = 1000; +true_A = 0.5*eye(p); +for j = 1:p-1 +true_A(j,j+1) = 0.25; true_A(j+1,j) = 0.25; +end +X = randn(total_session_duration,p); +for t = 2:total_session_duration +X(t,:) = X(t,:) + X(t-1,:) * true_A; +end +Now, if we sample many data sets like this and fit autoregressive models with the code +above, we would obtain distributions of estimated parameters as in Figure 6. If instead of +being able to sample many data sets, we had a limited number of subjects, we could still +perform a bootstrapped estimation (where we create pseudo data sets by randomly +sampling subjects with repetitions). + +Figure 6 Estimation of autoregressive coefficients over 1000 data sets generated through an idealised +linear dynamical system. +Of course, this is an extremely idealised scenario. In practice, the ground-truth is not an +autoregressive model, and autoregressive coefficients are rarely zero in real data. The +concept of Granger causality emerged as a dedicated statistical test in the field of economics +to deal with real data under a definition of causality based on temporal precedence. + +Region 1 +0 +0.2 +0.4 +0.6 +0 +0.2 +0.4 +0.6 +0.2 +0.4 +0.6 +0.2 +0.4 +0.6 +0 +0.2 +0.40.6 +Region 2 +0 +0.2 +0.4 +0.6 +0 +0.2 +0.40.6 +0.20.40.6 +0 +0.2 +0.40.6 +0 +0.2 +0.40.6 +Region 3 +0 +0.2 +0.4 +0.6 +0.2 +0.4 +0.6 +0 +0.2 +0.4 +0.6 +0 +0.2 +0.4 +0.6 +0.2 +0.4 +0.6 +Region 4 +0.2 +0.4 +0.6 +0.2 +0.4 +0.6 +0.20.40.6 +0 +0.2 +0.40.6 +0.20.40.6 +Region 5 +0.2 +0.4 +0.6 +0 +0.2 +0.4 +0.6 +0 +0.2 +0.4 +0.6 +0 +0.2 +0.40.6 +0 +0.20.40.6 +Region1 +Region2 +Region3 +Region4 +Region533 +However, the use of Granger causality in fMRI data for inferring effective connectivity has +been criticised because of the differences in latency in the hemodynamic response function +across brain regions, as well as other reasons. +Still, multivariate autoregressive modelling is a useful FC statistic in fMRI (as opposed to an +effective connectivity statistic), because it is easy to implement, it is relatively light in +assumptions while still informative, and, used as a predictor of behavioural traits, it can +surpass the accuracy of static FC estimates (Liégeois et al., 2019). +4 Evaluating dynamic FC models +In the previous section, we outlined some representative methods for dynamic FC +estimation. In this section, we discuss how to validate these models. +4.1 Testing against null models +A prevalent discussion in dynamic FC research (particularly in the sense of time-varying) is +the question whether FC meaningfully fluctuates over time, or whether it is actually stable +over time and temporal fluctuations are mainly due to noise. Depending on the criteria, it has +both been argued that real fMRI data fluctuates over time (Zalesky et al., 2014) and that it is +difficult to reject the statistical hypothesis that FC is stationary (Hindriks et al., 2016; Liégeois +et al., 2017; Lindquist et al., 2014). An argument for the latter view is that what may be +interpreted as dynamic FC can in fact be explained, for instance, by sampling variability +(Laumann et al., 2017), which may in turn be affected by denoising strategies. Simulation +studies also address whether static FC is being driven by transient (dynamic) events +(Zamani Esfahlani et al., 2020) or whether the presence of transient events in FC may be +driven by static FC (Ladwig et al., 2022; Novelli & Razi, 2022). Although some time-varying +FC methods are in principle equipped to find when the data is not time-varying (for example, +in a state-based model by removing all states but one), in practice they may find temporal + +34 +variation in FC due to statistical noise and sampling variability even if FC in the data was +really stationary. +We can see this effect of sampling variability by simulating data where FC does not vary +over time, using real data as a starting point. In this case, a null model is a model that +assumes stationarity of FC while preserving as many of the general distribution properties of +real fMRI data. We can, for example, generate data from a multivariate Gaussian distribution +that preserves the mean and variance of the real fMRI data: +T = 1000; +n_areas = size(tc_real,2); +mu = mean(tc_real); +Sigma = cov(tc_real); +tc_mvn = mvnrnd(mu, Sigma,T); +We can now compare the sliding window correlations between the real data and the +simulated data (for simplicity, we only look at the correlation between the first two areas): +window_length = 10; +real_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1); +mvn_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1); +for t = 1:T - window_length + 1 + real_dFC_tmp(:,:,t) = corr(tc_real(t:t+window_length-1,:)); + real_dFC(1,t) = real_dFC_tmp(1,2,t); + mvn_dFC_tmp(:,:,t) = corr(tc_mvn(t:t+window_length-1,:)); + mvn_dFC(1,t) = mvn_dFC_tmp(1,2,t); +end +As we show in Figure 7, for a sufficiently small window size, the sliding-window estimation of +FC in the surrogate data will vary considerably over time (see Figure 7B, middle panel; and +see also Figure 5), even though the true covariance of the data is static. We can increase +the window to see if we can recover the true (static) nature of the correlation: + +35 +window_length = 200; +real_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1); +mvn_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1); +for t = 1:T - window_length + 1 + real_dFC_tmp(:,:,t) = corr(tc_real(t:t+window_length-1,:)); + real_dFC(1,t) = real_dFC_tmp(1,2,t); + mvn_dFC_tmp(:,:,t) = corr(tc_mvn(t:t+window_length-1,:)); + mvn_dFC(1,t) = mvn_dFC_tmp(1,2,t); +end +As we increase the window size, the correlation in the surrogate data flattens out (see +Figure 7B, bottom panel), while we can still see some temporal variation in the sliding +window correlations of the real data (see Figure 7A, bottom panel). Here, it may seem as +though FC in the real fMRI data varies more over time than FC in the surrogate data. To test +this hypothesis, we would generate many random time courses from the multivariate +Gaussian distribution and test for significant differences in variance between the real fMRI +data and the surrogate data. However, multivariate Gaussian null data is not a very +appropriate null model in this case, since it does not capture any of the autocorrelations +present in real fMRI data. + +36 + +Figure 7 Is FC actually dynamic or are temporally changing estimates of FC a result of sampling variability? A) +Real fMRI timecourses for two regions (top row) and their corresponding sliding window correlations at a window +size of 10 TRs (middle row) and 200 TRs (bottom row). B) Surrogate timecourses for two regions, randomly +generated from a multivariate Gaussian distribution with the same mean and covariance as the real fMRI data +(top row), and their corresponding sliding window correlations at a window size of 10 TRs (middle row) and 200 +TRs (bottom row). +An important step in the development and evaluation of dynamic FC methods is therefore to +have suitable null models. However, defining the distribution of real fMRI data and setting +criteria for stationarity is not straightforward. +As a more adequate alternative to multivariate Gaussian models, we can use null models +that preserve not only the mean and variance, but also the autocorrelation structure of the +real fMRI data. We will demonstrate this following the example in Liégeois et al. (2017) to +generate surrogate data from a first order autoregressive (AR-1) model. To do this, we first +estimate the 1-lag autoregressive distribution parameters of the real fMRI data: + +A +Real data +B +Surrogate data (multivariate Gaussian) +300 +300 +200 +200 +de +100 +100 +plituc +100 +-100 +-200 +-200 +-300 +-300 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +0 +100 +200 +300 +400 +500 +600 +700 +800 +9001000 +Time +Time +Sliding window correlations (window size: 10 TR) +icient +Correlation coefficient +coeffic +0.5 +0.5 +Correlation +0.5 +-0.5 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +Time +Time +Sliding window correlations (window size: 200 TR) +icient +Correlation coefficient +coeffi +0.5 +rrelation +-0.5 +-0.5 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +Time +Time37 +Y = tc_real(2:end,:)'; +X(1,:) = ones(1,T-1); +X(2:n_areas+1,:) = tc_real(1:T-1, :)'; +beta = (Y*X')/(X*X'); +residuals = Y-beta*X; +We then generate synthetic data with these distribution parameters: +c = beta(:,1); +weights = beta(:, 2:n_areas+1); +tc_real_flip = tc_real'; +tc_arr = zeros(T, n_areas); +tc_arr(1,:) = tc_real_flip(:,randi(T-1))'; +rand_t = randperm(T-1); +res_mu = mean(residuals,2); +res_Sigma = cov(residuals'); +noise = (mvnrnd(res_mu,res_Sigma,T-1))'; +for i = 2:T + tc_arr(i,:) = c' + (weights*tc_arr(i-1,:)')' + ... + noise(:,rand_t(i-1))'; +end +Finally, we again compute the sliding window correlations in the surrogate timeseries and +compare them to the sliding window correlations of the real fMRI data we computed earlier: +window_length = 200; +arr_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1); +for t = 1:T - window_length + 1 + arr_dFC_tmp(:,:,t) = corr(tc_arr(t:t + window_length-1,:)); + arr_dFC(1,t) = arr_dFC_tmp(1,2,t); +end +The resulting surrogate timeseries and corresponding sliding window correlations are shown +in Figure 8B. Here, the sliding window correlations of the surrogate data have similar +temporal variance to the real fMRI data. + + +38 + +Figure 8 When generating surrogate data from a model that preserves the autocorrelation structure of the real +fMRI data, such as a 1-lag autoregressive randomisation model, the null hypothesis cannot easily be rejected. +In fact, several studies (Hindriks et al., 2016; Liégeois et al., 2017; Prichard & Theiler, 1994) +showed that for real fMRI data, the null hypothesis often cannot be rejected when using +surrogate data that preserves the autocorrelation structure of the real data, such as ARR +models or models based on phase randomisation (PR). The difference between the ARR +model and the PR model is how much of the autocovariance structure they preserve (the PR +model preserves the entire autocovariance structure, while the p-th order ARR model +preserves only the first p + 1 autocovariances). +It is however important to note that, since the assumption of both the AR model and the PR +model is that fMRI data are realisations of a linear, Gaussian, and stationary process, a +rejection of the null hypothesis in this case could be due to any of these assumptions being +untrue (i.e., not necessarily stationarity). That is, not being able to reject the null hypothesis +does not necessarily imply that the data is stationary. Liégeois et al. (2017) also showed that +the opposite conclusion also does not hold, i.e., being able to reject the null hypothesis does +not necessarily imply that the data is dynamic. While testing fMRI data for non-stationarity is +therefore not trivial, tests against null models can be useful to understand which information +different models use in their estimation of FC dynamics (Liégeois et al., 2021). + +A +Real data +B +Surrogate data (1st order autoregressive randomisation model) +300 +400 +200 +300 +200 +itude +Amplitude +100 +A +.100 +100 +-200 +-200 +-300 +-300 +U +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +1000 +Time +Time +Sliding window correlations (window size: 200 TR) +fficient +Correlation coefficient +0.5 +0.5 +coel +relation +Cor +-0.5 +-0.5 +100 +200 +300 +400 +500 +600 +700 +800 +900 +0 +100 +200 +300 +400 +500 +600 +700 +800 +900 +Time +Time39 +4.2 Comparing models with simulated data +An important question is how we can test the advantages and disadvantages of the different +methods. A powerful way to do this is through the use of synthetic data, where we can know +and control the ground-truth generative model. There are many ways to generate synthetic +data; here, we discuss as an illustration a simple procedure to compare the basic clustering +approach to the HMM. The idea is to use real fMRI data as a starting point. This approach is +described in detail in Ahrends et al. (2022). For example, let us assume that we have data +from ten regions in the brain from one subject. We can compute the covariance matrix of +those real data and generate data from a Gaussian distribution, which we can smooth to +make it look a bit more like real fMRI data: +some_time = 5000; +p = 10; +C = cov(real_data); +X = mvnrnd(zeros(some_time,p),C); +for j = 1:p +X(:,j) = smooth(X(:,j),10); +end +Now, from this empirical covariance matrix we can generate covariance matrices with small +variations. For this, we can perform a singular value decomposition of the empirical +covariance matrix, permute the low-order eigenvectors, and reassemble the matrix. +Depending on how many eigenvectors we permute (in the following code the last J +eigenvectors) and their corresponding eigenvalues, we can make the variations larger or +smaller: + +40 +[U,S,~] = svd(C); +e = diag(S); e = cumsum(e) / sum(e); +C_synth = zeros(size(C)); +% less than 90% variation +J = find(e>=0.9,1); +for j = 1:J +C_synth = C_synth + U(:,j) * S(j,j) * U(:,j)'; +end +for j = J+1:10 + +c = U(randperm(size(U,1)),j); +C_synth = C_synth + c * S(j,j) * c'; +end +We can use this code to generate a set of states, each with a different (but still relatively +similar) covariance matrix. This way, the larger the variations (i.e., the largest is the sum of +the eigenvalues of the permuted eigenvectors), the stronger the between-state differences +and the easier it is to find the ground-truth states by the HMM or any other method. +We can then sample ground-truth state time courses, for four states for example, as +total_session_duration = 20000; +max_duration = 100; min_duration = 5; +true_stc = zeros(total_session_duration,1); +t=1; +while t <= total_session_duration +k = randi(4,1); +L = randi(max_duration-min_duration,1) + min_duration; +if t+L-1 > total_session_duration + + +L = total_session_duration - t + 1; +end +true_stc(t:t+L-1,k) = 1; +t = t + L + 1; +end +In this case, we randomly sampled the identity of the state from a categorical distribution and +the duration of the state visits uniformly from the interval [min_duration,max_duration], +which corresponds to a form of semi-Markov process. Many other sampling schemes are +possible, with which we can interrogate the practical relevance of the models’ assumptions +(for instance, the HMM’s Markovianity; see below). + +41 +Altogether, we can put these different pieces together to sample synthetic data sets: sample +of covariance matrices, sample of state time courses, sample of data, and final smoothing. +Next, we will use simulated data from this scheme to compare the basic clustering approach +and the HMM. We followed these steps to generate 200 data sets for each of six different +scenarios: +- +Slower transitions (max_duration = 1000; min_duration = 50) and larger state +variations (J = 1). +- +Faster transitions (max_duration = 100; min_duration = 5) and larger state +variations (J = 1). +- +Slower transitions (max_duration = 1000; min_duration = 50) and moderate +state variations (J = 3). +- +Faster transitions (max_duration = 100; min_duration = 5) and moderate state +variations (J = 3). +- +Slower transitions (max_duration = 1000; min_duration = 50) and smaller state +variations (J = 6). +- +Faster transitions (max_duration = 100; min_duration = 5) and smaller state +variations (J = 6). +On each of these 4 x 200 data sets, we ran the basic clustering approach and the HMM, +both endowed with four states, matching the ground truth. Each method produced an +estimated state time course. Since the order of the states is not identifiable (i.e., the first +state in the ground-truth state time courses could correspond to the third state of the +estimated state time courses), we used the Hungarian algorithm (Kuhn, 1955) to align them +(as implemented in the HMM-MAR toolbox). We then computed the accuracy as the +correlation of the state probabilities with the ground-truth state time courses + +42 +options = struct(); +options.K = 4; +options.covtype = 'full'; +options.zeromean = 1; % states only have a covariance matrix +[hmm,stc] = hmmmar(X,size(X,1),options); +[assig,cost] = munkres(1-corr(true_stc,stc)); +stc = stc(:,assig); +accuracy = corr(stc(:),true_stc(:)); +Figure 9 shows the accuracy of each run per method and scenario (dots where a jitter was +introduced in the x-axis for ease of visualisation), together with the across-run average +(bars). As observed, the HMM fared better than the basic clustering approach for two +reasons. First, because the estimation of the states for the clustering method is based on +noisy sliding window estimates, whereas the HMM estimates the states directly from the +data using a much larger number of time points and, therefore, does it so much more +precisely; for this reason, the difference between the HMM and the clustering approach is +considerably smaller when the state covariance matrices are very different between states +(upper row of panels). Second, because the clustering approach cannot detect fast changes +(i.e., faster than the length of the window), whereas the HMM states can switch as quickly as +necessary; that is why the accuracies for clustering approach are lower in the right panels, +whereas the speed of the ground-truth state transitions makes less of a difference for the +HMM. + +43 + +Figure 9 Performance of the standard clustering approach vs. the HMM in estimating the state +transitions sampled from a state-based generative model where the state distributions are covariance +matrices. Both how different the states are, and the speed of the transitions are manipulated to create +simpler or harder estimation problems. Each dot represents a run (a sampled data set and the +corresponding estimates) and the bars are averages. +Although not shown here, LEiDA performed notably worse than the other methods. The +reason is that, while LEiDA’s states are based on the first principal component of matrices of +phase relationships, the states of both the HMM and the clustering approach are actual +covariance matrices, matching the ground-truth state model. This is a general principle of +this type of simulations: the more accurate the assumptions of a method are with respect to +the ground-truth generative process, the more accurate its estimates will be. It is important to +note that, in real data, where the generator of the data is the brain, LEiDA will just offer a +different perspective: in this case, we can hardly say that one model is more correct than +another. + +Slower transitions, larger state variations +Fastertransitions,largerstate variations +0.8 +0.8 +Accuracy +racy +0.6 +0.6 +Accur +0.4 +0.4 +0.2 +0.2 +0 +0 +Clustering +HMM +Clustering +HMM +Slower transitions, medium state variations +Faster transitions, medium state variations +1 +0.8 +.0.8 +racy +0.6 +0.6 +Accur +ccur +0.4 +0.4 +A +0.2 +0.2 +0 +0 +Clustering +HMM +Clustering +HMM +Slower transitions, smaller state variations +Faster transitions, smaller state variations +0.8 +0.8 +Accuracy +0.6 +0.6 +0.4 +0.4 +0.2 +0.2 +0 +0 +Clustering +HMM +Clustering +HMM44 +5 Challenges & perspectives +Having discussed the most important practical aspects of conducting dynamic FC analyses, +we will now outline some challenges and perspectives in the current development of +dynamic FC. Challenges and questions in the dynamic FC research field are also +extensively discussed in Lurie et al. (2019), so we here only highlight a few key ideas related +to reliability and reproducibility, and the use of dynamic FC to predict behavioural and clinical +traits. +5.1 Reliability & reproducibility +A key challenge in the study of dynamic FC is the reliability and reproducibility of different +methods. As a consequence, it can be difficult to compare results from dynamic FC studies. +Reliability and reproducibility in dynamic FC studies may mean several things. Reliability +may pertain to the question: When we have several scanning sessions of a subject, do we +reliably find the same patterns of dynamic FC in all sessions? We consider this the question +of test-retest reliability. Studies addressing this question have shown that test-retest +reliability of many dynamic FC methods is low, particularly for sliding window-based +approaches (Choe et al., 2017; Zhang et al., 2018). To what extent this is due to +methodological limitations or systematic within-subject variability is an important topic for +future research into dynamic FC reliability (Geerligs et al., 2015). +Reliability may also refer to the question: How reliable is the method in detecting dynamic +FC? This question can be addressed by simulation studies or by studying how biased a +method is by secondary parameters. For instance, Ahrends et al. (2022) showed how a +HMM’s ability to detect temporal changes in FC depends on several parameters that are +determined by preprocessing strategies. Optimising preprocessing strategies (as described + +45 +in the previous section Preparing data for dynamic FC analyses) is therefore an important +step in improving reliability. +Other studies have investigated the reproducibility of dynamic FC methods. Reproducibility +may refer to the question: Do we find consistent, canonical patterns of resting-state dynamic +FC across multiple datasets? For instance, Abrol et al. (2017) tested this question across +datasets from different scanning sites. Using two different state-based approaches, they +could show that the same basic connectivity patterns emerge in all datasets, indicating that +general patterns in dynamic FC are reproducible and robust to variations in the specific +datasets. +In some methods, reproducibility also entails the question: Do we get the same results when +we run the model several times? In models that depend on random initialisation, solutions +may to some extent be different every time the model is run (Vidaurre et al., 2019). This +issue may also be referred to as robustness. Vidaurre et al. (2018) have shown that dynamic +FC results from HMMs which use random initialisation for the inference are reproducible +across different runs of the model to some extent, but that the similarity between runs +depends on the dataset. Robustness may also refer to the issue of how the model deals with +noise, e.g., in terms of noisy fMRI recordings, or in terms of outliers in the dataset, i.e., does +the estimation of dynamic FC in a group of subjects suffer if single subjects display +anomalous patterns of dynamic FC. This question is only starting to be addressed for +different dynamic FC methods. +Assessing and improving reliability, reproducibility, and robustness of dynamic FC should be +a central task in future dynamic FC research. This may not only improve the estimation of +dynamic FC but also make methods and results more comparable. + +46 +5.2 Predicting behaviour and individual traits from dynamic FC +Time-varying FC measures have been shown to be highly individual, suggesting that they +could be used for “fingerprinting” an individual (J. Liu et al., 2018). Beyond explaining +behaviour, there is thus growing interest in using dynamic FC to predict individual traits and +behaviour with a potential to develop dynamic FC-based biomarkers. +So far, studies using features derived from dynamic FC were able to predict various +phenotypes and behavioural measures at moderate to fair accuracy, such as cognitive +performance or intelligence (J. Liu et al., 2018; Sen & Parhi, 2021; Shine et al., 2019), task +performance (Fong et al., 2019), or sleep quality (Zhou et al., 2020), as well as to classify +clinical diagnoses, such as schizophrenia (Bhinge et al., 2019; Cetin et al., 2016) or post- +traumatic stress disorder (PTSD) (Ou et al., 2015). This is an interesting avenue, because +dynamic FC contains information complementary to structural measures or static FC (Ge et +al., 2017; Liégeois et al., 2019; Vidaurre et al., 2021), and so it may be able to predict +distinct aspects of behaviour. It has in fact been shown that dynamic FC can outperform +structural and static measures in predicting behaviour (Saha et al., 2021; Vergara et al., +2020; Vidaurre et al., 2021) and in classifying neurological and psychiatric disease (Jin et al., +2017; Rashid et al., 2016). For a review on prediction and classification from static and +dynamic FC in clinical applications, see Du et al. (2018). +But how can we best use the information from dynamic FC models in the context of +prediction problems? One option is to select a feature of interest, such as the occurrence of +a specific network, as a predictor. This is the approach taken in most studies so far. +However, not only is it not obvious how to choose the relevant feature in the context of +neuroimaging (Du et al., 2018; Wolfers et al., 2015), doing this would also mean losing +information that was originally contained in the dynamic FC model. Instead, we may want to +use all information from subject-level dynamic FC models in a linear predictive model to +predict these subjects’ phenotype. While static measures can relatively straightforwardly + +47 +(with the exception of geometric constraints) be used in linear predictive models, the +temporal dimension and complexity of dynamic FC models make this more difficult. +Developing new approaches to predicting from dynamic FC models is a central goal in +current dynamic FC research. One promising avenue are kernel methods (Shawe-Taylor & +Cristianini, 2004). These methods have several advantages: They allow working with high- +dimensional features such as dynamic FC features in a computationally efficient way by +using a subject-by-subject similarity matrix for the prediction rather than the features +themselves. Kernels can be constructed in a way that preserves the structure of the features +(e.g. the scales and relationships between parameters of an underlying dynamic FC model). +Furthermore, kernels can be used straightforwardly in linear prediction models or classifiers. +Indeed, by using kernel functions that apply a nonlinear transformation to the data, these +linear prediction models or classifiers can be used to detect nonlinear decision boundaries. +This elegantly combines the advantages of linear models, which are computationally +efficient, easy to implement, and can be more readily interpreted, with nonlinear models, +which allow finding more complex relationships and patterns in data. Kernel methods also +open the door to predicting from multiple modalities, such as structural and functional, static +and dynamic information, or MRI and MEG/EEG (Engemann et al., 2020; Schouten et al., +2016). Using a Multi-kernel learning (MKL) approach (Gönen & Alpaydın, 2011), separately +constructed kernels for each modality can be combined within a single prediction model. +This approach has been shown to improve prediction accuracy, for example when combining +MRI and MEG (Vaghari et al., 2022) or several levels of FC estimation (Zhang et al., 2017). +6 Conclusions +In this chapter, we have given a practical introduction to different methods of dynamic FC. +We also discussed data preparation and model evaluation, as well as some challenges and +perspectives in the dynamic FC research field. As we have shown, there is a variety of + +48 +methods to estimate dynamic FC, each with unique advantages but also shortcomings. +Recent years have seen an explosion of dynamic FC studies, indicating the great potential of +a dynamic view of FC in answering specific research questions and in contributing to our +general understanding of brain function. However, we also stress the importance of +methodological rigour in the further development of these methods. For instance, dynamic +FC models may be more heavily biased by preprocessing strategies than other methods, +because they are more sensitive to temporal noise. Another not fully resolved issue is the +comparison to null models that assume stationarity, an important limitation that should be +kept in mind in both methods development and the conceptualisation of dynamic FC. Going +forward, we hope to see further work on these issues, as well as explicit testing and +improvement of the reliability and reproducibility of the different methods. +References +Abrol, A., Damaraju, E., Miller, R. L., Stephen, J. M., Claus, E. D., Mayer, A. R., & Calhoun, +V. D. (2017). Replicability of time-varying connectivity patterns in large resting state +fMRI samples. Neuroimage, 163, 160-176. +https://doi.org/10.1016/j.neuroimage.2017.09.020 +Afyouni, S., Smith, S. M., & Nichols, T. E. (2019). Effective degrees of freedom of the +Pearson's correlation coefficient under autocorrelation. Neuroimage, 199, 609-625. +https://doi.org/10.1016/j.neuroimage.2019.05.011 +Ahrends, C., Stevner, A., Pervaiz, U., Kringelbach, M. L., Vuust, P., Woolrich, M. W., & +Vidaurre, D. (2022). Data and model considerations for estimating time-varying +functional connectivity in fMRI. Neuroimage, 252, 119026. +https://doi.org/10.1016/j.neuroimage.2022.119026 +Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on +Automatic Control, 19(6), 716-723. https://doi.org/10.1109/TAC.1974.1100705 +Allen, E. A., Damaraju, E., Plis, S. M., Erhardt, E. B., Eichele, T., & Calhoun, V. D. (2014). +Tracking Whole-Brain Connectivity Dynamics in the Resting State. Cerebral Cortex, +24(3), 663-676. https://doi.org/10.1093/cercor/bhs352 +Beckmann, C. F., Mackay, C. E., Filippini, N., & Smith, S. M. (2009). Group comparison of +resting-state FMRI data using multi-subject ICA and dual regression. Neuroimage, +47, S148. https://doi.org/10.1016/S1053-8119(09)71511-3 +Bhinge, S., Long, Q., Calhoun, V. D., & Adali, T. (2019). Spatial Dynamic Functional +Connectivity Analysis Identifies Distinctive Biomarkers in Schizophrenia. Front +Neurosci, 13, 1006. https://doi.org/10.3389/fnins.2019.01006 + +49 +Bijsterbosch, J., Harrison, S., Duff, E., Alfaro-Almagro, F., Woolrich, M., & Smith, S. (2017). +Investigations into within- and between-subject resting-state amplitude variations. +Neuroimage, 159, 57-69. https://doi.org/10.1016/j.neuroimage.2017.07.014 +Bishop, C. (2006). Pattern Recognition and Machine Learning. Springer-Verlag. +Cabral, J., Vidaurre, D., Marques, P., Magalhaes, R., Silva Moreira, P., Miguel Soares, J., +Deco, G., Sousa, N., & Kringelbach, M. L. (2017). Cognitive performance in healthy +older adults relates to spontaneous switching between states of functional +connectivity during rest. Sci Rep, 7(1), 5135. https://doi.org/10.1038/s41598-017- +05425-7 +Cai, B., Zhang, G., Zhang, A., Stephen, J. M., Wilson, T. W., Calhoun, V. D., & Wang, Y. +(2018). Capturing Dynamic Connectivity from Resting State fMRI using Time-Varying +Graphical Lasso. IEEE Trans Biomed Eng. +https://doi.org/10.1109/tbme.2018.2880428 +Calhoun, Vince D., Miller, R., Pearlson, G., & Adalı, T. (2014). The Chronnectome: Time- +Varying Connectivity Networks as the Next Frontier in fMRI Data Discovery. Neuron, +84(2), 262-274. https://doi.org/10.1016/j.neuron.2014.10.015 +Cetin, M. S., Houck, J. M., Rashid, B., Agacoglu, O., Stephen, J. M., Sui, J., Canive, J., +Mayer, A., Aine, C., Bustillo, J. R., & Calhoun, V. D. (2016). Multimodal Classification +of Schizophrenia Patients with MEG and fMRI Data Using Static and Dynamic +Connectivity Measures. Front Neurosci, 10, 466. +https://doi.org/10.3389/fnins.2016.00466 +Choe, A. S., Nebel, M. B., Barber, A. D., Cohen, J. R., Xu, Y., Pekar, J. J., Caffo, B., & +Lindquist, M. A. (2017). Comparing test-retest reliability of dynamic functional +connectivity methods. Neuroimage, 158, 155-175. +https://doi.org/10.1016/j.neuroimage.2017.07.005 +Ciric, R., Wolf, D. H., Power, J. D., Roalf, D. R., Baum, G. L., Ruparel, K., Shinohara, R. T., +Elliott, M. A., Eickhoff, S. B., Davatzikos, C., Gur, R. C., Gur, R. E., Bassett, D. S., & +Satterthwaite, T. D. (2017). Benchmarking of participant-level confound regression +strategies for the control of motion artifact in studies of functional connectivity. +Neuroimage, 154, 174-187. https://doi.org/10.1016/j.neuroimage.2017.03.020 +Du, Y., Fu, Z., & Calhoun, V. D. (2018). Classification and Prediction of Brain Disorders +Using Functional Connectivity: Promising but Challenging. Front Neurosci, 12, 525. +https://doi.org/10.3389/fnins.2018.00525 +Engemann, D. A., Kozynets, O., Sabbagh, D., Lemaître, G., Varoquaux, G., Liem, F., & +Gramfort, A. (2020). Combining magnetoencephalography with magnetic resonance +imaging enhances learning of surrogate-biomarkers. eLife, 9, e54055. +https://doi.org/10.7554/eLife.54055 +Faes, L., & Nollo, G. (2011). Multivariate frequency domain analysis of causal interactions in +physiological time series. In A. Laskovski (Ed.), Biomedical Engineering, Trends in +Electronics, Communications and Software (Vol. 8, pp. 403-428). InTech. +http://www.intechopen.com/books/biomedical-engineering-trends-in-electronics- +communications-and-software/multivariate-frequency-domain-analysis-of-causal- +interactions-in-physiological-time-series +Filippi, M. (2016). fMRI Techniques and Protocols (2nd 2016. ed.). Springer New York. +https://doi.org/10.1007/978-1-4939-5611-1 +Fong, A. H. C., Yoo, K., Rosenberg, M. D., Zhang, S., Li, C.-S. R., Scheinost, D., Constable, +R. T., & Chun, M. M. (2019). Dynamic functional connectivity during task + +50 +performance and rest predicts individual differences in attention across studies. +Neuroimage, 188, 14-25. https://doi.org/10.1016/j.neuroimage.2018.11.057 +Frahm, J., Bruhn, H., Merboldt, K. D., & Hänicke, W. (1992). Dynamic MR imaging of human +brain oxygenation during rest and photic stimulation. J Magn Reson Imaging, 2(5), +501-505. https://doi.org/10.1002/jmri.1880020505 +Friedman, J., Hastie, T., & Tibshirani, R. (2008). Sparse inverse covariance estimation with +the graphical lasso. Biostatistics, 9(3), 432-441. +https://doi.org/10.1093/biostatistics/kxm045 +Friston, K. J. (2011). Functional and effective connectivity: a review. Brain Connect, 1(1), 13- +36. https://doi.org/10.1089/brain.2011.0008 +Friston, K. J., Ashburner, J. T., Kiebel, S. J., Nichols, T. E., & Penny, W. D. (2006). +Statistical Parametric Mapping: The Analysis of Functional Brain Images. Elsevier +Science & Technology. https://doi.org/10.1016/B978-0-12-372560-8.X5000-1 +Ge, T., Holmes, A. J., Buckner, R. L., Smoller, J. W., & Sabuncu, M. R. (2017). Heritability +analysis with repeat measurements and its application to resting-state functional +connectivity. Proc Natl Acad Sci U S A, 114(21), 5521-5526. +https://doi.org/10.1073/pnas.1700765114 +Geerligs, L., Rubinov, M., Cam, C., & Henson, R. N. (2015). State and Trait Components of +Functional Connectivity: Individual Differences Vary with Mental State. J Neurosci, +35(41), 13949-13961. https://doi.org/10.1523/jneurosci.1324-15.2015 +Glasser, M. F., Sotiropoulos, S. N., Wilson, J. A., Coalson, T. S., Fischl, B., Andersson, J. L., +Xu, J., Jbabdi, S., Webster, M., Polimeni, J. R., Van Essen, D. C., & Jenkinson, M. +(2013). The minimal preprocessing pipelines for the Human Connectome Project. +Neuroimage, 80, 105-124. https://doi.org/10.1016/j.neuroimage.2013.04.127 +Griffanti, L., Salimi-Khorshidi, G., Beckmann, C. F., Auerbach, E. J., Douaud, G., Sexton, C. +E., Zsoldos, E., Ebmeier, K. P., Filippini, N., Mackay, C. E., Moeller, S., Xu, J., +Yacoub, E., Baselli, G., Ugurbil, K., Miller, K. L., & Smith, S. M. (2014). ICA-based +artefact removal and accelerated fMRI acquisition for improved resting state network +imaging. Neuroimage, 95, 232-247. +https://doi.org/10.1016/j.neuroimage.2014.03.034 +Gönen, M., & Alpaydın, E. (2011). Multiple kernel learning algorithms. The Journal of +Machine Learning Research, 12, 2211-2268. +http://jmlr.org/papers/v12/gonen11a.html +Hair, J. F. (2009). Multivariate data analysis (7. ed.). Prentice Hall. +https://digitalcommons.kennesaw.edu/facpubs/2925/ +Hair, J. F. (2009). Multivariate data analysis. +Harrison, L., Penny, W. D., & Friston, K. (2003). Multivariate autoregressive modeling of +fMRI time series. Neuroimage, 19(4), 1477-1491. https://doi.org/10.1016/s1053- +8119(03)00160-5 +Hindriks, R., Adhikari, M. H., Murayama, Y., Ganzetti, M., Mantini, D., Logothetis, N. K., & +Deco, G. (2016). Can sliding-window correlations reveal dynamic functional +connectivity in resting-state fMRI? Neuroimage, 127, 242-256. +https://doi.org/10.1016/j.neuroimage.2015.11.055 +Hutchison, R. M., Womelsdorf, T., Allen, E. A., Bandettini, P. A., Calhoun, V. D., Corbetta, +M., Della Penna, S., Duyn, J. H., Glover, G. H., Gonzalez-Castillo, J., Handwerker, D. + +51 +A., Keilholz, S., Kiviniemi, V., Leopold, D. A., de Pasquale, F., Sporns, O., Walter, M., +& Chang, C. (2013). Dynamic functional connectivity: Promise, issues, and +interpretations [Article]. Neuroimage, 80, 360-378. +https://doi.org/10.1016/j.neuroimage.2013.05.079 +Iraji, A., Fu, Z., DeRamus, T., Qi, S., Rachakonda, S., Du, Y., & Calhoun, V. (2020). Ultra- +high-order ICA: fine overlapping functional parcellations and spatiotemporal +reconfiguration. bioRxiv, 2020.2003.2011.986299. +https://doi.org/10.1101/2020.03.11.986299 +Jin, C., Jia, H., Lanka, P., Rangaprakash, D., Li, L., Liu, T., Hu, X., & Deshpande, G. (2017). +Dynamic brain connectivity is a better predictor of PTSD than static connectivity +[Article]. Human Brain Mapping, 38(9), 4479-4496. +https://doi.org/10.1002/hbm.23676 +Kalaba, R., & Tesfatsion, L. (1989). Time-varying linear regression via flexible least squares. +Computers & Mathematics with Applications, 17(8), 1215-1245. +https://doi.org/10.1016/0898-1221(89)90091-6 +Kuhn, H. W. (1955). The Hungarian method for the assignment problem. Naval Research +Logistics (NRL), 2, 83–97. https://doi.org/10.1007/978-3-540-68279-0_2 +Kwong, K. K., Belliveau, J. W., Chesler, D. A., Goldberg, I. E., Weisskoff, R. M., Poncelet, B. +P., Kennedy, D. N., Hoppel, B. E., Cohen, M. S., Turner, R., Cheng, H.-M., Brady, T. +J., & Rosen, B. R. (1992). Dynamic Magnetic Resonance Imaging of Human Brain +Activity During Primary Sensory Stimulation. Proceedings of the National Academy of +Sciences of the United States of America, 89(12), 5675-5679. +http://www.jstor.org/stable/2359716 +Ladwig, Z., Seitzman, B. A., Dworetsky, A., Yu, Y., Adeyemo, B., Smith, D. M., Petersen, S. +E., & Gratton, C. (2022). BOLD cofluctuation ‘events’ are predicted from static +functional connectivity. Neuroimage, 260, 119476. +https://doi.org/10.1016/j.neuroimage.2022.119476 +Laumann, T. O., Snyder, A. Z., Mitra, A., Gordon, E. M., Gratton, C., Adeyemo, B., Gilmore, +A. W., Nelson, S. M., Berg, J. J., Greene, D. J., McCarthy, J. E., Tagliazucchi, E., +Laufs, H., Schlaggar, B. L., Dosenbach, N. U. F., & Petersen, S. E. (2017). On the +Stability of BOLD fMRI Correlations. Cerebral cortex (New York, N.Y. : 1991), 27(10), +4719-4732. https://doi.org/10.1093/cercor/bhw265 +Lee, M. H., Smyser, C. D., & Shimony, J. S. (2013). Resting-state fMRI: a review of methods +and clinical applications. AJNR Am J Neuroradiol, 34(10), 1866-1872. +https://doi.org/10.3174/ajnr.A3263 +Lehmann, B. C. L., White, S. R., Henson, R. N., Cam, C. A. N., & Geerligs, L. (2017). +Assessing dynamic functional connectivity in heterogeneous samples. Neuroimage, +157, 635-647. https://doi.org/10.1016/j.neuroimage.2017.05.065 +Liao, W., Wu, G. R., Xu, Q., Ji, G. J., Zhang, Z., Zang, Y. F., & Lu, G. (2014). DynamicBC: a +MATLAB toolbox for dynamic brain connectome analysis. Brain Connect, 4(10), 780- +790. https://doi.org/10.1089/brain.2014.0253 +Liégeois, R., Laumann, T. O., Snyder, A. Z., Zhou, J., & Yeo, B. T. T. (2017). Interpreting +temporal fluctuations in resting-state functional connectivity MRI. Neuroimage, 163, +437-455. https://doi.org/10.1016/j.neuroimage.2017.09.012 +Liégeois, R., Li, J., Kong, R., Orban, C., Van De Ville, D., Ge, T., Sabuncu, M. R., & Yeo, B. +T. T. (2019). Resting brain dynamics at different timescales capture distinct aspects + +52 +of human behavior. Nat Commun, 10(1), 2317. https://doi.org/10.1038/s41467-019- +10317-7 +Liégeois, R., Yeo, B. T. T., & Van De Ville, D. (2021). Interpreting null models of resting-state +functional MRI dynamics: not throwing the model out with the hypothesis. +Neuroimage, 243, 118518. https://doi.org/10.1016/j.neuroimage.2021.118518 +Lindquist, M. A., Xu, Y., Nebel, M. B., & Caffo, B. S. (2014). Evaluating dynamic bivariate +correlations in resting-state fMRI: A comparison study and a new approach. +Neuroimage, 101, 531-546. https://doi.org/10.1016/j.neuroimage.2014.06.052 +Liu, J., Liao, X., Xia, M., & He, Y. (2018). Chronnectome fingerprinting: Identifying individuals +and predicting higher cognitive functions using dynamic brain connectivity patterns. +Hum Brain Mapp, 39(2), 902-915. https://doi.org/10.1002/hbm.23890 +Liu, X., Zhang, N., Chang, C., & Duyn, J. H. (2018). Co-activation patterns in resting-state +fMRI signals. Neuroimage, 180(Pt B), 485-494. +https://doi.org/10.1016/j.neuroimage.2018.01.041 +Lurie, D. J., Kessler, D., Bassett, D. S., Betzel, R. F., Breakspear, M., Kheilholz, S., Kucyi, +A., Liégeois, R., Lindquist, M. A., McIntosh, A. R., Poldrack, R. A., Shine, J. M., +Thompson, W. H., Bielczyk, N. Z., Douw, L., Kraft, D., Miller, R. L., Muthuraman, M., +Pasquini, L., . . . Calhoun, V. D. (2019). Questions and controversies in the study of +time-varying functional connectivity in resting fMRI. Network Neuroscience, 4(1), 30- +69. https://doi.org/10.1162/netn_a_00116 +Lydon-Staley, D. M., Ciric, R., Satterthwaite, T. D., & Bassett, D. S. (2019). Evaluation of +confound regression strategies for the mitigation of micromovement artifact in studies +of dynamic resting-state functional connectivity and multilayer network modularity. +Network Neuroscience, 3(2), 427-454. https://doi.org/10.1162/netn_a_00071 +Margulies, D. S., Ghosh, S. S., Goulas, A., Falkiewicz, M., Huntenburg, J. M., Langs, G., +Bezgin, G., Eickhoff, S. B., Castellanos, F. X., Petrides, M., Jefferies, E., & +Smallwood, J. (2016). Situating the default-mode network along a principal gradient +of macroscale cortical organization. Proceedings of the National Academy of +Sciences, 113(44), 12574-12579. https://doi.org/10.1073/pnas.1608282113 +Monti, R. P., Hellyer, P., Sharp, D., Leech, R., Anagnostopoulos, C., & Montana, G. (2014). +Estimating time-varying brain connectivity networks from functional MRI time series. +Neuroimage, 103, 427-443. https://doi.org/10.1016/j.neuroimage.2014.07.033 +Murphy, K., & Fox, M. D. (2017). Towards a consensus regarding global signal regression +for resting state functional connectivity MRI. Neuroimage, 154, 169-173. +https://doi.org/10.1016/j.neuroimage.2016.11.052 +Nalci, A., Rao, B. D., & Liu, T. T. (2019). Nuisance effects and the limitations of nuisance +regression in dynamic functional connectivity fMRI. Neuroimage, 184, 1005-1031. +https://doi.org/10.1016/j.neuroimage.2018.09.024 +Novelli, L., & Razi, A. (2022). A mathematical perspective on edge-centric brain functional +connectivity. Nature Communications, 13(1), 2693. https://doi.org/10.1038/s41467- +022-29775-7 +Ogawa, S., Tank, D. W., Menon, R., Ellermann, J. M., Kim, S. G., Merkle, H., & Ugurbil, K. +(1992). Intrinsic signal changes accompanying sensory stimulation: functional brain +mapping with magnetic resonance imaging. Proc Natl Acad Sci U S A, 89(13), 5951- +5955. https://doi.org/10.1073/pnas.89.13.5951 + +53 +Ou, J., Xie, L., Jin, C., Li, X., Zhu, D., Jiang, R., Chen, Y., Zhang, J., Li, L., & Liu, T. (2015). +Characterizing and Differentiating Brain State Dynamics via Hidden Markov Models. +Brain Topography, 28(5), 666-679. https://doi.org/10.1007/s10548-014-0406-2 +Parkes, L., Fulcher, B., Yücel, M., & Fornito, A. (2018). An evaluation of the efficacy, +reliability, and sensitivity of motion correction strategies for resting-state functional +MRI. Neuroimage, 171, 415-436. https://doi.org/10.1016/j.neuroimage.2017.12.073 +Pervaiz, U., Vidaurre, D., Woolrich, M. W., & Smith, S. M. (2020). Optimising network +modelling methods for fMRI. Neuroimage, 211, 116604. +https://doi.org/10.1016/j.neuroimage.2020.116604 +Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2012). +Spurious but systematic correlations in functional connectivity MRI networks arise +from subject motion. Neuroimage, 59(3), 2142-2154. +https://doi.org/10.1016/j.neuroimage.2011.10.018 +Power, Jonathan D., Cohen, Alexander L., Nelson, Steven M., Wig, Gagan S., Barnes, +Kelly A., Church, Jessica A., Vogel, Alecia C., Laumann, Timothy O., Miezin, +Fran M., Schlaggar, Bradley L., & Petersen, Steven E. (2011). Functional Network +Organization of the Human Brain. Neuron, 72(4), 665-678. +https://doi.org/10.1016/j.neuron.2011.09.006 +Preti, M. G., Bolton, T. A. W., & Van De Ville, D. (2017). The dynamic functional +connectome: State-of-the-art and perspectives. Neuroimage, 160, 41-54. +https://doi.org/10.1016/j.neuroimage.2016.12.061 +Preti, M. G., & Van De Ville, D. (2017). Dynamics of functional connectivity at high spatial +resolution reveal long-range interactions and fine-scale organization. Scientific +Reports, 7(1), 12773. https://doi.org/10.1038/s41598-017-12993-1 +Prichard, D., & Theiler, J. (1994). Generating surrogate data for time series with several +simultaneously measured variables. Phys Rev Lett, 73(7), 951-954. +https://doi.org/10.1103/PhysRevLett.73.951 +Rashid, B., Arbabshirani, M. R., Damaraju, E., Cetin, M. S., Miller, R., Pearlson, G. D., & +Calhoun, V. D. (2016). Classification of schizophrenia and bipolar patients using +static and dynamic resting-state fMRI brain connectivity. Neuroimage, 134, 645-657. +https://doi.org/10.1016/j.neuroimage.2016.04.051 +Rogers, B. P., Katwal, S. B., Morgan, V. L., Asplund, C. L., & Gore, J. C. (2010). Functional +MRI and multivariate autoregressive models. Magn Reson Imaging, 28(8), 1058- +1065. https://doi.org/10.1016/j.mri.2010.03.002 +Saha, D. K., Damaraju, E., Rashid, B., Abrol, A., Plis, S. M., & Calhoun, V. D. (2021). A +Classification-Based Approach to Estimate the Number of Resting Functional +Magnetic Resonance Imaging Dynamic Functional Connectivity States. Brain +Connect, 11(2), 132-145. https://doi.org/10.1089/brain.2020.0794 +Schouten, T. M., Koini, M., de Vos, F., Seiler, S., van der Grond, J., Lechner, A., +Hafkemeijer, A., Möller, C., Schmidt, R., de Rooij, M., & Rombouts, S. A. R. B. +(2016). Combining anatomical, diffusion, and resting state functional magnetic +resonance imaging for individual classification of mild and moderate Alzheimer's +disease. NeuroImage: Clinical, 11, 46-51. https://doi.org/10.1016/j.nicl.2016.01.002 +Sen, B., & Parhi, K. K. (2021). Predicting Biological Gender and Intelligence From fMRI via +Dynamic Functional Connectivity. IEEE Transactions on Biomedical Engineering, +68(3), 815-825. https://doi.org/10.1109/TBME.2020.3011363 + +54 +Shappell, H., Caffo, B. S., Pekar, J. J., & Lindquist, M. A. (2019). Improved state change +estimation in dynamic functional connectivity using hidden semi-Markov models. +Neuroimage, 191, 243-257. https://doi.org/10.1016/j.neuroimage.2019.02.013 +Shawe-Taylor, J., & Cristianini, N. (2004). Kernel Methods for Pattern Analysis. Cambridge +University Press. https://doi.org/10.1017/CBO9780511809682 +Shine, J. M., Breakspear, M., Bell, P. T., Ehgoetz Martens, K. A., Shine, R., Koyejo, O., +Sporns, O., & Poldrack, R. A. (2019). Human cognition involves the dynamic +integration of neural activity and neuromodulatory systems. Nat Neurosci, 22(2), 289- +296. https://doi.org/10.1038/s41593-018-0312-0 +Smith, S. M., Beckmann, C. F., Andersson, J., Auerbach, E. J., Bijsterbosch, J., Douaud, G., +Duff, E., Feinberg, D. A., Griffanti, L., Harms, M. P., Kelly, M., Laumann, T., Miller, K. +L., Moeller, S., Petersen, S., Power, J., Salimi-Khorshidi, G., Snyder, A. Z., Vu, A. T., +. . . Glasser, M. F. (2013). Resting-state fMRI in the Human Connectome Project. +Neuroimage, 80, 144-168. https://doi.org/10.1016/j.neuroimage.2013.05.039 +Smith, S. M., Vidaurre, D., Beckmann, C. F., Glasser, M. F., Jenkinson, M., Miller, K. L., +Nichols, T. E., Robinson, E. C., Salimi-Khorshidi, G., Woolrich, M. W., Barch, D. M., +Uğurbil, K., & Van Essen, D. C. (2013). Functional connectomics from resting-state +fMRI. Trends in Cognitive Sciences, 17(12), 666-682. +https://doi.org/10.1016/j.tics.2013.09.016 +Vaghari, D., Kabir, E., & Henson, R. N. (2022). Late combination shows that MEG adds to +MRI in classifying MCI versus controls. Neuroimage, 252, 119054. +https://doi.org/10.1016/j.neuroimage.2022.119054 +Vergara, V. M., Mayer, A. R., Damaraju, E., & Calhoun, V. D. (2017). The effect of +preprocessing in dynamic functional network connectivity used to classify mild +traumatic brain injury. Brain and Behavior, 7(10), e00809. +https://doi.org/10.1002/brb3.809 +Vergara, V. M., Salman, M., Abrol, A., Espinoza, F. A., & Calhoun, V. D. (2020). Determining +the number of states in dynamic functional connectivity using cluster validity indexes. +Journal of Neuroscience Methods, 337, 108651. +https://doi.org/10.1016/j.jneumeth.2020.108651 +Vidaurre, D. (2021). A new model for simultaneous dimensionality reduction and time- +varying functional connectivity estimation. PLOS Computational Biology, 17(4), +e1008580. https://doi.org/10.1371/journal.pcbi.1008580 +Vidaurre, D., Abeysuriya, R., Becker, R., Quinn, A. J., Alfaro-Almagro, F., Smith, S. M., & +Woolrich, M. W. (2018). Discovering dynamic brain networks from big data in rest +and task. Neuroimage, 180, 646-656. +https://doi.org/10.1016/j.neuroimage.2017.06.077 +Vidaurre, D., Llera, A., Smith, S. M., & Woolrich, M. W. (2021). Behavioural relevance of +spontaneous, transient brain network interactions in fMRI. Neuroimage, 229, 117713. +https://doi.org/10.1016/j.neuroimage.2020.117713 +Vidaurre, D., Smith, S. M., & Woolrich, M. W. (2017). Brain network dynamics are +hierarchically organized in time. Proc Natl Acad Sci U S A, 114(48), 12827-12832. +https://doi.org/10.1073/pnas.1705120114 +Vidaurre, D., Woolrich, M. W., Winkler, A. M., Karapanagiotidis, T., Smallwood, J., & Nichols, +T. E. (2019). Stable between-subject statistical inference from unstable within-subject +functional connectivity estimates. Human Brain Mapping, 40(4), 1234-1243. +https://doi.org/10.1002/hbm.24442 + +55 +Wolfers, T., Buitelaar, J. K., Beckmann, C. F., Franke, B., & Marquand, A. F. (2015). From +estimating activation locality to predicting disorder: A review of pattern recognition for +neuroimaging-based psychiatric diagnostics. Neuroscience & Biobehavioral Reviews, +57, 328-349. https://doi.org/10.1016/j.neubiorev.2015.08.001 +Yang, Z., Craddock, R. C., Margulies, D. S., Yan, C.-G., & Milham, M. P. (2014). Common +intrinsic connectivity states among posteromedial cortex subdivisions: Insights from +analysis of temporal dynamics. Neuroimage, 93, 124-137. +https://doi.org/10.1016/j.neuroimage.2014.02.014 +Zalesky, A., Fornito, A., Cocchi, L., Gollo, L. L., & Breakspear, M. (2014). Time-resolved +resting-state brain networks. Proceedings of the National Academy of Sciences, +111(28), 10341-10346. https://doi.org/10.1073/pnas.1400181111 +Zamani Esfahlani, F., Jo, Y., Faskowitz, J., Byrge, L., Kennedy, D. P., Sporns, O., & Betzel, +R. F. (2020). High-amplitude cofluctuations in cortical activity drive functional +connectivity. Proceedings of the National Academy of Sciences, 117(45), 28393- +28401. https://doi.org/10.1073/pnas.2005531117 +Zhang, C., Baum, S. A., Adduru, V. R., Biswal, B. B., & Michael, A. M. (2018). Test-retest +reliability of dynamic functional connectivity in resting state fMRI. Neuroimage, 183, +907-918. https://doi.org/10.1016/j.neuroimage.2018.08.021 +Zhang, Y., Zhang, H., Chen, X., Lee, S. W., & Shen, D. (2017). Hybrid High-order Functional +Connectivity Networks Using Resting-state Functional MRI for Mild Cognitive +Impairment Diagnosis [Article]. Scientific Reports, 7(1), Article 6530. +https://doi.org/10.1038/s41598-017-06509-0 +Zhou, Z., Cai, B., Zhang, G., Zhang, A., Calhoun, V. D., & Wang, Y.-P. (2020). Prediction +and classification of sleep quality based on phase synchronization related whole- +brain dynamic connectivity using resting state fMRI. Neuroimage, 221, 117190. +https://doi.org/10.1016/j.neuroimage.2020.117190 + + diff --git a/YtE1T4oBgHgl3EQfwQVD/content/tmp_files/load_file.txt b/YtE1T4oBgHgl3EQfwQVD/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..63a1924a019da8caa54337ac044d5e545b2e6533 --- /dev/null +++ b/YtE1T4oBgHgl3EQfwQVD/content/tmp_files/load_file.txt @@ -0,0 +1,2516 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf,len=2515 +page_content='1 Dynamic Functional Connectivity Christine Ahrends1 and Diego Vidaurre1,2 1 Center of Functionally Integrative Neuroscience, Department of Clinical Medicine, Aarhus University, Denmark 2 Department of Psychiatry, University of Oxford, United Kingdom Abstract Most generally, dynamic functional connectivity (FC) refers to the non-instantaneous couplings across timeseries from a set of brain areas, here as measured by fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is in contrast to static FC, which is defined as purely instantaneous relations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this chapter, we provide a hands-on description of a non-exhaustive selection of different methods used to estimate dynamic FC (such as sliding windows, clustering approaches, Hidden Markov Models, and multivariate autoregressive models), and we explain, using practical examples, how data should be prepared for dynamic FC analyses and how models of dynamic FC can be evaluated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We also discuss current developments in the dynamic FC research field, including challenges of reliability and reproducibility, and perspectives of using dynamic FC for prediction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Keywords Functional connectivity, dynamics, sliding windows, clustering, Hidden Markov Model 1 Background & motivation The most traditional use of fMRI in research has been to detect changes in the signal according to some task design, for example in response to a visual stimulus (Filippi, 2016;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Frahm et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 1992;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Friston et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2006;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Kwong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 1992;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Ogawa et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 1992).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is a first-order statistic of the signal —i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' the average value of the signal for a given region.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Since the scale of the BOLD signal is not necessarily interpretable, it makes sense only 2 relative to a baseline.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, is the signal in a given region higher on average after a stimulus presentation than before?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In the last decade, however, there has been an explosion of studies aiming to characterise individual differences in fMRI activity, not only in task but also in rest —that is, in the absence of an explicit, constrained task (Lee et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2013;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Lurie et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Smith, Vidaurre, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In rest, first order statistics are less useful because there is no baseline to compare to, and there are no observable conditions to build a contrast.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is one of the reasons why we look at second-order statistics, such as functional connectivity (FC), which can be interpreted above and beyond the (arbitrary) scale of the signal (Power et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Another reason is that we expect FC to capture aspects of neural function and structure that are not directly linked to, or visible in, first-order statistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' One example is neural communication, which however might be reflected in FC in ways that are not necessarily direct or linear.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' FC can also reveal aspects of topographical organisation at large brain scales that cannot be retrieved from first-order statistics (Margulies et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' FC is defined as a statistical relationship between two brain areas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The most common measures of FC are linear, namely Pearson’s correlation or unnormalised covariance, or the inverse of these, which serves to partialise the information;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' that is, in the inverse of a covariance matrix (also called a precision matrix), the coefficient relating area i and area j defines their (linear) relation after removing the influence of all the other areas (Joseph F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Hair, 2009;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Joseph F Hair, 2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' There are other measures of FC that aim at capturing nonlinear information, most remarkably based in information-theory, such as mutual information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Although these measures are more general and can potentially capture other aspects of connectivity that escape a covariance or correlation matrix, they either require more data to obtain a clean estimation or impose their own assumptions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this chapter, we will focus our discussion on linear FC, since it is by far the most common choice, not just in general but particularly when the interest is in characterising the dynamic aspects of FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3 Another useful distinction is between FC and effective connectivity (EC) (Friston, 2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As opposed to FC, which is a mere statistical description of the data, EC is defined as the direct, causal influence that an area has over another in the context of the network to which they belong, and under specific models of biophysical constraints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Whereas EC can capture aspects of the data that FC cannot, the estimation of EC is also more dependent on specific assumptions and the inference of the parameters is more difficult.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This chapter only focuses on FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' But what is dynamic FC precisely?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Most broadly, we can define it as any second-order, cross-region information that is not captured by the so-called static FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Static FC is a usual term to refer to the correlation between voxels or regions as computed within the entire duration of the scanning session.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Note that, since the standard practice is to standardise the signal per scanning session to have a mean zero and standard deviation one, the covariance and correlation matrices are mathematically equal when we speak about static FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Critically, static FC, as per this definition, only captures instantaneous relationships;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' that is, within a scanning session, if we permute the time points within a session such that the permutation is the same for all voxels or areas, the static FC estimate would not change.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' But there is other information that would be lost by permuting;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' this information exists above and beyond the static FC, and that is what we refer to as dynamic FC here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' One of the problems in the study of dynamic FC is that the literature uses the same term for different types of measures, which has considerably muddled the discussion about the topic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' By basing our discussion on a broad definition of dynamic FC, we intend to clarify the different aspects of dynamic FC that the literature has covered.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Under the umbrella of this definition, there are two different aspects of dynamic FC: time-varying instantaneous FC, and FC in the context of a linear dynamical system that models non-instantaneous aspects of FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For now, we will discuss them conceptually.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Later, when we cover the different approaches to model dynamic FC, we will specify which type of information each approach is aiming to model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 4 Time-varying instantaneous FC, commonly referred to as just time-varying FC, reflects within-session modulations of the covariance matrix across regions, which captures between-area linear couplings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We note that, after standardising the signal, the mean is zero and the variance is one across the entire session, but this is not necessarily the case for shorter periods of the signal, where the signal, for example, could transiently have a higher variance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This means that covariance and correlation are not exactly equivalent anymore, opening different possibilities in how we model and interpret the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These practical aspects and what they mean conceptually will also be discussed below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Apart from instantaneous FC, the signal has other temporal information at different time scales, including the effect imposed by the hemodynamic response function as well as other factors of neural and non-neural origins.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These non-instantaneous couplings between areas, which are also dynamic FC, can be modelled as a linear dynamical system, like the autoregressive model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In short, an autoregressive model models the multivariate signal at time point ������������ (namely ������������������������) as a linear function of previous time points plus (Gaussian) noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This linear function is embodied by a set of autoregressive coefficients A, which, however linear, contain rich information about the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We will cover autoregressive models in some detail in the Practical Approaches to dynamic FC section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For now, it suffices to say that A, with a single set of parameters, can capture non-instantaneous phenomena such as travelling waves, chaotic behaviour, and oscillations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Different approaches of analysis can capture one or the other aspect, or both.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' But importantly, they are not independent from each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, if the signal contains both types of information, but we use an analysis approach that is focussed on only one, information of the other aspect will necessarily leak into the estimation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For this reason, they are not trivial to separate, and arguments about which one is a better-grounded description of the data are not easy to make.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' More practically, different approaches to dynamic FC allow us to address different research questions in complementary ways;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' see Calhoun et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 5 (2014);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Hutchison et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2013);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Lurie et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Preti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017) for some examples of applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This chapter is mainly devoted to practical aspects of dynamic FC analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This includes: data preparation prior to the analysis, a non-exhaustive description of existing approaches to characterise dynamic FC (with an emphasis on implementation), and some discussion on how to validate these models after they have been estimated, followed by conclusions and perspectives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Where applicable, we will demonstrate examples with Matlab code or refer to existing software packages that can be used for the implementation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We chose Matlab because it may still be the most common language in neuroimaging, but a translation of the code examples to other languages like Python or Julia is straightforward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' All code examples can also be found at https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='com/ahrends/DynamicFC_examples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 2 Preparing data for dynamic FC analysis When preparing fMRI data for dynamic FC analyses, a good starting point is to follow preprocessing guidelines for resting state fMRI, such as the Human Connectome Project’s (HCP) resting state preprocessing pipeline (Glasser et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2013;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Smith, Beckmann, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Resting state preprocessing guidelines may be more suitable than task-specific preprocessing recommendations since traditional analyses of task data typically take advantage of averaging over trials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' There are a few considerations specific to dynamic FC approaches, which we will outline here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For further reference, the issue of preprocessing for dynamic FC analyses has also been tested and discussed in Lydon-Staley et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019) and Vergara et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 Temporal preprocessing Most importantly, dynamic approaches are more heavily affected by temporal noise than time-averaged types of analyses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' With temporal noise, we here refer to any type of artefact 6 that varies over time, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', head motion, other physiological artefacts such as cardiac or respiratory artefacts, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While these artefacts may almost disappear when averaging over timepoints, such as in time-averaged or trial-averaged approaches, they can drastically influence the estimation of dynamic FC (Nalci et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For instance, if the dominant temporal fluctuations in an fMRI timeseries are due to head motion, a dynamic FC model may use its explanatory power to describe these movement-related variations rather than the more subtle signal fluctuations stemming from neural activity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In a state-based model, this may result in one or more states being actually motion states and not “brain” states, while in a continuously-varying FC estimation, each FC estimate may to some extent be biased by motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' On the other hand, dynamic FC analyses also suffer from too aggressive clean-up in the time domain, as some of the meaningful temporal variability can be removed along with the temporal artefacts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This may be the case, for instance, by applying preprocessing approaches that average over or censor time points, such as motion scrubbing (Power et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2012), or when regressing out temporal noise components using a full variance clean-up approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Also global signal regression can affect temporal variability, both positively and negatively, and so its use in preprocessing data for dynamic FC analysis is controversial (Murphy & Fox, 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The goal in terms of temporal variability when preprocessing fMRI data for dynamic FC analyses should therefore be to remove temporal artefacts while retaining non-artefactual temporal variability as much as possible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This may be achieved by non-aggressive temporal preprocessing strategies such as independent component analysis (ICA) in combination with unique variance clean-up of noise-related components (Griffanti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An additional consideration as regards temporal variability is temporal filtering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While a relatively lenient high-pass filter can be useful in removing ultra-slow fluctuations, such as scanner drifts, a very narrow filter will restrict the timescale on which dynamic changes in FC can be detected by a model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In general, since temporal noise and the meaningful aspect of the signal are not 7 perfectly separable, achieving the right balance can be a difficult task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A sensible approach may be to start with a relatively lenient temporal clean-up and to test post-hoc whether the dynamic FC model was affected by known temporal artefacts, such as head motion, and decide whether more aggressive temporal clean-up is necessary (Ciric et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Parkes et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An additional factor when doing dynamic FC analyses, and in particular time-varying FC analyses, is the variability of time-averaged FC between subjects;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' how different subjects are in terms of their average between-region correlations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Whether this variability between subjects is artefactual (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', due to poor registration) or non-artefactual (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' meaningful individual differences), it may mask the more subtle temporal variations in FC, making dynamic FC difficult to detect in a group-level timeseries (Lehmann et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While non- artefactual between-subject variability may be of interest in the analyses, preprocessing should strive to minimise artefactual differences between subjects when planning a group- level dynamic FC analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this case, it is recommended to test how representative FC patterns are of the group of subjects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A poor balance of low temporal variability and high between-subject variability can lead a dynamic FC model to converge to a static solution, which only describes differences between subjects (Ahrends et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The question of between-subject variability in FC is discussed in more detail in Bijsterbosch et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 Parcellations and timecourse extraction Another important consideration when preparing fMRI data for dynamic FC analyses is the choice of parcellation and method for timecourse extraction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' One may argue that, at least to some extent, the choice of parcellation depends on basic beliefs about brain organisation that have little to do with the analysis at hand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, the way in which we divide the brain into parcels and how we extract timecourses in practice determines important aspects of the data, such as the amount of temporal and between-subject variability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Consequently, the choice of parcellation and method of timecourse extraction also greatly affect the estimation 8 of dynamic FC (Ahrends et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2022;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Iraji et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Pervaiz et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For simplicity, we will here only demonstrate the effects of two types of parcellations on temporal variability and dynamic FC: a priori (structural or functional) binary parcellations and data-driven functional weighted parcellations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In a binary parcellation, the region of interest is outlined with a discrete boundary and each voxel can either belong (1) or not belong (0) to the parcel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In a data-driven functional weighted parcellation, each voxel has a specific (continuous) weight associated with each parcel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These weights are estimated from the data based on functional activity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' It should also be noted that the interpretation of FC itself changes depending on how parcels are defined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For instance, if parcels are binary (meaning that each voxel can either belong to a given parcel or not) and non-overlapping (meaning that each voxel can only belong to one parcel), FC can be interpreted as connectivity between a pair of distinct regions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If, on the other hand, parcels are weighted (meaning that each voxel has a certain value in each parcel) and overlapping (meaning that each voxel belongs to some degree to several or even all parcels), FC may be interpreted as connectivity between distributed pairs of networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' There are also several methods for extracting timecourses from voxels in these parcellations, some of which we will demonstrate below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In a binary parcellation, parcel timecourses can be extracted as the mean over voxels or the first principal component (PC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In a weighted parcellation, timecourses can be extracted using a dual regression approach to obtain individual spatial maps and individual timecourses of parcels (Beckmann et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' To understand the effects of the parcellation and timecourse extraction method on temporal variability and dynamic FC estimation, we will simulate a few different scenarios.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Let’s first consider the optimal case, in which, within each parcel, voxel activity is relatively well homogenous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is an implicit assumption of binary parcellations, but it is not usually tested explicitly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This case is illustrated in Figure 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We simulate two functional clusters, A and B, each with a consistent timecourse whose peak is located at the centre of the cluster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 9 In this scenario, both clusters’ timecourses are simply sine-waves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The only difference between the clusters is the overall signal amplitude.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A = zeros(10,10);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for i = 1:5 A(i, i:(end-(i-1))) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2*i;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A(i:(end-(i-1)), i) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2*i;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A(end-(i-1), i+1:end-(i-1)) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2*i;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A(i+1:end-(i-1), end-(i-1)) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2*i;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end B = A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T = 1:100;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yA = A(:)*sin(T);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB = B(:)*sin(T)*0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='7;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, yA are the timecourses for the voxels in cluster A and yB are the timecourses in cluster B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The voxel weights for cluster A and cluster B as well as the empirical timecourses yA and yB are shown in Figure 1, top row.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 10 Figure 1 Parcellation and timecourse extraction for scenario 1, where voxel clusters in parcel have homogeneous timecourses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We will consider the case where both clusters are contained within the same parcel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In binary, non-overlapping parcellations, timecourses are often extracted by simply computing the mean over all voxels belonging to each parcel at each timepoint: for t = T parcel_mean(t) = mean([yA(:,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(:,t)]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end The resulting timecourses are shown in Figure 1, second row.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this (optimal) case, the average activity across voxels captures the overall pattern in the region adequately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, while this method is fast and easy to compute, it can have substantial shortcomings depending on the actual temporal variance of voxel clusters in the parcel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Empirical voxel timecourses Empirical voxel weights Cluster A 5 Cluster A 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 10 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Cluster B Cluster B 15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 20 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0 10 20 30 40 50 60 70 80 90 100 2 6 10 Binary parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mean extracted Voxel weights binary parcel 5 Parcel 1 10 (binary) 0 15 20 0 10 20 30 40 50 60 70 80 90 100 2 6 10 Binary parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 1st PC extracted Voxel weights binary parcel 5 Parcel 1 10 (binary) 0 15 20 0 10 20 30 40 50 60 70 80 90 100 2 6 10 Weighted parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' regression coefficients extracted Voxel weights weighted parcel 5 Parcel 1 10 (weighted) 15 20 0 10 20 30 40 50 60 70 80 90 100 2 610 Time11 Consider the case where the two functional clusters’ activity contained within the parcel is heterogeneous,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', they are negatively correlated (see Figure 2, top row): yA = A(:)*sin(T);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB = B(:)*-sin(T);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If we here extract the parcel timecourse as mean over all voxels, we will completely flatten the timecourse so that the original temporal variance will be lost (see Figure 2, second row).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An alternative to extracting the parcel timecourse as the mean over all voxels is to use the first principal component (PC) of the voxel timecourses within the given parcel: parcel_pc_tmp = pca([yA;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_pc = parcel_pc_tmp(:,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The resulting timecourses are shown in Figure 1 and Figure 2, third rows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Both in the case of homogeneous activity within the parcel (Figure 1) and in the case of heterogeneous activity within the parcel (Figure 2), the first PC captures the dominant pattern of voxel activity over time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this way, we can avoid the issue of the parcel timecourse flattening out and instead preserve the most dominant temporal variance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, using this approach, some information about the temporal variance within the region of interest would still be lost, namely that of cluster B in the example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 12 Figure 2 Parcellations and timecourse extraction, scenario 2, where voxel clusters in parcel have heterogeneous timecourses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Using a weighted parcellation gives a more nuanced view of the activity within a parcel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can create a weighted parcel from the cluster timecourses by decomposing the data, for instance, using independent component analysis (ICA): ica_mdl = rica([yA;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='yB],1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_ic = ica_mdl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='TransformWeights;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, parcel_ic is the timecourse of a single IC extracted from the voxel-level timecourses of clusters A and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can access each voxel’s individual weight within the parcel: Empirical voxel timecourses Empirical voxel weights Cluster A 5 Cluster A 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 10 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Cluster B Cluster B 15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 20 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0 10 20 30 40 50 60 70 80 90 100 2 6 10 Binary parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mean extracted Voxel weights binary parcel 5 Parcel 1 10 (binary) 0 15 20 0 10 20 30 40 50 60 70 80 90 100 2 Binary parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 1st PC extracted Voxel weights binary parcel 5 Parcel 1 10 (binary) 0 15 20 0 10 20 30 40 50 60 70 80 90 100 2 6 10 Weighted parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' regression coefficients extracted Voxel weights weighted parce 5 Parcel 1 10 (weighted) 15 20 0 10 20 30 40 50 60 70 80 90 100 2 610 Time13 z = transform(ica_mdl,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='[yA;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='yB]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights_ICA_A = reshape(z(1:100),[10,10]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights_ICA_B = reshape(z(101:end),[10,10]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In scenario 1 (Figure 1, fourth row), the IC parcel timecourse corresponds closely to the timecourse of cluster A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The voxel weights capture the original spatial distribution of the clusters and are higher for the voxels belonging to cluster A than the voxels belonging to cluster B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this way, the parcel timecourse in combination with the voxel weights capture not only the most dominant temporal pattern, but also the small amplitude difference between clusters A and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In scenario 2 (Figure 2, fourth row), the IC parcel timecourse again corresponds closely to the timecourse of cluster A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, while the estimated voxel weights for the voxels from cluster A are positive, the voxels from cluster B have negative weights assigned, which captures how cluster B’s voxels’ activity is negatively correlated with the activity of cluster A voxels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This method thus retains almost all spatiotemporal details of the original data, so that by multiplying the voxel weights with the parcel timecourses, we could almost perfectly recreate the empirical voxel timecourses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An additional advantage of estimating the parcellation based on the functional activity from the dataset at hand is that we can avoid the issue of suboptimal parcel boundaries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is especially important in the context of dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Consider the case of two functional clusters, which are positively correlated at the beginning of a timeseries, then fall out of synchrony in the middle of the timeseries, and are correlated again at the end of the timeseries: yA = A(:)*sin(T);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(:,1:30) = B(:)*sin(T(1:30));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(:,31:70) = B(:)*sin(31:0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5:50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(:,71:100) = B(:)*sin(T(71:100));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can calculate the average correlation between the two clusters, using e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', a sliding window approach (see section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 for details).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is shown in Figure 3, top row.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Binary parcellations that are defined a priori on a different dataset can have the problem of suboptimal parcel boundaries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this case, rather than dividing the data into the “true” 14 clusters A and B, an a priori defined parcellation may split both clusters in the middle so that half of cluster A and half of cluster B belong to parcel 1 and the other half of each cluster belongs to parcel 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this case, whether timecourses are extracted using the mean or the first PC, the two clusters’ activity will be contained equally in both parcels, making it impossible to disambiguate the original functional clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = T parcel_mean(1,t) = mean([yA(1:50,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(1:50,t)]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_mean(2,t) = mean([yA(51:100,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(51:100,t)]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end parcel_pc_tmp_1 = pca([yA(1:50,:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(1:50,:)]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_pc(:,1) = parcel_pc_tmp_1(:,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_pc_tmp_2 = pca([yA(51:100,:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' yB(51:100,:)]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_pc(:,2) = parcel_pc_tmp_2(:,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is shown in Figure 3, second and third row for the mean timecourse extraction and first PC timecourse extraction methods, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If we estimate dynamic FC between the two parcels using these timecourses, FC would be high throughout the timecourse and not capture the change in FC in the middle of the timeseries, since we are essentially computing dynamic FC of both functional clusters with themselves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The data-driven functional parcellation (ICA) is able to capture the true parcel boundaries more accurately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We estimate ICA with two components: ica_mdl = rica([yA;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='yB],2);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' parcel_ic = ica_mdl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='TransformWeights;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We then have a separate set of voxel weights for each component (parcel): z = transform(ica_mdl,[yA;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='yB]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights_ICA_A_1 = reshape(z(1:100,1),[10,10]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights_ICA_B_1 = reshape(z(101:end,1),[10,10]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights_ICA_A_2 = reshape(z(1:100,2), [10,10]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights_ICA_B_2 = reshape(z(101:end,2),[10,10]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These components capture the separate timecourses for clusters A and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Estimating dynamic FC between these two components more closely resembles the empirical dynamic FC between the functional clusters (see Figure 3, fourth row).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 15 Figure 3 Dynamic FC between two cluster timecourses (empirical: row 1) extracted from binary parcels with suboptimal parcel boundaries (row 2 and 3) vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' extracted from weighted, data-driven parcels (row 4) The examples showed how a suboptimal parcellation or timecourse extraction method can lead to loss of temporal information and wrong estimation of the temporal relationships between parcels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' When preparing data for dynamic FC analysis, it is therefore crucial to choose a suitable parcellation and timecourse extraction method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Extracting timecourses as the first PC rather than the mean can preserve temporal variability in a binary parcellation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For dynamic FC analyses, several studies found that data-driven functional parcellations, such as group ICA approaches, are preferred over binary parcellations (Ahrends et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2022;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Iraji et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Pervaiz et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020), as data-driven parcellations not only preserve temporal variability, but also avoid the issue of suboptimal parcel boundaries that can occur Empiricalvoxel timecourses Empirical voxel weights Empirical dFC A Cluster 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 Cluster A 10 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 B Cluster B luster 15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 -1 20 0 20 40 60 80 100 0 2 20 40 60 80 100 Binary parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mean extracted Voxel weights binary parcels dFc binary parcels,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mean 5 Parcel 1 1 10 0 ■0 Parcel 2 15 20 1F 0 20 40 60 80 100 2610 2610 0 20 40 60 80 100 Parcel 1 Parcel 2 Binary parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 1st Pc extracted Voxel weights binary parcels dFC binary parcels,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 1st PC 5 Parcel 1 1 10 0 0 Parcel 2 15 20 1F 0 20 40 60 80 100 2610 2610 0 20 40 60 80 100 Parcel 1 Parcel 2 Weighted parcellation,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' regression coefficient extracted Voxel weights weighted parcels dFC weighted parcels,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' regression coefficient 5 Parcel 1 10 0 15 Parcel 2 6 20 0 20 40 60 80 100 2610 2 610 0 20 40 60 80 100 Time Parcel 1 Parcel 2 Time16 with a priori binary parcellations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' It should be mentioned that alternative approaches exist that compute time-varying FC on the voxel-level or define a parcellation based on time- varying FC (Preti & Van De Ville, 2017), but they are less common.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Beyond the type of parcellation, its coarseness needs to be taken into account.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While it may be tempting to choose a parcellation to be as fine-grained as possible to maximise spatial resolution, a too fine-grained parcellation increases the number of parameters in a dynamic FC model, which can lead to noisy or even ill-posed estimations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Broadly speaking, the number of parameters to estimate in the model should be in balance with the number of observations, which are often limited in common fMRI datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An alternative to using a coarser parcellation is to use a dimensionality reduction technique, such as Principal Component Analysis (PCA), on the extracted timecourses of a fine-grained parcellation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, working with transformed data in PCA-space may also introduce a bias for certain dynamic FC models (Vidaurre, 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3 Practical Approaches to dynamic FC There is a plethora of approaches to evaluate dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, we focus on methods that fit our definition of dynamic FC, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', that are based on second-order statistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This excludes, for instance, popular methods based on co-activation maps (CAPs) (X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018), which are based on first-order information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Also, we prioritise a practical understanding of the methods at the expense of breadth, so only a subset of representative methods will be discussed and analysed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For each approach there often are several software packages available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We present methods and software to estimate single-subject, single-session dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, all the principles are applicable to compute the estimates over a range of sessions and subjects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This would be done by concatenating all the time series and taking into account when a session finishes and the next starts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We will assume that the data for a 17 given scanning session have been formatted into a matrix X, with as many rows as fMRI volumes (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', time points) and as many columns as voxels, regions, or components.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 Continuously-varying estimators We refer as continuously-varying estimators to methods that produce an estimate of FC per time point or window, such that the parameters that define such estimations vary smoothly in the time axis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Because the dynamics are encoded by changes in these parameters, these approaches are time-varying estimates according to the above classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can separate two kinds of continuously-varying estimators: those based on sliding windows, which divide the data set in contiguous pieces and perform the estimation separately per piece;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' and all-data estimators, which use the data set all at once.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Sliding windows are simpler and much more common.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 Sliding windows The most basic approach to evaluate dynamic FC is sliding windows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, given a certain (typically predefined) choice of window length in terms of number of frames, we perform the estimation within this window and then slide the window by one or more time points and repeat the estimation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is done across the entire time series.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, given p number of brain regions, a basic estimation would be window_length = 100;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C = zeros(p,p,total_session_duration - window_length + 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = 1:total_session_duration - window_length + 1 C(:,:,t) = corr(X(t:t+window_length-1,:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end In this code, we slide the window by one time point;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for computational efficiency, the window can be slid by more than one time point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A family of variations of this scheme is related to the shape of the window: here a squared window was used, but there are alternatives that can improve the estimation in practice such as a tapered or an exponentially decaying window.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In contrast to a static FC estimation that uses the entire timeseries (Figure 4A), 18 Figure 4B illustrates how FC is estimated by sliding windows (Figure 4C refers to a state- based estimation, which will be described later).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Figure 4 Static and time-varying FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We refer to second-order statistics of fMRI timeseries, such as correlations between regions of interest (ROIs), as functional connectivity (FC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A) Time-averaged (static) FC can be computed, for instance, by correlating all pairs of ROIs over the entire timeseries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The resulting FC can be illustrated as (ROI x ROI) FC matrix or as FC map in the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B) Continuously varying dynamic FC, such as sliding window approaches, estimate FC separately on portions (windows) of the timeseries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Each window has an associated FC matrix that can also be projected into the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C) State-based FC models estimate recurring patterns of FC over the timeseries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Each state has an associated FC matrix that can also be projected into the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (Abbreviations: ROI - region of interest;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' FC - functional connectivity;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' TR - repetition time).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A Time-averaged (static) fMRI timeseries FC matrix FC brain map ROI 1 ROI 1 w ROI2 RO12 ww wiy Mmw Mr ROI n ROI1 ROIn ROIn ROI2 Time (TR) ROI 1 ROI n ROI x ROI correlation B Time-varying FC: continuously-varying estimation FC matrix ROI 1 ROI2 ROI n ROlxROI correlation ROI 1 ROI2 Myny W ROIn W Time (TR) C Time-varying FC: state-based estimation State 1 State 2 State 3 State 4 State k ROI 1 ROI2 ROIn FC matrix19 Here,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' we used correlations,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' but unnormalised covariance is also possible by using cov() instead of corr();' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' in this case, we will also be including information about the variance of the signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Now, if we wish to compute partial correlation (or covariance), where the linear influence of all the other regions is removed from the coefficient between region i and j, we can consider the following code: for t = 1:total_session_duration - window_length + 1 inv_mat = inv(C(:,:,t) + lambda * eye(p));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' inv_mat = - (inv_mat .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='/ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' repmat(sqrt(abs(diag(inv_mat))),1,p)) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='/ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" repmat(sqrt(abs(diag(inv_mat)))',p,1);" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' inv_mat(inv_mat(p)>0)=0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' iC(:,:,t) = inv_mat;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end Inverse covariance matrices (here, inv_mat) are also referred to as precision matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If the number of regions is comparatively large, we need to regularise the matrix inversion to avoid badly scaled results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is what we did here by adding the identity matrix multiplied by the regularisation constant lambda.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Other types of regularisation that impose sparsity, by driving some partial correlations to exactly zero, are also possible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this case, the resulting precision matrices can be mapped to a graph where non-zero coefficients relate to edges between two nodes (voxels, areas, or components), and zero coefficients relate to conditional independence between node i and j, given all the rest of the nodes (Cai et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Friedman et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2008).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Sliding-window analyses can be very dependent on the choice of the window length, in the sense that too short windows will render very unstable estimates and too long windows will over-smooth the estimation and miss relatively fast changes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Although there are data-driven approaches to adaptively optimise the window length, the resulting window length is itself based on assumptions and subject to estimation noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A critical weakness of the sliding window approach is that, except perhaps for very long windows, a large part of the variability observed across windows will inevitably be due to 20 estimation noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is because the number of time points within a window is not large enough to yield a stable measure;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' this gets exacerbated by the autocorrelations in the signal, which further reduce the effective degrees of freedom within the window (Afyouni et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This can be easily verified empirically with the following code total_session_duration = 2000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' p = 10;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X = mvnrnd(zeros(total_session_duration,p),true_C);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j = 1:p X(:,j) = smooth(X(:,j),10);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end X = X(101:end-100,:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, true_C was a static covariance matrix obtained from real fMRI data, which we used to sample Gaussian noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We then applied some smoothing to generate some autocorrelation in the data (imitating what we would observe in real fMRI data at a very basic level).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If we then apply a sliding window analysis, we will observe broadly varying estimations of time- varying FC with either correlation or partial correlation, even though the underlying covariance matrix is not time-varying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is illustrated in the left panel of Figure 5A, where we show, for a pair of signals, the results of performing a sliding window analysis on data generated as per the above code;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' the dotted and solid grey lines correspond, respectively, to the ground-truth correlation (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' true_C(1,2)) and empirical static correlation (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' corr(X(:,1),X(:,2))) —note the large difference between the two due to smoothing, even when using the entire length of the signals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For further discussion and examples on model validation, see Section 4 below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For comparison, we generated a second data set where there are two different covariance matrices underlying the generation of the data;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' the middle section of the time series was generated using one, and the beginning and end of the time series was generated using the other: 21 total_session_duration = 2000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' p = 10;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X1a = mvnrnd(zeros(total_session_duration/4,p),true_C1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X1b = mvnrnd(zeros(total_session_duration/4,p),true_C1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X2 = mvnrnd(zeros(total_session_duration/2,p),true_C2);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X = [X1a;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X2;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X1b];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j = 1:p X(:,j) = smooth(X(:,j),10);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end X = X(101:end-100,:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Figure 5B shows that, while the real modulation in FC is apparent, the fluctuations around it are also quite large.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Figure 5 Result of performing a sliding-window (AB) or all-data, flexible time squares (CD) analysis on synthetic data whose ground truth generative process either does not contain a time-varying component (AC), or does contain it (CD).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This problem has put dynamic FC on the spot of criticism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Different approaches to generate surrogate data and statistical tests have been proposed (see below), which tend to agree that sliding-window approaches struggle to find genuine time-varying FC in most data, as the variability of interest may drown in estimation noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A Amount of regularisation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 Correlation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Correlation/Partial Correlation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 Volumes Volumes Empiricalcorrelation Staticcorrelation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 Amount of regularisation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Correlation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Volumes Volumes B D22 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 All-data continuous estimators One alternative to sliding windows within the paradigm of continuous estimators are methods that estimate one correlation coefficient per time point by considering all the data set at once.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These methods do not use a predefined window, but the estimate at a given time point still depends on neighbouring time points in an adaptive fashion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is achieved by imposing a regularisation constraint consisting of penalising the coefficient differences between contiguous time points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' One such method is flexible least squares (Kalaba & Tesfatsion, 1989), which is implemented for example in the Dynamic BC toolbox (Liao et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multivariate approaches that estimate continuously-varying partial correlation matrices also exist based on the same idea (Monti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Focussing on just two signals, Figure 5C shows the behaviour of the flexible least squares method on the synthetic data generated above, where there was no actual fluctuation of FC in the underlying generative model of the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As observed, choices with a low amount of regularisation (redder lines) result also in largely varying estimates (even above the 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='0 boundary), while higher amounts of regularisation converge to the empirical, non-varying correlation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' When there is actual variability in FC, Figure 5D shows that for some choices of the regularisation parameter, the method is able to capture the modulation in the middle section of the time series above and beyond the noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As we can see, while continuous approaches of this kind do not necessitate the specification of a window length parameter, they are still quite dependent on the choice of the regularisation parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, the fact that, unlike sliding windows, these models have a well-defined loss function lends to the use of quantitative metrics for selecting this parameter, for instance based on cross-validation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Further, as the above simulation shows, they can improve over the sliding-window estimator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' On the other hand, they are more complicated to implement and they are less readily available in standard software packages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 23 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 State-based estimators As opposed to the continuous estimators discussed above, state-based estimators assume that the data can be reasonably described using a discrete set of states (which can still be large).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This does not mean that by using these methods we assume any one-to-one, state- to-mechanism mapping, or that there is any specific biophysical significance in the number of states;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' this assumption is merely descriptive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Another important assumption of these methods is exclusivity: only one state can be active at a given time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A very important difference between continuous and state-based estimators is that, while for a continuous estimator the FC parameters are essentially all we have to estimate, for a state-based estimator we also have to estimate when the states occur (namely, the state time courses), together with each state’s FC information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Another critical difference is that, because the occurrences of these states can happen anytime in the time series, a state can (and will probably) be assigned to several non-contiguous segments of the signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Consequently, while continuous estimators (like sliding windows) do not pool information across subjects in any way, state-based estimators can have states that are shared across subjects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' State-based estimators are illustrated in Figure 4C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, each of the five states’ FC is estimated based on several occurrences of varying length (dwell times) throughout the timeseries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These differences have an important practical implication: that the estimation of a state’s FC is based on much more data than the estimation of a sliding window’s FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, let us suppose that we have 100 subjects with 20min of data per subject, and we run some state-based estimator with 20 states;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' each state would have on average 100 minutes of available data for the estimation of its FC parameters (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', a covariance or precision matrix), as opposed to a typical sliding window of 1-2min.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Now, if we scale up to larger data sets like the HCP or the UK Biobank, states will have many hours of data available, effectively making the estimation of FC extremely precise except for residual states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' But, as mentioned, 24 in these methods we also have to estimate the state time courses, which is of course also subject to estimation noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We will briefly discuss below how to characterise and deal with estimation noise at different levels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We divide the state-based estimators into two classes: clustering methods (Allen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014) and generative models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Within the generative models, we focus on the hidden Markov models (HMM) (Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Conceptually, the main difference is that the former has fewer assumptions and does not set up a model from which we can sample data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We will compare them empirically later on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' There are other state-based estimators, such as the Kalman filter, that do not belong to any of these categories, but they are somewhat less common in the field of neuroimaging and will not be discussed here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 Clustering approaches The most common clustering method builds upon sliding window estimates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Assuming we have a pool of FC matrices (one per window), the simplest procedure would be to vectorise the off-diagonal elements of the FC matrices, constructing a (no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' of windows by pairs of regions) matrix, on which we will then apply a clustering technique such as k-means (Allen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014) or hierarchical clustering (Yang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' By doing this, we would assign each window to a different state, which constitutes a categorical state time course.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Probabilistic alternatives to k-means, such as a mixture of distributions, are also possible, in which case the state time courses will be made of probabilities (that is, per window and state).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Together with the estimated state time courses, the state FC estimations would correspond to the centre of the clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The hope of this method is that, even though the individual sliding window estimates are quite noisy, the state FC matrices are made of averaging across many windows and therefore will be less noisy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A mathematically more principled version of this approach is based on the use of Riemannian distances between the window FC matrices (instead of Euclidean, as results from vectorising the FC matrices) (Pervaiz et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 25 For simplicity, we will here illustrate only the simplest (and most common) approach with code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Assuming we have already computed sliding-window estimates on data generated using the code above, the following code implements the most basic clustering approach: N_windows = total_session_duration - window_length + 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C_unwrapped = zeros(N_windows,p*(p-1)/2);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j = 1:N_windows Cj = C(:,:,j);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cj = Cj(triu(true(p),1));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C_unwrapped(j,:) = Cj(:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end K = 4;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' idx = kmeans(C_unwrapped,K);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' state_time_courses = zeros(length(T),K);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for k = 1:K state_time_courses(idx==k,k) = 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end We can now get the fractional occupancies (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' the proportion of time taken by each state), and display the state time courses as mean(state_time_courses) area(state_time_courses) Another clustering approach that does not necessitate the specification of a window length hyperparameter is Leading Eigenvector Dynamics Analysis (LEiDA) (Cabral et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In short, LEiDA computes functional connectivity estimates at each time point by first computing the signals’ instantaneous phase (by means of the Hilbert transform) and then calculating the cosine similarity between each pair of signals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Note that even though phase is instantaneous, it still needs information from neighbouring time points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This yields a relatively noisy estimation of functional connectivity per time point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' To reduce the amount of noise, the first eigenvector of each FC matrix is extracted using a singular value decomposition, producing a (no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' of time points by no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' of regions) matrix containing FC information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can then apply a k-means algorithm as usual.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For illustration, this procedure is implemented in the code below: 26 Phase = zeros(total_session_duration,p);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j=1:p Phase(:,j) = angle(hilbert(X(:,j)));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end Eigenvectors = zeros(total_session_duration,p);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = 1:total_session_duration C = zeros(p);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j1 = 1:p-1 for j2 = j1+1:p d = abs(Phase(t,j1) - Phase(t,j2));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' if d > pi, d = 2*pi-d;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end C(j1,j2) = cos(d);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C(j2,j1) = C(j1,j2);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end end [v,d] = eig(C);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' [~,i] = max(diag(d));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Eigenvectors(t,:) = v(:,i);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end [idx,eigen_centroids] = kmeans(Eigenvectors,K);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, eigen_centroids define the states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Each state essentially projects the brain areas into a one-dimensional axis, such that areas close to each other in one extreme of the axis (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', having the same sign in eigen_centroids) are in-phase, and areas that belong to opposite extremes of the axis (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', having opposite signs) are anti-phase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Note that the sign in eigen_centroids is arbitrary, and only the sign relationships between areas are meaningful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Overall, clustering approaches are generally simple to implement but they are not generative models, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', we do not have a compact set of parameters to sample data or perform statistical testing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The hidden Markov model (HMM), discussed in the next section, is a generative model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 Generative models: Hidden Markov Models (HMMs) Generative models are those that specify a full mathematical structure from which we can sample new data sets (Bishop, 2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As such, they define a probability distribution over the data, and have a well-defined set of parameters over which we can do statistical inference 27 and testing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, we discuss the HMM, a generative model specifically designed to deal with temporal (and other sequential) data (Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The HMM is, rather than a single model, a family of probability distributions where each state is in itself a probability distribution, defined by a set of state parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' By choosing one or another state distribution (also called observation models), the HMM can adapt to different kinds of data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, state distributions can be Gaussian, multinomial, Wishart, Poisson, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, we will focus on the Gaussian distribution because it is the common choice for fMRI data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A Gaussian distribution is parametrised by a mean vector and a covariance matrix containing FC information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This allows three main variants of the Gaussian HMM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' First, having a shared covariance matrix for all states and one mean vector per state;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' this configuration however does not model changes in FC and will not be discussed here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Second, we can have one mean vector and covariance matrix per state, which is the most common approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Third, in order to focus on FC, we can pin the state means to zero and have only a covariance matrix per state (which is equivalent to having a Wishart state distribution) (Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The remaining parameters of the HMM are the initial probabilities, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', the probability of the trials to commence with a given state;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' and the transition probability matrix, with elements (j,k) encoding the probability of transitioning from state j to state k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The initial probabilities, as well as each of the rows of the transition probability matrix, are modelled as Dirichlet distributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Apart from state exclusivity, and the fact that we use a discrete number of states, the use of a transition probability matrix implies that the HMM builds upon a third assumption: Markovianity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In the temporal domain, this means that which state is active in the present time point depends on which state was active in the previous time point;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' or in more rigorous terms, that the state at time point t is conditional independent to all the rest of the time points given t-1 and t+1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In practice, this typically leads to having smoother, and not too abrupt, 28 state transitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Note that even though the HMM does not model longer term dependencies, the model does not forbid them either, and the resulting state time courses might still exhibit long-term dependencies (Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Versions of the HMM that explicitly model the state visit durations, therefore dealing with longer term dependencies, are referred to as semi-Markovian (HsMM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These have also been applied to fMRI data, in practice yielding similar estimations to the standard HMM (Shappell et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Now, the inference of an HMM from data implies not only the estimation of the HMM parameters just described but also the state time courses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While the HMM model can be ported to different data sets, the state time courses are specific to the data set on which they were estimated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' To illustrate the use of the HMM and its inference, we will use the HMM- MAR1 toolbox in Matlab.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Assuming that we have added the toolbox to the Matlab path, we can estimate a model by options = struct();' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' options.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='K = 8;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' options.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content="covtype = 'full';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' [hmm,Gamma,~,vpath] = hmmmar(X,T,options);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' With this code, we are estimating an HMM model with K=8 states, and state time courses Gamma, on data X, which might be composed of different concatenated fMRI sessions or subjects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The length of each session is indicated in vector T (in number of time points).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If we wish to pin the mean of the state Gaussian distributions to zero such that the states are defined only as FC matrices, we would use options.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='zeromean = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 1 The name of the toolbox has a historical reason: the first observation model available was the multivariate regressive model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' But it now contains other distributions, such as Gaussian, Poisson, Wishart, probabilistic PCA, or the multivariate regression model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' All documentation is available at https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='com/OHBA-analysis/HMM-MAR/wiki 29 Using the same variables for T and options that we used for training, we can query the fractional occupancies (the proportion of activation per state and subject) and plot the state time courses as FO = getFractionalOccupancy(Gamma,T,options) for k = 1:8 state_time_courses(vpath==k,k) = 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end figure;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' imagesc(FO) figure;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' area(state_time_courses(1:T(1),:)) For interpretation purposes, it is important to note that, unlike the sliding window estimates (which are typically correlations matrices), here a FC matrix is a covariance matrix and therefore also conveys information about the variance of the signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As mentioned above, although the variance of an entire scanning session is one because of standardisation, the variance for sub-periods of the signal might not be exactly one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As a consequence, the HMM could be capturing changes in amplitude or variance in certain cases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A final remark is about the estimation of the number of states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This could be done in different ways, for example using the free energy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In real data this question is not very relevant because the ground-truth number of states does not exist;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' more states will simply result in a finer-grained estimation, and replicability is often a more sensible criterion (Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='3 Multivariate autoregressive models So far, we have been considering dynamic functional connectivity in terms of time-varying instantaneous FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this section, we will describe the multivariate autoregressive model (MAR), a linear dynamical system that models non-instantaneous aspects of FC with a single set of parameters —i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', without the use of states or windows (Harrison et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2003;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Rogers et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Mathematically, we can define the MAR model as the following generative model: 30 ������������������������ = � ������������������������−������������ ������������������������ + ������������������������ ������������ ������������=1 Where ������������������������ is Gaussian noise, ������������������������ are matrices of autoregressive parameters and ������������ is the order of the MAR model, which defines the complexity of the model and which we have to choose beforehand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Order selection can make use of a pre-specified criterion, such Akaike’s (Akaike, 1974), or use cross-validation;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' we will however not discuss this here since, in practice, a precise estimation of the optimal order is not that relevant in neuroimaging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The MAR model can be expanded with non-linear terms (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', area interactions), but we will not discuss this possibility here either.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can reexpress the above equation in a standard regression form as ������������������������ = ������������������������ ������������∗ + ������������������������ where ������������������������ is defined as [������������������������−������������ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' ������������������������] and ������������∗ is a (������������ ������������ by ������������) matrix that reunites all the ������������������������ matrices across the ������������ channels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This allows us to straightforwardly apply any standard regression estimation method to find the autoregressive parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, assuming that there is only one subject in X, we can compute the autoregressive coefficients using least squares: Y = zeros(size(X,1) - L , L * p);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j = 1:L Y(:,(1:p)+(j-1)*p) = X(L-j+1:end-j,:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end A_star = pinv(Y) * X(L+1:end,:);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Although very simple to estimate, the MAR model contains rich information about the signal on the frequency domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Power, coherence, partial directed coherence, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' can be readily computed by Fourier transforming ������������∗ (in the code, A_star), and the richness of these spectral estimates depends on the order of the model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This falls out of the scope of this chapter;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' but see Faes and Nollo (2011) for a comprehensive account.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 31 According to these equations, the diagonal elements of the ������������������������ matrices (and the corresponding elements in ������������∗) refer to “self-connections”, and the off-diagonal elements refer to cross-area connections, which we can interpret as dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Note that, similarly to the elements in a covariance matrix, the autoregressive coefficients are second-order statistics, but are not instantaneous.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Conveniently, the autoregressive coefficients share with the precision matrix the property that, for a very large amount of data, the coefficient(s) between variable i and variable j will be exactly zero if they do not hold a direct linear (in this case lagged) relationship, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', after we regress out the other variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This has opened the window for the MAR model to be used as a tool to infer effective connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Critically, this is under the assumption that there are no unobserved variables in the system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, we could get a non-zero coefficient between two variables if these were connected through an unobserved third variable but not directly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Of course, this assumption does not often hold in real neuroimaging data, but this property could still be useful because, at least, if a coefficient is zero then we can assert that the variables are not linearly connected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Let us show the most basic scenario with an example, where we assume the following linear dynamical system of order 1 with variables (brain areas) ������������(1),.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', ������������(5): ������������(1)������������ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 ������������(1)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(2)������������−1 + ������������(1)������������ ������������(2)������������ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 ������������(2)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(1)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(3)������������−1 + ������������(2)������������ ������������(3)������������ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 ������������(3)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(2)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(4)������������−1 + ������������(3)������������ ������������(4)������������ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 ������������(4)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(3)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(5)������������−1 + ������������(4)������������ ������������(5)������������ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 ������������(5)������������−1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25 ������������(4)������������−1 + ������������(5)������������ And let us assume we sample data from this system with the following code: 32 p = 5;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' total_session_duration = 1000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' true_A = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5*eye(p);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j = 1:p-1 true_A(j,j+1) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' true_A(j+1,j) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='25;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end X = randn(total_session_duration,p);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = 2:total_session_duration X(t,:) = X(t,:) + X(t-1,:) * true_A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end Now, if we sample many data sets like this and fit autoregressive models with the code above, we would obtain distributions of estimated parameters as in Figure 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' If instead of being able to sample many data sets, we had a limited number of subjects, we could still perform a bootstrapped estimation (where we create pseudo data sets by randomly sampling subjects with repetitions).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Figure 6 Estimation of autoregressive coefficients over 1000 data sets generated through an idealised linear dynamical system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Of course, this is an extremely idealised scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In practice, the ground-truth is not an autoregressive model, and autoregressive coefficients are rarely zero in real data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The concept of Granger causality emerged as a dedicated statistical test in the field of economics to deal with real data under a definition of causality based on temporal precedence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Region 1 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Region 2 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Region 3 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Region 4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Region 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Region1 Region2 Region3 Region4 Region533 However, the use of Granger causality in fMRI data for inferring effective connectivity has been criticised because of the differences in latency in the hemodynamic response function across brain regions, as well as other reasons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Still, multivariate autoregressive modelling is a useful FC statistic in fMRI (as opposed to an effective connectivity statistic), because it is easy to implement, it is relatively light in assumptions while still informative, and, used as a predictor of behavioural traits, it can surpass the accuracy of static FC estimates (Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 4 Evaluating dynamic FC models In the previous section, we outlined some representative methods for dynamic FC estimation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this section, we discuss how to validate these models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 Testing against null models A prevalent discussion in dynamic FC research (particularly in the sense of time-varying) is the question whether FC meaningfully fluctuates over time, or whether it is actually stable over time and temporal fluctuations are mainly due to noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Depending on the criteria, it has both been argued that real fMRI data fluctuates over time (Zalesky et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014) and that it is difficult to reject the statistical hypothesis that FC is stationary (Hindriks et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2016;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Lindquist et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An argument for the latter view is that what may be interpreted as dynamic FC can in fact be explained, for instance, by sampling variability (Laumann et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017), which may in turn be affected by denoising strategies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Simulation studies also address whether static FC is being driven by transient (dynamic) events (Zamani Esfahlani et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020) or whether the presence of transient events in FC may be driven by static FC (Ladwig et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2022;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Novelli & Razi, 2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Although some time-varying FC methods are in principle equipped to find when the data is not time-varying (for example, in a state-based model by removing all states but one), in practice they may find temporal 34 variation in FC due to statistical noise and sampling variability even if FC in the data was really stationary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can see this effect of sampling variability by simulating data where FC does not vary over time, using real data as a starting point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In this case, a null model is a model that assumes stationarity of FC while preserving as many of the general distribution properties of real fMRI data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can, for example, generate data from a multivariate Gaussian distribution that preserves the mean and variance of the real fMRI data: T = 1000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' n_areas = size(tc_real,2);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mu = mean(tc_real);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Sigma = cov(tc_real);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' tc_mvn = mvnrnd(mu, Sigma,T);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can now compare the sliding window correlations between the real data and the simulated data (for simplicity, we only look at the correlation between the first two areas): window_length = 10;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' real_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mvn_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = 1:T - window_length + 1 real_dFC_tmp(:,:,t) = corr(tc_real(t:t+window_length-1,:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' real_dFC(1,t) = real_dFC_tmp(1,2,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mvn_dFC_tmp(:,:,t) = corr(tc_mvn(t:t+window_length-1,:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mvn_dFC(1,t) = mvn_dFC_tmp(1,2,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end As we show in Figure 7, for a sufficiently small window size, the sliding-window estimation of FC in the surrogate data will vary considerably over time (see Figure 7B, middle panel;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' and see also Figure 5), even though the true covariance of the data is static.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can increase the window to see if we can recover the true (static) nature of the correlation: 35 window_length = 200;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' real_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mvn_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = 1:T - window_length + 1 real_dFC_tmp(:,:,t) = corr(tc_real(t:t+window_length-1,:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' real_dFC(1,t) = real_dFC_tmp(1,2,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mvn_dFC_tmp(:,:,t) = corr(tc_mvn(t:t+window_length-1,:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' mvn_dFC(1,t) = mvn_dFC_tmp(1,2,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end As we increase the window size, the correlation in the surrogate data flattens out (see Figure 7B, bottom panel), while we can still see some temporal variation in the sliding window correlations of the real data (see Figure 7A, bottom panel).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, it may seem as though FC in the real fMRI data varies more over time than FC in the surrogate data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' To test this hypothesis, we would generate many random time courses from the multivariate Gaussian distribution and test for significant differences in variance between the real fMRI data and the surrogate data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, multivariate Gaussian null data is not a very appropriate null model in this case, since it does not capture any of the autocorrelations present in real fMRI data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 36 Figure 7 Is FC actually dynamic or are temporally changing estimates of FC a result of sampling variability?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A) Real fMRI timecourses for two regions (top row) and their corresponding sliding window correlations at a window size of 10 TRs (middle row) and 200 TRs (bottom row).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B) Surrogate timecourses for two regions, randomly generated from a multivariate Gaussian distribution with the same mean and covariance as the real fMRI data (top row), and their corresponding sliding window correlations at a window size of 10 TRs (middle row) and 200 TRs (bottom row).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An important step in the development and evaluation of dynamic FC methods is therefore to have suitable null models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, defining the distribution of real fMRI data and setting criteria for stationarity is not straightforward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As a more adequate alternative to multivariate Gaussian models, we can use null models that preserve not only the mean and variance, but also the autocorrelation structure of the real fMRI data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We will demonstrate this following the example in Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017) to generate surrogate data from a first order autoregressive (AR-1) model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' To do this, we first estimate the 1-lag autoregressive distribution parameters of the real fMRI data: A Real data B Surrogate data (multivariate Gaussian) 300 300 200 200 de 100 100 plituc 100 100 200 200 300 300 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 9001000 Time Time Sliding window correlations (window size: 10 TR) icient Correlation coefficient coeffic 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 Correlation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 Time Time Sliding window correlations (window size: 200 TR) icient Correlation coefficient coeffi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 rrelation 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content="5 0 100 200 300 400 500 600 700 800 900 0 100 200 300 400 500 600 700 800 900 Time Time37 Y = tc_real(2:end,:)';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X(1,:) = ones(1,T-1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" X(2:n_areas+1,:) = tc_real(1:T-1, :)';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" beta = (Y*X')/(X*X');" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' residuals = Y-beta*X;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We then generate synthetic data with these distribution parameters: c = beta(:,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' weights = beta(:, 2:n_areas+1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" tc_real_flip = tc_real';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' tc_arr = zeros(T, n_areas);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" tc_arr(1,:) = tc_real_flip(:,randi(T-1))';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' rand_t = randperm(T-1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' res_mu = mean(residuals,2);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" res_Sigma = cov(residuals');" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" noise = (mvnrnd(res_mu,res_Sigma,T-1))';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" for i = 2:T tc_arr(i,:) = c' + (weights*tc_arr(i-1,:)')' + ." metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" noise(:,rand_t(i-1))';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end Finally, we again compute the sliding window correlations in the surrogate timeseries and compare them to the sliding window correlations of the real fMRI data we computed earlier: window_length = 200;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' arr_dFC_tmp = zeros(n_areas, n_areas, T - window_length + 1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for t = 1:T - window_length + 1 arr_dFC_tmp(:,:,t) = corr(tc_arr(t:t + window_length-1,:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' arr_dFC(1,t) = arr_dFC_tmp(1,2,t);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end The resulting surrogate timeseries and corresponding sliding window correlations are shown in Figure 8B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Here, the sliding window correlations of the surrogate data have similar temporal variance to the real fMRI data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 38 Figure 8 When generating surrogate data from a model that preserves the autocorrelation structure of the real fMRI data, such as a 1-lag autoregressive randomisation model, the null hypothesis cannot easily be rejected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In fact, several studies (Hindriks et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2016;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Prichard & Theiler, 1994) showed that for real fMRI data, the null hypothesis often cannot be rejected when using surrogate data that preserves the autocorrelation structure of the real data, such as ARR models or models based on phase randomisation (PR).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The difference between the ARR model and the PR model is how much of the autocovariance structure they preserve (the PR model preserves the entire autocovariance structure, while the p-th order ARR model preserves only the first p + 1 autocovariances).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' It is however important to note that, since the assumption of both the AR model and the PR model is that fMRI data are realisations of a linear, Gaussian, and stationary process, a rejection of the null hypothesis in this case could be due to any of these assumptions being untrue (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', not necessarily stationarity).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' That is, not being able to reject the null hypothesis does not necessarily imply that the data is stationary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017) also showed that the opposite conclusion also does not hold, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', being able to reject the null hypothesis does not necessarily imply that the data is dynamic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While testing fMRI data for non-stationarity is therefore not trivial, tests against null models can be useful to understand which information different models use in their estimation of FC dynamics (Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A Real data B Surrogate data (1st order autoregressive randomisation model) 300 400 200 300 200 itude Amplitude 100 A .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='100 100 200 200 300 300 U 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 Time Time Sliding window correlations (window size: 200 TR) fficient Correlation coefficient 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 coel relation Cor 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5 100 200 300 400 500 600 700 800 900 0 100 200 300 400 500 600 700 800 900 Time Time39 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 Comparing models with simulated data An important question is how we can test the advantages and disadvantages of the different methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A powerful way to do this is through the use of synthetic data, where we can know and control the ground-truth generative model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' There are many ways to generate synthetic data;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' here, we discuss as an illustration a simple procedure to compare the basic clustering approach to the HMM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The idea is to use real fMRI data as a starting point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This approach is described in detail in Ahrends et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For example, let us assume that we have data from ten regions in the brain from one subject.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can compute the covariance matrix of those real data and generate data from a Gaussian distribution, which we can smooth to make it look a bit more like real fMRI data: some_time = 5000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' p = 10;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C = cov(real_data);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X = mvnrnd(zeros(some_time,p),C);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for j = 1:p X(:,j) = smooth(X(:,j),10);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end Now, from this empirical covariance matrix we can generate covariance matrices with small variations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For this, we can perform a singular value decomposition of the empirical covariance matrix, permute the low-order eigenvectors, and reassemble the matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Depending on how many eigenvectors we permute (in the following code the last J eigenvectors) and their corresponding eigenvalues, we can make the variations larger or smaller: 40 [U,S,~] = svd(C);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' e = diag(S);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' e = cumsum(e) / sum(e);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C_synth = zeros(size(C));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' % less than 90% variation J = find(e>=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='9,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" for j = 1:J C_synth = C_synth + U(:,j) * S(j,j) * U(:,j)';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end for j = J+1:10 c = U(randperm(size(U,1)),j);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" C_synth = C_synth + c * S(j,j) * c';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end We can use this code to generate a set of states, each with a different (but still relatively similar) covariance matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This way, the larger the variations (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', the largest is the sum of the eigenvalues of the permuted eigenvectors), the stronger the between-state differences and the easier it is to find the ground-truth states by the HMM or any other method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We can then sample ground-truth state time courses, for four states for example, as total_session_duration = 20000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' max_duration = 100;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 5;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' true_stc = zeros(total_session_duration,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' t=1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' while t <= total_session_duration k = randi(4,1);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L = randi(max_duration-min_duration,1) + min_duration;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' if t+L-1 > total_session_duration L = total_session_duration - t + 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end true_stc(t:t+L-1,k) = 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' t = t + L + 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' end In this case, we randomly sampled the identity of the state from a categorical distribution and the duration of the state visits uniformly from the interval [min_duration,max_duration], which corresponds to a form of semi-Markov process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Many other sampling schemes are possible, with which we can interrogate the practical relevance of the models’ assumptions (for instance, the HMM’s Markovianity;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' see below).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 41 Altogether, we can put these different pieces together to sample synthetic data sets: sample of covariance matrices, sample of state time courses, sample of data, and final smoothing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Next, we will use simulated data from this scheme to compare the basic clustering approach and the HMM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We followed these steps to generate 200 data sets for each of six different scenarios: Slower transitions (max_duration = 1000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 50) and larger state variations (J = 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Faster transitions (max_duration = 100;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 5) and larger state variations (J = 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Slower transitions (max_duration = 1000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 50) and moderate state variations (J = 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Faster transitions (max_duration = 100;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 5) and moderate state variations (J = 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Slower transitions (max_duration = 1000;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 50) and smaller state variations (J = 6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Faster transitions (max_duration = 100;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' min_duration = 5) and smaller state variations (J = 6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' On each of these 4 x 200 data sets, we ran the basic clustering approach and the HMM, both endowed with four states, matching the ground truth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Each method produced an estimated state time course.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Since the order of the states is not identifiable (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', the first state in the ground-truth state time courses could correspond to the third state of the estimated state time courses), we used the Hungarian algorithm (Kuhn, 1955) to align them (as implemented in the HMM-MAR toolbox).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We then computed the accuracy as the correlation of the state probabilities with the ground-truth state time courses 42 options = struct();' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' options.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='K = 4;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' options.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content="covtype = 'full';" metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' options.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='zeromean = 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' % states only have a covariance matrix [hmm,stc] = hmmmar(X,size(X,1),options);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' [assig,cost] = munkres(1-corr(true_stc,stc));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' stc = stc(:,assig);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' accuracy = corr(stc(:),true_stc(:));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Figure 9 shows the accuracy of each run per method and scenario (dots where a jitter was introduced in the x-axis for ease of visualisation), together with the across-run average (bars).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As observed, the HMM fared better than the basic clustering approach for two reasons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' First, because the estimation of the states for the clustering method is based on noisy sliding window estimates, whereas the HMM estimates the states directly from the data using a much larger number of time points and, therefore, does it so much more precisely;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' for this reason, the difference between the HMM and the clustering approach is considerably smaller when the state covariance matrices are very different between states (upper row of panels).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Second, because the clustering approach cannot detect fast changes (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', faster than the length of the window), whereas the HMM states can switch as quickly as necessary;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' that is why the accuracies for clustering approach are lower in the right panels, whereas the speed of the ground-truth state transitions makes less of a difference for the HMM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 43 Figure 9 Performance of the standard clustering approach vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' the HMM in estimating the state transitions sampled from a state-based generative model where the state distributions are covariance matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Both how different the states are, and the speed of the transitions are manipulated to create simpler or harder estimation problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Each dot represents a run (a sampled data set and the corresponding estimates) and the bars are averages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Although not shown here, LEiDA performed notably worse than the other methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The reason is that, while LEiDA’s states are based on the first principal component of matrices of phase relationships, the states of both the HMM and the clustering approach are actual covariance matrices, matching the ground-truth state model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is a general principle of this type of simulations: the more accurate the assumptions of a method are with respect to the ground-truth generative process, the more accurate its estimates will be.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' It is important to note that, in real data, where the generator of the data is the brain, LEiDA will just offer a different perspective: in this case, we can hardly say that one model is more correct than another.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Slower transitions, larger state variations Fastertransitions,largerstate variations 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 Accuracy racy 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Accur 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0 0 Clustering HMM Clustering HMM Slower transitions, medium state variations Faster transitions, medium state variations 1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 racy 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 Accur ccur 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 A 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0 0 Clustering HMM Clustering HMM Slower transitions, smaller state variations Faster transitions, smaller state variations 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='8 Accuracy 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 0 0 Clustering HMM Clustering HMM44 5 Challenges & perspectives Having discussed the most important practical aspects of conducting dynamic FC analyses, we will now outline some challenges and perspectives in the current development of dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Challenges and questions in the dynamic FC research field are also extensively discussed in Lurie et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019), so we here only highlight a few key ideas related to reliability and reproducibility, and the use of dynamic FC to predict behavioural and clinical traits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1 Reliability & reproducibility A key challenge in the study of dynamic FC is the reliability and reproducibility of different methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As a consequence, it can be difficult to compare results from dynamic FC studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Reliability and reproducibility in dynamic FC studies may mean several things.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Reliability may pertain to the question: When we have several scanning sessions of a subject, do we reliably find the same patterns of dynamic FC in all sessions?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We consider this the question of test-retest reliability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Studies addressing this question have shown that test-retest reliability of many dynamic FC methods is low, particularly for sliding window-based approaches (Choe et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Zhang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' To what extent this is due to methodological limitations or systematic within-subject variability is an important topic for future research into dynamic FC reliability (Geerligs et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Reliability may also refer to the question: How reliable is the method in detecting dynamic FC?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This question can be addressed by simulation studies or by studying how biased a method is by secondary parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For instance, Ahrends et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2022) showed how a HMM’s ability to detect temporal changes in FC depends on several parameters that are determined by preprocessing strategies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Optimising preprocessing strategies (as described 45 in the previous section Preparing data for dynamic FC analyses) is therefore an important step in improving reliability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Other studies have investigated the reproducibility of dynamic FC methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Reproducibility may refer to the question: Do we find consistent, canonical patterns of resting-state dynamic FC across multiple datasets?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For instance, Abrol et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017) tested this question across datasets from different scanning sites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Using two different state-based approaches, they could show that the same basic connectivity patterns emerge in all datasets, indicating that general patterns in dynamic FC are reproducible and robust to variations in the specific datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In some methods, reproducibility also entails the question: Do we get the same results when we run the model several times?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In models that depend on random initialisation, solutions may to some extent be different every time the model is run (Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This issue may also be referred to as robustness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018) have shown that dynamic FC results from HMMs which use random initialisation for the inference are reproducible across different runs of the model to some extent, but that the similarity between runs depends on the dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Robustness may also refer to the issue of how the model deals with noise, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', in terms of noisy fMRI recordings, or in terms of outliers in the dataset, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', does the estimation of dynamic FC in a group of subjects suffer if single subjects display anomalous patterns of dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This question is only starting to be addressed for different dynamic FC methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Assessing and improving reliability, reproducibility, and robustness of dynamic FC should be a central task in future dynamic FC research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This may not only improve the estimation of dynamic FC but also make methods and results more comparable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 46 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2 Predicting behaviour and individual traits from dynamic FC Time-varying FC measures have been shown to be highly individual, suggesting that they could be used for “fingerprinting” an individual (J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Beyond explaining behaviour, there is thus growing interest in using dynamic FC to predict individual traits and behaviour with a potential to develop dynamic FC-based biomarkers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' So far, studies using features derived from dynamic FC were able to predict various phenotypes and behavioural measures at moderate to fair accuracy, such as cognitive performance or intelligence (J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Sen & Parhi, 2021;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Shine et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019), task performance (Fong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019), or sleep quality (Zhou et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020), as well as to classify clinical diagnoses, such as schizophrenia (Bhinge et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cetin et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2016) or post- traumatic stress disorder (PTSD) (Ou et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is an interesting avenue, because dynamic FC contains information complementary to structural measures or static FC (Ge et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Liégeois et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2021), and so it may be able to predict distinct aspects of behaviour.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' It has in fact been shown that dynamic FC can outperform structural and static measures in predicting behaviour (Saha et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2021;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Vergara et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Vidaurre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2021) and in classifying neurological and psychiatric disease (Jin et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Rashid et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For a review on prediction and classification from static and dynamic FC in clinical applications, see Du et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' But how can we best use the information from dynamic FC models in the context of prediction problems?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' One option is to select a feature of interest, such as the occurrence of a specific network, as a predictor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This is the approach taken in most studies so far.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, not only is it not obvious how to choose the relevant feature in the context of neuroimaging (Du et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2018;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Wolfers et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2015), doing this would also mean losing information that was originally contained in the dynamic FC model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Instead, we may want to use all information from subject-level dynamic FC models in a linear predictive model to predict these subjects’ phenotype.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' While static measures can relatively straightforwardly 47 (with the exception of geometric constraints) be used in linear predictive models, the temporal dimension and complexity of dynamic FC models make this more difficult.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Developing new approaches to predicting from dynamic FC models is a central goal in current dynamic FC research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' One promising avenue are kernel methods (Shawe-Taylor & Cristianini, 2004).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' These methods have several advantages: They allow working with high- dimensional features such as dynamic FC features in a computationally efficient way by using a subject-by-subject similarity matrix for the prediction rather than the features themselves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Kernels can be constructed in a way that preserves the structure of the features (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' the scales and relationships between parameters of an underlying dynamic FC model).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Furthermore, kernels can be used straightforwardly in linear prediction models or classifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Indeed, by using kernel functions that apply a nonlinear transformation to the data, these linear prediction models or classifiers can be used to detect nonlinear decision boundaries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This elegantly combines the advantages of linear models, which are computationally efficient, easy to implement, and can be more readily interpreted, with nonlinear models, which allow finding more complex relationships and patterns in data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Kernel methods also open the door to predicting from multiple modalities, such as structural and functional, static and dynamic information, or MRI and MEG/EEG (Engemann et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Schouten et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Using a Multi-kernel learning (MKL) approach (Gönen & Alpaydın, 2011), separately constructed kernels for each modality can be combined within a single prediction model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' This approach has been shown to improve prediction accuracy, for example when combining MRI and MEG (Vaghari et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2022) or several levels of FC estimation (Zhang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', 2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 6 Conclusions In this chapter, we have given a practical introduction to different methods of dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' We also discussed data preparation and model evaluation, as well as some challenges and perspectives in the dynamic FC research field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' As we have shown, there is a variety of 48 methods to estimate dynamic FC, each with unique advantages but also shortcomings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Recent years have seen an explosion of dynamic FC studies, indicating the great potential of a dynamic view of FC in answering specific research questions and in contributing to our general understanding of brain function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' However, we also stress the importance of methodological rigour in the further development of these methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' For instance, dynamic FC models may be more heavily biased by preprocessing strategies than other methods, because they are more sensitive to temporal noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Another not fully resolved issue is the comparison to null models that assume stationarity, an important limitation that should be kept in mind in both methods development and the conceptualisation of dynamic FC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Going forward, we hope to see further work on these issues, as well as explicit testing and improvement of the reliability and reproducibility of the different methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' References Abrol, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Damaraju, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Stephen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Claus, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mayer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Replicability of time-varying connectivity patterns in large resting state fMRI samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 163, 160-176.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='09.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='020 Afyouni, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Nichols, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" Effective degrees of freedom of the Pearson's correlation coefficient under autocorrelation." metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 199, 609-625.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='011 Ahrends, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Stevner, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Pervaiz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kringelbach, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Vuust, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Data and model considerations for estimating time-varying functional connectivity in fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 252, 119026.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='119026 Akaike, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1974).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A new look at the statistical model identification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' IEEE Transactions on Automatic Control, 19(6), 716-723.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1109/TAC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1974.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1100705 Allen, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Damaraju, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Plis, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Erhardt, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Eichele, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Tracking Whole-Brain Connectivity Dynamics in the Resting State.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cerebral Cortex, 24(3), 663-676.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1093/cercor/bhs352 Beckmann, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mackay, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Filippini, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Group comparison of resting-state FMRI data using multi-subject ICA and dual regression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 47, S148.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/S1053-8119(09)71511-3 Bhinge, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Long, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Adali, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Spatial Dynamic Functional Connectivity Analysis Identifies Distinctive Biomarkers in Schizophrenia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Front Neurosci, 13, 1006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='3389/fnins.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='01006 49 Bijsterbosch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Harrison, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Duff, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Alfaro-Almagro, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Investigations into within- and between-subject resting-state amplitude variations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 159, 57-69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='07.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='014 Bishop, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Pattern Recognition and Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Springer-Verlag.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cabral, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Marques, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Magalhaes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Silva Moreira, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miguel Soares, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Deco, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sousa, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Kringelbach, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cognitive performance in healthy older adults relates to spontaneous switching between states of functional connectivity during rest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Sci Rep, 7(1), 5135.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1038/s41598-017- 05425-7 Cai, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Stephen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Wilson, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Capturing Dynamic Connectivity from Resting State fMRI using Time-Varying Graphical Lasso.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' IEEE Trans Biomed Eng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1109/tbme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2880428 Calhoun, Vince D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Pearlson, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Adalı, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The Chronnectome: Time- Varying Connectivity Networks as the Next Frontier in fMRI Data Discovery.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuron, 84(2), 262-274.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuron.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='015 Cetin, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Houck, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rashid, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Agacoglu, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Stephen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sui, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Canive, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mayer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Aine, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bustillo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multimodal Classification of Schizophrenia Patients with MEG and fMRI Data Using Static and Dynamic Connectivity Measures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Front Neurosci, 10, 466.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='3389/fnins.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='00466 Choe, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Nebel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Barber, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cohen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Pekar, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Caffo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Lindquist, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Comparing test-retest reliability of dynamic functional connectivity methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 158, 155-175.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='07.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='005 Ciric, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Wolf, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Power, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Roalf, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Baum, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ruparel, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Shinohara, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Elliott, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Eickhoff, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Davatzikos, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gur, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gur, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bassett, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Satterthwaite, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Benchmarking of participant-level confound regression strategies for the control of motion artifact in studies of functional connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 154, 174-187.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='03.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='020 Du, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Fu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Classification and Prediction of Brain Disorders Using Functional Connectivity: Promising but Challenging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Front Neurosci, 12, 525.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='3389/fnins.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='00525 Engemann, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kozynets, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sabbagh, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Lemaître, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Varoquaux, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Liem, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Gramfort, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Combining magnetoencephalography with magnetic resonance imaging enhances learning of surrogate-biomarkers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' eLife, 9, e54055.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='7554/eLife.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='54055 Faes, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Nollo, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multivariate frequency domain analysis of causal interactions in physiological time series.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' In A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Laskovski (Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' ), Biomedical Engineering, Trends in Electronics, Communications and Software (Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 8, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 403-428).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' InTech.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='intechopen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='com/books/biomedical-engineering-trends-in-electronics- communications-and-software/multivariate-frequency-domain-analysis-of-causal- interactions-in-physiological-time-series Filippi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' fMRI Techniques and Protocols (2nd 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Springer New York.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1007/978-1-4939-5611-1 Fong, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Yoo, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rosenberg, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Li, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='-S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Scheinost, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Constable, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Chun, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Dynamic functional connectivity during task 50 performance and rest predicts individual differences in attention across studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 188, 14-25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='057 Frahm, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bruhn, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Merboldt, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Hänicke, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1992).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Dynamic MR imaging of human brain oxygenation during rest and photic stimulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J Magn Reson Imaging, 2(5), 501-505.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1002/jmri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1880020505 Friedman, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Hastie, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Tibshirani, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2008).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Sparse inverse covariance estimation with the graphical lasso.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Biostatistics, 9(3), 432-441.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1093/biostatistics/kxm045 Friston, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Functional and effective connectivity: a review.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Brain Connect, 1(1), 13- 36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1089/brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='0008 Friston, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ashburner, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kiebel, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Nichols, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Penny, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Statistical Parametric Mapping: The Analysis of Functional Brain Images.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Elsevier Science & Technology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/B978-0-12-372560-8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='X5000-1 Ge, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Holmes, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Buckner, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smoller, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Sabuncu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Heritability analysis with repeat measurements and its application to resting-state functional connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proc Natl Acad Sci U S A, 114(21), 5521-5526.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1700765114 Geerligs, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rubinov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cam, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Henson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' State and Trait Components of Functional Connectivity: Individual Differences Vary with Mental State.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J Neurosci, 35(41), 13949-13961.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1523/jneurosci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1324-15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2015 Glasser, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sotiropoulos, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Wilson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Coalson, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Fischl, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Andersson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jbabdi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Webster, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Polimeni, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Van Essen, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Jenkinson, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The minimal preprocessing pipelines for the Human Connectome Project.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 80, 105-124.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='04.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='127 Griffanti, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Salimi-Khorshidi, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Beckmann, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Auerbach, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Douaud, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sexton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zsoldos, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ebmeier, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Filippini, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mackay, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Moeller, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Yacoub, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Baselli, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ugurbil, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' ICA-based artefact removal and accelerated fMRI acquisition for improved resting state network imaging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 95, 232-247.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='03.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='034 Gönen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Alpaydın, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multiple kernel learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The Journal of Machine Learning Research, 12, 2211-2268.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' http://jmlr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/papers/v12/gonen11a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='html Hair, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multivariate data analysis (7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Prentice Hall.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://digitalcommons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='kennesaw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='edu/facpubs/2925/ Hair, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multivariate data analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Harrison, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Penny, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Friston, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Multivariate autoregressive modeling of fMRI time series.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 19(4), 1477-1491.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/s1053- 8119(03)00160-5 Hindriks, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Adhikari, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Murayama, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ganzetti, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mantini, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Logothetis, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Deco, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Can sliding-window correlations reveal dynamic functional connectivity in resting-state fMRI?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 127, 242-256.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='055 Hutchison, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Womelsdorf, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Allen, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bandettini, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Corbetta, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Della Penna, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Duyn, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Glover, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gonzalez-Castillo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Handwerker, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' 51 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Keilholz, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kiviniemi, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Leopold, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', de Pasquale, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sporns, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Walter, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Chang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Dynamic functional connectivity: Promise, issues, and interpretations [Article].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 80, 360-378.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='079 Iraji, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Fu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', DeRamus, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Qi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rachakonda, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Du, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Ultra- high-order ICA: fine overlapping functional parcellations and spatiotemporal reconfiguration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' bioRxiv, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='986299.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1101/2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='03.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='986299 Jin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jia, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Lanka, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rangaprakash, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Li, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Liu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Hu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Deshpande, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Dynamic brain connectivity is a better predictor of PTSD than static connectivity [Article].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Human Brain Mapping, 38(9), 4479-4496.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1002/hbm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='23676 Kalaba, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Tesfatsion, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1989).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Time-varying linear regression via flexible least squares.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Computers & Mathematics with Applications, 17(8), 1215-1245.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/0898-1221(89)90091-6 Kuhn, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1955).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The Hungarian method for the assignment problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Naval Research Logistics (NRL), 2, 83–97.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1007/978-3-540-68279-0_2 Kwong, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Belliveau, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Chesler, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Goldberg, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Weisskoff, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Poncelet, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kennedy, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Hoppel, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cohen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Turner, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cheng, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Brady, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Rosen, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1992).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Dynamic Magnetic Resonance Imaging of Human Brain Activity During Primary Sensory Stimulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proceedings of the National Academy of Sciences of the United States of America, 89(12), 5675-5679.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='jstor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/stable/2359716 Ladwig, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Seitzman, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Dworetsky, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Yu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Adeyemo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smith, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Petersen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Gratton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' BOLD cofluctuation ‘events’ are predicted from static functional connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 260, 119476.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='119476 Laumann, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Snyder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mitra, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gordon, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gratton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Adeyemo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gilmore, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Nelson, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Berg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Greene, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', McCarthy, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Tagliazucchi, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Laufs, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Schlaggar, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Dosenbach, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Petersen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' On the Stability of BOLD fMRI Correlations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cerebral cortex (New York, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' : 1991), 27(10), 4719-4732.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1093/cercor/bhw265 Lee, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smyser, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Shimony, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Resting-state fMRI: a review of methods and clinical applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' AJNR Am J Neuroradiol, 34(10), 1866-1872.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='3174/ajnr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='A3263 Lehmann, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', White, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Henson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cam, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Geerligs, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Assessing dynamic functional connectivity in heterogeneous samples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 157, 635-647.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='065 Liao, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Wu, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ji, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Lu, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' DynamicBC: a MATLAB toolbox for dynamic brain connectome analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Brain Connect, 4(10), 780- 790.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1089/brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='0253 Liégeois, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Laumann, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Snyder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhou, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Yeo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Interpreting temporal fluctuations in resting-state functional connectivity MRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 163, 437-455.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='09.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='012 Liégeois, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Li, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kong, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Orban, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Van De Ville, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ge, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sabuncu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Yeo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Resting brain dynamics at different timescales capture distinct aspects 52 of human behavior.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Nat Commun, 10(1), 2317.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1038/s41467-019- 10317-7 Liégeois, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Yeo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Van De Ville, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Interpreting null models of resting-state functional MRI dynamics: not throwing the model out with the hypothesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 243, 118518.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='118518 Lindquist, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Nebel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Caffo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Evaluating dynamic bivariate correlations in resting-state fMRI: A comparison study and a new approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 101, 531-546.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='06.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='052 Liu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Liao, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xia, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & He, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Chronnectome fingerprinting: Identifying individuals and predicting higher cognitive functions using dynamic brain connectivity patterns.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Hum Brain Mapp, 39(2), 902-915.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1002/hbm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='23890 Liu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Chang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Duyn, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Co-activation patterns in resting-state fMRI signals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 180(Pt B), 485-494.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='041 Lurie, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kessler, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bassett, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Betzel, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Breakspear, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kheilholz, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kucyi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Liégeois, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Lindquist, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', McIntosh, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Poldrack, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Shine, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Thompson, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bielczyk, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Douw, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kraft, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Muthuraman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Pasquini, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Questions and controversies in the study of time-varying functional connectivity in resting fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Network Neuroscience, 4(1), 30- 69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1162/netn_a_00116 Lydon-Staley, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ciric, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Satterthwaite, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Bassett, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Evaluation of confound regression strategies for the mitigation of micromovement artifact in studies of dynamic resting-state functional connectivity and multilayer network modularity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Network Neuroscience, 3(2), 427-454.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1162/netn_a_00071 Margulies, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ghosh, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Goulas, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Falkiewicz, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Huntenburg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Langs, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bezgin, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Eickhoff, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Castellanos, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Petrides, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jefferies, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Smallwood, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Situating the default-mode network along a principal gradient of macroscale cortical organization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proceedings of the National Academy of Sciences, 113(44), 12574-12579.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1608282113 Monti, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Hellyer, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sharp, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Leech, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Anagnostopoulos, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Montana, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Estimating time-varying brain connectivity networks from functional MRI time series.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 103, 427-443.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='07.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='033 Murphy, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Fox, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Towards a consensus regarding global signal regression for resting state functional connectivity MRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 154, 169-173.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='052 Nalci, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rao, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Liu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Nuisance effects and the limitations of nuisance regression in dynamic functional connectivity fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 184, 1005-1031.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='09.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='024 Novelli, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Razi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A mathematical perspective on edge-centric brain functional connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Nature Communications, 13(1), 2693.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1038/s41467- 022-29775-7 Ogawa, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Tank, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Menon, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ellermann, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Merkle, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Ugurbil, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1992).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proc Natl Acad Sci U S A, 89(13), 5951- 5955.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='89.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='5951 53 Ou, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Xie, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jiang, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Chen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Li, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Liu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Characterizing and Differentiating Brain State Dynamics via Hidden Markov Models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Brain Topography, 28(5), 666-679.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1007/s10548-014-0406-2 Parkes, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Fulcher, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Yücel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Fornito, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 171, 415-436.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='073 Pervaiz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Optimising network modelling methods for fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 211, 116604.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='116604 Power, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Barnes, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Snyder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Schlaggar, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Petersen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 59(3), 2142-2154.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='018 Power, Jonathan D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cohen, Alexander L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Nelson, Steven M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Wig, Gagan S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Barnes, Kelly A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Church, Jessica A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Vogel, Alecia C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Laumann, Timothy O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miezin, Fran M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Schlaggar, Bradley L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Petersen, Steven E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Functional Network Organization of the Human Brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuron, 72(4), 665-678.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuron.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='09.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='006 Preti, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bolton, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Van De Ville, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The dynamic functional connectome: State-of-the-art and perspectives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 160, 41-54.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='061 Preti, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Van De Ville, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Dynamics of functional connectivity at high spatial resolution reveal long-range interactions and fine-scale organization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Scientific Reports, 7(1), 12773.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1038/s41598-017-12993-1 Prichard, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Theiler, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (1994).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Generating surrogate data for time series with several simultaneously measured variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Phys Rev Lett, 73(7), 951-954.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1103/PhysRevLett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='951 Rashid, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Arbabshirani, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Damaraju, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cetin, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Pearlson, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Classification of schizophrenia and bipolar patients using static and dynamic resting-state fMRI brain connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 134, 645-657.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='04.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='051 Rogers, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Katwal, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Morgan, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Asplund, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Gore, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Functional MRI and multivariate autoregressive models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Magn Reson Imaging, 28(8), 1058- 1065.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='mri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='03.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='002 Saha, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Damaraju, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Rashid, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Abrol, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Plis, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A Classification-Based Approach to Estimate the Number of Resting Functional Magnetic Resonance Imaging Dynamic Functional Connectivity States.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Brain Connect, 11(2), 132-145.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1089/brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='0794 Schouten, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Koini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', de Vos, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Seiler, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', van der Grond, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Lechner, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Hafkemeijer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Möller, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Schmidt, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', de Rooij, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Rombouts, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=" Combining anatomical, diffusion, and resting state functional magnetic resonance imaging for individual classification of mild and moderate Alzheimer's disease." metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' NeuroImage: Clinical, 11, 46-51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='nicl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='002 Sen, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Parhi, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Predicting Biological Gender and Intelligence From fMRI via Dynamic Functional Connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' IEEE Transactions on Biomedical Engineering, 68(3), 815-825.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1109/TBME.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='3011363 54 Shappell, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Caffo, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Pekar, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Lindquist, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Improved state change estimation in dynamic functional connectivity using hidden semi-Markov models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 191, 243-257.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='02.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='013 Shawe-Taylor, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Cristianini, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2004).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Kernel Methods for Pattern Analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Cambridge University Press.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1017/CBO9780511809682 Shine, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Breakspear, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bell, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Ehgoetz Martens, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Shine, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Koyejo, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sporns, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Poldrack, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Human cognition involves the dynamic integration of neural activity and neuromodulatory systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Nat Neurosci, 22(2), 289- 296.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1038/s41593-018-0312-0 Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Beckmann, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Andersson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Auerbach, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Bijsterbosch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Douaud, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Duff, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Feinberg, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Griffanti, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Harms, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kelly, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Laumann, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Moeller, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Petersen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Power, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Salimi-Khorshidi, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Snyder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Vu, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Glasser, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Resting-state fMRI in the Human Connectome Project.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 80, 144-168.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='039 Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Beckmann, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Glasser, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jenkinson, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Miller, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Nichols, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Robinson, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Salimi-Khorshidi, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Barch, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Uğurbil, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Van Essen, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Functional connectomics from resting-state fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Trends in Cognitive Sciences, 17(12), 666-682.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='tics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='09.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='016 Vaghari, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kabir, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Henson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Late combination shows that MEG adds to MRI in classifying MCI versus controls.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 252, 119054.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='119054 Vergara, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Mayer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Damaraju, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' The effect of preprocessing in dynamic functional network connectivity used to classify mild traumatic brain injury.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Brain and Behavior, 7(10), e00809.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1002/brb3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='809 Vergara, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Salman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Abrol, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Espinoza, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Determining the number of states in dynamic functional connectivity using cluster validity indexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Journal of Neuroscience Methods, 337, 108651.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='jneumeth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='108651 Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A new model for simultaneous dimensionality reduction and time- varying functional connectivity estimation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' PLOS Computational Biology, 17(4), e1008580.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1371/journal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='pcbi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1008580 Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Abeysuriya, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Becker, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Quinn, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Alfaro-Almagro, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Discovering dynamic brain networks from big data in rest and task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 180, 646-656.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='06.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='077 Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Llera, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Behavioural relevance of spontaneous, transient brain network interactions in fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 229, 117713.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='117713 Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smith, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Brain network dynamics are hierarchically organized in time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proc Natl Acad Sci U S A, 114(48), 12827-12832.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1705120114 Vidaurre, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Woolrich, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Winkler, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Karapanagiotidis, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Smallwood, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Nichols, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Stable between-subject statistical inference from unstable within-subject functional connectivity estimates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Human Brain Mapping, 40(4), 1234-1243.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1002/hbm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='24442 55 Wolfers, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Buitelaar, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Beckmann, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Franke, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Marquand, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' From estimating activation locality to predicting disorder: A review of pattern recognition for neuroimaging-based psychiatric diagnostics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroscience & Biobehavioral Reviews, 57, 328-349.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neubiorev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='08.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='001 Yang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Craddock, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Margulies, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Yan, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='-G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Milham, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Common intrinsic connectivity states among posteromedial cortex subdivisions: Insights from analysis of temporal dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 93, 124-137.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='02.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='014 Zalesky, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Fornito, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cocchi, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Gollo, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Breakspear, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Time-resolved resting-state brain networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proceedings of the National Academy of Sciences, 111(28), 10341-10346.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1400181111 Zamani Esfahlani, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Jo, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Faskowitz, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Byrge, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Kennedy, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Sporns, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Betzel, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' High-amplitude cofluctuations in cortical activity drive functional connectivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Proceedings of the National Academy of Sciences, 117(45), 28393- 28401.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2005531117 Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Baum, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Adduru, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Biswal, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Michael, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Test-retest reliability of dynamic functional connectivity in resting state fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 183, 907-918.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='08.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='021 Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Chen, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Lee, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Shen, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Hybrid High-order Functional Connectivity Networks Using Resting-state Functional MRI for Mild Cognitive Impairment Diagnosis [Article].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Scientific Reports, 7(1), Article 6530.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1038/s41598-017-06509-0 Zhou, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Cai, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Zhang, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', Calhoun, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=', & Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='-P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Prediction and classification of sleep quality based on phase synchronization related whole- brain dynamic connectivity using resting state fMRI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' Neuroimage, 221, 117190.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='neuroimage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} +page_content='117190' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/YtE1T4oBgHgl3EQfwQVD/content/2301.03408v1.pdf'} diff --git a/ZNAzT4oBgHgl3EQfKvv0/content/tmp_files/2301.01104v1.pdf.txt b/ZNAzT4oBgHgl3EQfKvv0/content/tmp_files/2301.01104v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..caf208c118e8869921f3cfef83d631d5aaa481be --- /dev/null +++ b/ZNAzT4oBgHgl3EQfKvv0/content/tmp_files/2301.01104v1.pdf.txt @@ -0,0 +1,3079 @@ +KoopmanLab: A PyTorch module of Koopman neural operator family for solving partial +differential equations +Wei Xionga,c, Muyuan Maa, Pei Sunb,∗, Yang Tianb,c,∗ +aDepartment of Earth System Science, Tsinghua University, Beijing, 100084, China. +bDepartment of Psychology & Tsinghua Laboratory of Brain and Intelligence, Tsinghua University, Beijing, 100084, China. +cLaboratory of Advanced Computing and Storage, Central Research Institute, 2012 Laboratories, Huawei Technologies Co. Ltd., Beijing, 100084, China. +Abstract +Given the increasingly intricate forms of partial differential equations (PDEs) in physics and related fields, computationally solving +PDEs without analytic solutions inevitably suffers from the trade-off between accuracy and efficiency. Recent advances in neural +operators, a kind of mesh-independent neural-network-based PDE solvers, have suggested the dawn of overcoming this challenge. +In this emerging direction, Koopman neural operator (KNO) is a representative demonstration and outperforms other state-of-the- +art alternatives in terms of accuracy and efficiency. Here we present KoopmanLab, a self-contained and user-friendly PyTorch +module of the Koopman neural operator family for solving partial differential equations. Beyond the original version of KNO, we +develop multiple new variants of KNO based on different neural network architectures to improve the general applicability of our +module. These variants are validated by mesh-independent and long-term prediction experiments implemented on representative +PDEs (e.g., the Navier-Stokes equation and the Bateman–Burgers equation) and ERA5 (i.e., one of the largest high-resolution +data sets of global-scale climate fields). These demonstrations suggest the potential of KoopmanLab to be considered in diverse +applications of partial differential equations. +Keywords: Partial differential equation; Koopman neural operator; Dynamic system +PROGRAM SUMMARY +Program Title: KoopmanLab +CPC Library link to program files: (to be added by Technical Editor) +Developer’s +repository +link: +https://github.com/ +Koopman-Laboratory/KoopmanLab +Code Ocean capsule: (to be added by Technical Editor) +Licensing provisions: GNU General Public License 3 (GPL) +Programming language: Python, PyTorch +Nature of problem: Computationally solving partial differential equa- +tions (PDEs) by classic frameworks (e.g., finite element method) faces +the trade-off between accuracy and efficiency because these numerical +systems are mesh-dependent. +Fine mesh granularity ensures high +accuracy but inevitably implies high computation complexity and time +cost. Although different neural-network-based PDE solvers have been +proposed to accelerate the processes of solving PDEs and maintain +solution accuracy, most pioneer frameworks are not mesh-independent +or limited to solving simple equations. +Solution method: +The proposed KoopmanLab provides multiple +variants of Koopman neural operator (KNO), a mesh-independent +neural-network-based PDE solver with optimal performance in across- +mesh-granularity and long-term prediction. The compact variants of +KNO (e.g., implemented on multi-layer perceptrons and convolutional +neural networks) can achieve high accuracy PDE solving with small +model sizes while the large variants of KNO (e.g., implemented +∗Corresponding author. +E-mail address: tiany20@mails.tsinghua.edu.cn & tyanyang04@gmail.com +(Y.T); peisun@tsinghua.edu.cn (P.S) +on visual transformers) are more competitive in predicting highly +complicated dynamic systems govern by unknown, high-dimensional, +and non-linear PDEs. Compared with classic numerical solvers (e.g., +finite element method), the prediction time costs of all variants of +KNO are reduced by multiple orders of magnitude. +1. Introduction +1.1. The rising of partial differential equation solvers +Solving partial differential equations (PDEs) essentially re- +quires characterizing an appropriate solution operator F that +relates Φ = Φ +� +D; Rdφ� +, a Banach space of inputs (i.e., initial +values), with Γ = Γ +� +D; Rdγ� +, a Banach space of solutions (i.e., +target values), for a typically time-dependent PDE family de- +fined on a bounded open set D ⊂ Rd +∂tγ (xt) = +� +Lφγ +� +(xt) + κ (xt) , xt ∈ D × T, +(1) +γ (xt) = γB, xt ∈ ∂D × T, +(2) +γ (x0) = γI, x0 ∈ D × {0}. +(3) +In Eq. (1-3), set T = [0, ∞) denotes the time domain. Notion +Lφ is a differential operator characterized by φ. Mapping κ (·) +is a function that lives in a function space determined by Lφ. +Mapping γ (·) is the solution of the PDE family that we attempt +to obtain. The boundary and initial conditions are denoted by +Preprint submitted to Computer Physics Communications +January 4, 2023 +arXiv:2301.01104v1 [cs.LG] 3 Jan 2023 + +γB and γI, respectively. Mathematically, driving an accurate +solution operator F : (φ, γB, γI) �→ γ is the key step to obtain +the PDE solution γ (·). However, even in the case where the +boundary and initial conditions are constant (i.e., the solution +operator F : (φ, γB, γI) �→ γ reduces to F : φ �→ γ), driving an +analytic expression of solution operator F can be highly non- +trivial [1, 2]. +The absence of analytic solutions of various important PDEs +in science and engineering naturally calls for the rapid develop- +ment of computational solvers, which attempt to approximate a +parametric counterpart Fθ ≃ F parameterized by θ to derive so- +lution γ (·) [1, 3, 4]. To date, the joint efforts of physics, math- +ematics, and computer science have given birth to two main- +stream families of PDE solvers [5]: +(1) The first family of solvers are classic numerical ones. +Typical instances of these solvers include finite element +(FEM) [6], finite difference (FDM) [7], and finite volume +(FVM) [8] methods. In general, these methods discretize +space and time domains following specific mesh designs +and solve parameterized PDEs on meshes by certain itera- +tive algorithms. Specifically, FEM subdivides the original +domain into a set of sub-domains defined by a collection +of element equations and recombines these element equa- +tions to derive the global solution [6]. FDM approximates +derivatives as finite differences measured on local values +[7]. FVM transforms the original problem into a series of +surface flux calculations on local volumes [8]. +(2) The second family of solvers are neural-network-based +ones. +With a pursuit of accelerating PDE solving and +improving the applicability on real data, three kinds of +neural-network-based solvers have been proposed: +(a) One kind of solvers discretize domains D and T into +x and y meshes and approximate a finite-dimensional +and mesh-dependent solution operator Fθ by a pa- +rameterized neural network between finite Euclidean +spaces, i.e., Fθ : Rx × Ry × Θ → Rx × Ry (e.g., see +Refs. [9, 10, 11]). Given an arbitrary input γ (xt), +the trained neural network can function as a solution +operator to predict γ (xt+τ) = Fθ (γ (xt)) for a certain +time difference τ. +(b) Another kind of solvers directly parameterize equa- +tion solution γ (·) as a neural network, i.e., Fθ : +D × T × Θ → R (e.g., see Refs. [12, 13, 14, 15]). +These solvers are mesh-independent and accurate in +learning a given PDE because they can directly trans- +form arbitrary domain and parameter setting to target +equation solution γ (·). +(c) The last kind of solvers, including neural opera- +tors, attempt to parameterize a mesh-dependent and +infinite-dimensional solution operator with neural +networks, i.e, Fθ : Φ × Θ → Γ (e.g., see Refs. +[16, 17, 18, 5, 19, 20, 21]). These mesh-independent +solvers can be flexibly implemented on different dis- +cretization schemes and only need to be trained once +for a given PDE family. The equation solution γ (·) +of different instances of the PDE family can be gen- +erated by a computationally reusable forward pass +of the network [5, 19], which can be further ac- +celerated by fast Fourier transform [19]. +Repre- +sentative demonstrations of this kind of solver are +Fourier neural operator [19] and its variants (e.g., +adaptive Fourier neural operator [22] and FourCast- +Net [23, 24]). +These frameworks not only solve +PDEs with known expressions but also be able to +predict complex dynamic systems governed by un- +known PDEs on real data sets (e.g., climate system +[23, 24]). +1.2. The limitation of previous partial differential equation +solvers +Although substantial progress has been accomplished by ex- +isting PDE solvers from various perspectives, there remain crit- +ical challenges in this booming direction. +In practice, the mesh-dependent property of classic numeri- +cal solvers has implied an inevitable trade-off between compu- +tation accuracy and efficiency, i.e., fine-grained meshes ensure +accuracy yet coarse-grained meshes are favorable for efficiency +[19, 25]. However, in many cases, the applications of PDE solv- +ing (e.g., numerical weather forecasting [26, 27]) require timely +and accurate computation. To ensure accuracy and speed, ev- +ery single time of computation in the downstream applications +supported by classic numerical solvers frequently costs large +amounts of computing resources. In cases with limited com- +puting power, a significant time delay may occur. Moreover, all +numerical solvers require the explicit definitions of target PDEs +as a priori knowledge and are less applicable to predict real data +generated by unknown PDEs [19]. +As for neural-network-based solvers, challenges still arise +from multiple perspectives, even though these solvers have out- +performed the classic numerical ones in prediction efficiency +significantly. Type (a) solvers, as we have suggested, are mesh- +dependent and lack generalization capacities across different +mesh designs [5]. Type (b) solvers are limited to learning a con- +crete instance of the PDE rather than the entire family and, con- +sequently, require restarted training given a different instance +and can not handle the data with unknown PDEs [5]. Although +type (c) solvers can learn the entire PDE family in a mesh- +independent manner [19, 5], they may face challenges in char- +acterizing the long-term behaviour of equation solution γ (·). To +understand these challenges, let us consider the iterative update +strategy of neural operators [5] +�γ (xt+ε) = σ +� +W�γ (xt) + +� +D×{t} +κθ (xt, yt, φ (xt) , φ (yt))�γ (yt) dyt +� +, +∀ xt ∈ D × {t}, +(4) +in which ε ∈ (0, ∞) denotes time difference, notion σ : R → R +is an arbitrary element-wise non-linear activation function, no- +tion W : Rd�γ → Rd�γ stands for a linear layer, function κθ : +R2(d+dφ) → Rd�γ is a neural network parameterized by θ, and +2 + +mapping �γ : D × T → Rd�γ denotes the parameterized coun- +terpart of equation solution γ generated by the neural network +(e.g., by embedding) [5]. In Eq. (4), the integral term associ- +ated with κθ defines an kernel integral operator to parameterize +the Green function Jφ : (D × T) × (D × T) → R +�γ (xt+ε) = +� +D×{t} +Jφ (xt, yt) η (yt) dyt, ∀ xt ∈ D × {t}, +(5) +where the Green function is determined by φ as well. One can +see a similar form of Eq. (5) in Ref. [5]. Computationally, the +iteration of Eq. (4) can be significantly accelerated by Fourier +transform, which leads to the well-known Fourier neural oper- +ator [19]. +From a dynamic system perspective, Eq. (4) is similar to +the iterative dynamics of an infinite-dimensional non-linear dy- +namic system of equation solution γt = γ (D × {t}), where each +snapshot γ (D × {t}) is generated after function γ acts on all el- +ements in set D × {t}. Mathematically, the dynamics is defined +as +γt+ε = γt + +� t+ε +t +ζ (γτ, τ) dτ, ∀t ∈ T, +(6) +or equivalently +∂tγt = ζ (γt, t) , ∀γt ∈ Rdγ × T, +(7) +in which ζ : Rdγ × T → Rdγ denotes the associated infinite- +dimensional evolution mapping. +The challenge faced by type (c) solvers lies in that evolution +mapping ζ (·, ·) maybe even more intricate than equation solu- +tion γ (·) itself. Let us consider the cocycle property of the flow +mapping θ associated with ζ (·, ·) according to modern dynamic +system theory [28] +θt+ε +t += θt+ε +t+τ ◦ θt+τ +t +, ∀t ≤ t + τ ≤ t + ε ∈ T. +(8) +Operator ◦ denotes the composition of mappings. In general, +Eq. (8) determines how equation solution γ (·) evolves across +adjoining time intervals. In a special case where ζ (·, ·) is time- +independent, i.e., ∂tζ (·, t) ≡ 0, Eq. +(8) reduces to the au- +tonomous case +θt+ε = θε ◦ θt, ∀t, ε ∈ T. +(9) +Otherwise, +Eq. +(8) generally corresponds to the non- +autonomous case where the underlying mechanisms governing +the evolution of γ (·) vary across time. Consequently, a large +ε may correspond to a highly non-trivial evolution process of +γ (·), making �γ (xt+ε) less predictable during iterative updat- +ing and reducing the precision of Eq. (4) significantly. This +phenomenon inevitably impedes the accurate prediction of the +long-term dynamics (i.e., ε → ∞) of diverse non-linear PDE +families (e.g., see those in epidemic prevention [29], economic +modelling [30], and weather forecast [23, 24]). To overcome +this obstacle, existing models are forced to improve accuracy at +the cost of efficiency. +1.3. Our contributions to partial differential equation solvers +In this paper, we build on Koopman neural operator (KNO), +one of our latest works [31], to develop an efficient module of +PDE solving and overcome the limitation in characterizing the +long-term behaviours of complicated PDE families. As a study +on computational physics programs, our research has the fol- +lowing contributions compared with our previous work [31]. +First, we generalize the original KNO to four kinds of vari- +ants. Beyond the original KNO, these differentiated variants of- +fer more possibilities for data-specific and task-oriented solver +designs. Specifically, the compact variants of KNO realized by +multi-layer perceptrons and convolutional neural networks can +accurately solve PDE with small model sizes. The large vari- +ants of KNO implemented on visual transformers can predict +highly intricate dynamic systems governed by unknown, high- +dimensional, and non-linear PDEs (e.g., climate system). +Second, we propose KoopmanLab, a PyTorch module of +Koopman neural operator family, as a self-contained and user- +friendly platform for PDE solving. All necessary tools, such as +those for data loading, model construction, parameter manipu- +lation, output visualization, and performance quantification, are +offered in a user-friendly manner to support customized appli- +cations. +Third, we offer comprehensive validation of the proposed +module on representative data sets, including those gener- +ated by important PDEs in fluid mechanics (e.g., the Navier- +Stokes equation and the Bateman–Burgers equation) or ob- +tained by global meteorological recording research (e.g., atmo- +spheric, land, and oceanic climate fields in ERA5 data set) [32]. +By measuring accuracy, quantifying efficiency, and comparing +all KNO variants with other state-of-the-art alternatives (e.g., +Fourier neural operator [19] and FourCastNet [23, 24]), we sug- +gest the potential of our module to serve as an ideal choice of +PDE solving and dynamic system prediction. +2. The initial version of Koopman neural operator +Although the original Koopman neural operator has been +proposed in our earlier work [31], here we elaborate on its +mechanisms for completeness. We further present more math- +ematical details that are not covered in Ref. [31] to analyze the +convergence of the original Koopman neural operator. +2.1. The original Koopman neural operator: Objective +Koopman neural operator (KNO) is proposed to deal with the +non-linear, and potentially non-autonomous, dynamic system in +Eqs. (6-7). The idea underlying KNO arises from the pursuit to +transform the non-linear system in Eqs. (6-7) to a sufficiently +simple linear one +∂tg (γt) = Ag (γt) , ∀t ∈ T, +(10) +where g (·) is an appropriate transform and A is a linear op- +erator. +In modern dynamic system theory [28], this pursuit +may be achieved if we can develop an approach to character- +ize the Koopman operator K, an infinite-dimensional linear op- +erator governing all possible observations of the dynamic sys- +tem of equation solution γ (·), to act on the flow mapping θ and +3 + +linearizing the dynamics of γ (·) in an appropriate observation +space. This idea has been extensively applied in plasma physics +[33], fluid dynamics [34], robot kinetics [35], and neuroscience +[36]. +Mathematically, we need to find a set of observation func- +tions (or named as measurement functions) [28] +G +� +Rdγ × T +� += {g|g : Rdγ × T → Cdγ} +(11) +such that a family of Koopman operators can be identified for +the autonomous (i.e., Kε : G +� +Rdγ × T +� +→ G +� +Rdγ × T +� +) or +the non-autonomous (i.e., Kt+ε +t +: G +� +Rdγ × T +� +→ G +� +Rdγ × T +� +) +case. These Koopman operators can function on the observa- +tions of γ (·) to update them +Kεg (γt) = g (θε (γt)) = g (γt+ε) , ∀t × T, +(12) +Kt+ε +t +g (γt) = g +� +θt+ε +t +(γt) +� += g (γt+ε) , ∀t ≤ t + ε ∈ T, +(13) +where Eqs. (12-13) correspond to the autonomous and non- +autonomous cases, respectively. The updating is implemented +in a linear manner, which can be illustrated by taking the non- +autonomous case as an example +∂tg (γt) = lim +ε→0 +Kt+ε +t +g (γt) − g (γt) +ε +. +(14) +Apart from the linear system of g (γt) in Eq. (14), one may +also consider the Lie operator (i.e., the Lie derivative of g (·) +along the vector field γ (·)), which is generator operator of such +a Koopman operator [37, 38, 39] +Ltg = lim +t+ε→t +Kt+ε +t +g (γt) − g (γt) +t + ε − t +. +(15) +Eq. (15) defines a linear system of g (γt) as well +∂tg (γt) = lim +t+ε→t +Kt+ε +t +g (γt) − g (γt) +ε += Ltg (γt) , +(16) +which can also be considered in the application. +To understand the linearization of g (γt) by the Koopman op- +erator from the perspective of PDE solving, let us consider the +Lax pair (M, N) of an integrable version of Eqs. (1-3) [40] +M = Dn +x + αγ (xt) I, α ∈ C, +(17) +Mψ (xt) = λψ (xt) , λ ∈ C, +(18) +∂tψ (xt) = Nψ (xt) , +(19) +where Dn +x denotes the n-th total derivative operator and I is an +identity operator. Eq. (18) denotes an eigenvalue problem at +moment t. A relation between linear operators M and N can be +identified if we calculate the time derivative of Eq. (18) +(∂tM + MN − NM) ψ (xt) = ∂tλψ (xt) , +(20) +which directly leads to +∂tM + [M, N] = 0, +(21) +where [M, N] = MN − NM denotes the commutator of op- +erators. Combining Eqs. (17-21) with Eq. (16), we can readily +see the close relation between N and Kt+ε +t +ψ (D × {t}) = g (γt) ⇒ N = lim +t+ε→t +Kt+ε +t +g (γt) − g (γt) +ε +, +(22) +which holds in the autonomous case as well. In sum, the lin- +earization of g (γt) is intrinsically related to the Lax pair and +the inverse scattering transform of integrable PDEs [40]. Note +that similar ideas have been comprehensively explored in math- +ematics and physics [41, 42, 43, 44]. +Once we find a way to derive the Koopman operator, we can +reformulate Eq. (4) as +�γt+ε = g−1 � +Kt+ε +t +g ��γt +�� +, ∀t ∈ T, +(23) +where �γt = �γ (D × {t}). Certainly, an infinite-dimensional lin- +ear operator is not operable in practice. To enable neural net- +works to learn a potential Koopman operator, we need to con- +sider � +K ∈ Rr×r, a finite matrix, as a counterpart of K that +acts on K = span +�� +G +� +, a finite invariant sub-space spanned by +� +G = {g1, . . . , gr} ⊂ G +� +Rdγ × T +� +� +Kgi = ⟨[ν1, . . . , νr] , �g1, . . . , gr +�⟩, ∀gi ∈ � +G, +(24) +where [ν1, . . . , νr] ∈ Rr and ⟨·, ·⟩ denotes the inner product. +Mathematically, any finite set of eigenfunctions of the Koop- +man operator K can span a finite invariant sub-space. +2.2. The original Koopman neural operator: Mathematics +There exist numerous previous works that pursue to char- +acterize the Koopman operator by machine-learning-based ap- +proaches. Some approaches are highly practical but limited to +the autonomous case (e.g., the case in Eq. (12)) [45, 46, 47, 48]. +Other approaches are more general in application but critically +depend on a priori knowledge about the eigenvalue spectrum +(e.g, the numbers of real and complex eigenvalues) of Koopman +operator to deal with the continuous spectrum problem [49]. +In practice, a balance should be reached between mathe- +matical completeness and computational practicality. An ideal +Koopman-operator-based PDE solver should fit in with both au- +tonomous and non-autonomous cases and limit the dependence +of a priori knowledge as much as possible (even though these +restraints inevitably reduce mathematical completeness). To ex- +plore such a balance, we introduce the Koopman neural opera- +tor (KNO), a flexible approach, in our previous work [31]. +The formalization of KNO begins with the Krylov sequence +[50] of the observable defined by a unit time step ε ∈ [0, ∞], +which is used in the Krylov subspace approach for comput- +ing the eigenvalues of large matrices [50]. +One can see its +application in Koopman-operator-related algorithms such as +the Hankel-DMD [51], sHankel-DMD [52], and HAVOK [53]. +Specifically, the Krylov sequence is given as +Rn = �g (γ0) , g (γε) , g (γ2ε) , . . . , g (γnε)� , +(25) +4 + +which is generated by K and g (γ0) +Rn = +� +g (γ0) , Kε +0g (γ0) , K2ε +ε Kε +0g (γ0) , . . . , Knε +(n−1)ε · · · Kε +0g (γ0) +� +. +(26) +Computationally, the Krylov sequence can be sampled by a +Hankel matrix of observations +Hm×n = +������������ +g (γ0) +g (γε) +· · · +g (γnε) +... +... +... +... +g �γ(m−1)ε +� +g (γmε) +· · · +g �γ(m+n−1)ε +� +������������ +, +(27) +where m ∈ N+ denotes the dimension of delay-embedding. In +Eq. (27), each column is a sampled result that approximates a +function in the Krylov subspace. +If the Koopman operator has a discrete spectrum (e.g., has +eigenvalues), there exists an invariant subspace K of the Koop- +man operator, which can be spanned by the Krylov subspace +K = span (Rn) ≃ span �H(m,n) +� +(28) +as long as n ≥ dim (K) − 1 (here dim (·) denotes the dimension- +ality). This property suggests the possibility of approximating +the actual Koopman operator to K by � +Kt+ε +t +: G +� +Rdγ × T +� +→ K, +a finite Koopman operator restricted to K for any t ∈ T. Math- +ematically, matrix � +K is required to satisfy the Galerkin projec- +tion relation +⟨ � +Kt+ε +t +h (γt) , g (γiε)⟩ = ⟨Kt+ε +t +h (γt) , g (γiε)⟩, ∀i = 0, . . . , m, +(29) +where h (·) ∈ G +� +Rdγ × T +� +is an arbitrary function [54, 55]. If +the target Koopman operator is bounded and H(m,n) spans its +invariant subspace, the approximation can be realized by +lim +m→∞ +� +G(Rdγ ×T) +∥ � +Kt+ε +t +h (γt) − Kt+ε +t +h (γt) ∥Fdµ = 0, +∀h (·) ∈ G +� +Rdγ × T +� +, +(30) +where µ is a measure on G +� +Rdγ × T +� +and ∥ · ∥F denotes the +Frobenius norm. Once a restricted Koopman operator is de- +rived, we can obtain the following iterative dynamics +Hm×n (k + 1) = � +K(k+1)ε +kε +Hm×n (k) , ∀k = 1, . . . , n, +(31) +in which Hm×n (k) is the k-th column of Hm×n. +As for the case where the Koopman operator has a continuous +spectrum (e.g., has no eigenvalue), there is no finite invariant +subspace to support computational approximation. Such an ill- +posed situation remains for future exploration. +The restricted Koopman operator � +K can be learned efficiently +if it corresponds to autonomous system, i.e., � +Kt+ε +t += � +Kε. How- +ever, an online optimization will be necessary if it corresponds +to non-autonomous system, i.e., � +Kt+ε +t +is time-varying. Lim- +ited by computing resources or data size, expensive online op- +timization may not always be available during PDE solving. +Therefore, we propose a compromised approach to realize off- +line training under the ergodic assumption [51, 56] of the dy- +namic system of γt, i.e., γt ultimately visits every possible sys- +tem states as t → ∞. Under this assumption, the proportion of +retention time of γt at a certain system state is equivalent to the +probability of this state in the space, making the time-averaging +equivalent to the actual expectation at the limit of infinite time. +Based on this property, we can define an expectation of the re- +stricted Koopman operator associated with ε +Kε = lim +t→∞ +1 +t +� +[0,t) +g (γτ)−1 g (γτ+ε) dτ, +(32) +≃ argmin +P∈R +n +� +k=1 +∥Hm×n (k + 1) − PHm×n (k) ∥F. +(33) +For a fixed time difference ε, the expected Koopman operator +Kε : G +� +Rdγ × T +� +→ K is a time-average of � +Kt+ε +t +that can be +learned during offline optimization. +2.3. The original Koopman neural operator: Convergence +Given an ideal setting of m → ∞, we can ensure the conver- +gence of the eigenvalues and eigenfunctions of � +K to those of +K under the assumption of ergodic property. Similar conclu- +sions can be seen in Ref. [54]. To understand this convergence, +we need to indicate several important properties. First, as we +have mentioned, there exists an equivalence relation between +time-averaging and the real expectation as the time approaches +to infinity (i.e., the Birkhoff ergodic theorem [51, 56]) +lim +m→∞ +1 +m +m−1 +� +i=0 +g (γi) = +� +K +gdµ. +(34) +Second, Eq. (34) directly implies that +lim +m→∞ +1 +m⟨Hm×n (i) , Hm×n ( j)⟩ = +� +K +Hm×n (i) �Hm×n ( j)�∗ dµ. +(35) += ⟨Hm×n (i) , Hm×n ( j)⟩K, +(36) +where ∗ denotes the complex conjugate and ⟨·, ·⟩K stands for the +inner product of functions in K. Given the learned Koopman +operator Kε, Eq. (36) coincides with the definition of the actual +Gramian matrix V associated with the inner product space K +Vi,j = ⟨K +i−1 +ε Rn, K +j−1 +ε +Rn⟩K, +(37) += ⟨Hm×n (i) , Hm×n ( j)⟩K, +(38) +where we mark K +i−1 +ε Rn = +� +K +i−1 +ε g (γ0) , . . . , K +i−1 +ε g (γnε) +� +for +convenience. Eq. (38) is derived from the fact that Hm×n serves +as the sampling of Rn. Meanwhile, the left side of Eq. (35) co- +incides with the empirical Gramian matrix � +V associated with +matrix Hm×n +� +Vi,j = 1 +m⟨Hm×n (i) , Hm×n ( j)⟩. +(39) +5 + +Our formal proof can be developed based on these two prop- +erties. Let us consider the first r < n element of Rn +Rr = +� +g (γ0) , Kεg (γ0) , K +2 +εg (γ0) , . . . , K +r−1 +ε g (γ0) +� +, +(40) +which defines a possible basis of K. +Theoretically, the learned Koopman operator restricted to K +can be represented by a companion matrix +C = +���������������������� +0 +0 +· · · +0 +c0 +1 +0 +· · · +0 +c1 +0 +1 +· · · +0 +c2 +... +... +... +... +... +0 +0 +· · · +1 +cr−1 +���������������������� +. +(41) +The last column of C denotes the coordinate of K +r +εg (γ0) defined +by the basis, which should be calculated as +C (r) = V−1 +��������������������������� +⟨g (γ0) , K +r +εg (γ0)⟩K +⟨Kεg (γ0) , K +r +εg (γ0)⟩K +⟨K +2 +εg (γ0) , K +r +εg (γ0)⟩K +... +⟨K +r−1 +ε g (γ0) , K +r +εg (γ0)⟩K +��������������������������� +. +(42) +Empirically, the learned Koopman operator restricted to +span �H(m,n) +� can be also represented by a companion matrix, +whose last column can be calculated as +� +C (r) = 1 +m +� +V−1 +��������������������������� +⟨Hm×n (1) , K +r +εHm×n (1)⟩ +⟨KεHm×n (1) , K +r +εHm×n (1)⟩ +⟨K +2 +εHm×n (1) , K +r +εHm×n (1)⟩ +... +⟨K +r−1 +ε g (γ0) , K +r +εg (γ0)⟩K +��������������������������� +. +(43) +It is trivial to prove +lim +m→∞ +� +V−1 = +� +lim +m→∞ +� +V +�−1 += V−1 +(44) +applying Eq. (36) and Eqs. (38-39). Similarly, we can know +lim +m→∞ +1 +m +��������������������������� +⟨Hm×n (1) , K +r +εHm×n (1)⟩ +⟨KεHm×n (1) , K +r +εHm×n (1)⟩ +⟨K +2 +εHm×n (1) , K +r +εHm×n (1)⟩ +... +⟨K +r−1 +ε g (γ0) , K +r +εg (γ0)⟩K +��������������������������� += +��������������������������� +⟨g (γ0) , K +r +εg (γ0)⟩K +⟨Kεg (γ0) , K +r +εg (γ0)⟩K +⟨K +2 +εg (γ0) , K +r +εg (γ0)⟩K +... +⟨K +r−1 +ε g (γ0) , K +r +εg (γ0)⟩K +��������������������������� +. +(45) +Therefore, we can derive +lim +m→∞ +� +Cij = Cij, ∀i, j ∈ {1, . . . , r}2 +(46) +based on Eqs. (44-45), implying that +lim +m→∞ +� +M∈PMk +�� +C +� M = +� +M∈PMk(C) +M, ∀k ∈ {1, . . . , r}. +(47) +In Eq. (47), notion PMk (·) denotes the set of all the k-order +principal minors of the corresponding matrix. Now, let us con- +sider the characteristic polynomials of � +C and C +P� +C (z) = zr + +r� +i=1 +(−1)i +������������ +� +M∈PMi +�� +C +� M +������������ +zr−i, +(48) +PC (z) = zr + +r� +i=1 +(−1)i +�������� +� +M∈PMi(C) +M +�������� zr−i, +(49) +whose distance at the limit of m → ∞ can be measured as +lim +m→∞ ∥P� +C (z) − PC (z) ∥ += lim +m→∞ max +1≤i≤r +������ (−1)i +������������ +� +M∈PMi +�� +C +� M − +� +M∈PMi(C) +M +������������ +������, +(50) +=0. +(51) +Because the roots of a given polynomial evolve continuously +as the function of the coefficients, we know that � +C and C share +the same set of eigenvalues at the limit of m → ∞ since their +characteristic polynomials converge to the same. Moreover, the +convergence of � +C to C and the convergence of the eigenvalues +� +C to those of C eventually imply the convergence of the eigen- +functions. +2.4. The original Koopman neural operator: Computation +In Ref. [31], we have proposed an architecture to implement +the original Koopman neural operator on neural networks. The +details of architecture designs are presented below: +• Part 1: Observation. An encoder (e.g., a single non- +linear layer with tanh (·) activation function in the original +Koopman neural operator) serves as observation function +g (·) to transform φt = φ (D × {t}), an arbitrary input of +the PDE family (e.g., φt can be directly set as γt), into +g ��γt +� ∈ G +� +Rd�γ × T +� +g ��γt +� = Encoder (φt) , ∀t ∈ T. +(52) +Please see Fig. 1 for illustrations. +• Part 2: Fourier transform. Similar to the Fourier neural +operator [19], the original Koopman neural operator also +utilizes the Fourier transform during the iterative update of +the Green function in Eq. (5). Given g ��γt +�, we derive the +Fourier transform gF (·) = F ◦ g (·), where we truncate the +Fourier series at ω, a maximum frequency +gF (ξ) = χ[0,ω] (ξ) +� +D×{t} +g ��γ (xt)� exp (−2πi⟨xt, ξ⟩) dxt. +(53) +Note that χ· (·) denotes the indicator function (i.e., χA (a) = +1 if a ∈ A, otherwise χA (a) = 0). Computationally, the +6 + +Figure 1: +Conceptual illustrations of neural network architectures of the original Koopman neural operator. (a) summarizes key mathematical transform in each +part, where r is the prediction length. (b) visualizes a prediction instance on the 2-dimensional Navier-Stokes equation. +above transform is implemented by fast Fourier transform. +For convenience, we mark +gF +��γt +� = F ◦ g ��γt +� = {gF (ξ) |ξ ∈ [0, ∞)} +(54) +as the transformed result of �γt. Different from Ref. [19], +our main motivation for using the truncated Fourier trans- +form is to extract the low-frequency information (i.e., +main system components) of the represented equation so- +lution g ��γt +�. +Certainly, frequency truncation inevitably +causes the loss of high-frequency information (i.e., high- +frequency perturbations or edges). In the original Koop- +man neural operator, Part 5 is designed to complement +the lost information associated with high-frequency. See +Fig. 1 for more details. +• Part 3: Hankel representation and offline Koopman +operator. Once we have derived gF +��γt +� for every t ∈ εN+, +a Hankel matrix � +Hm×n of gF +��γt +� will be generated follow- +ing m ∈ N, a dimension of delay-embedding (note that +n ∈ N is the number of all accessible samples) +� +Hm×n = +������������ +gF +��γ0 +� +gF +��γε +� +· · · +gF +��γnε +� +... +... +... +... +gF +��γ(m−1)ε +� +gF +��γmε +� +· · · +gF +��γ(m+n−1)ε +� +������������ +. +(55) +We train a o × o linear layer to learn a neural network rep- +resentation of Koopman operator Kε : G +� +Rd�γ × T +� +→ �K +following Eqs. (32-33), where �K is spanned by � +Hm×n. +The learned Kε can be used to predict the future state of +gF +��γ(m+n−1)ε +� as +gF +��γ(m+n+r−1)ε +� = +� +K +r +ε � +Hm×n (n) +�T (m) , r ∈ N+ +(56) +where notion T denotes the transpose of a matrix. Please +see Fig. 1 for illustrations. +• Part 4: Inverse Fourier transform. After gF +��γ(m+n)ε +� +is predicted in Part 3, it is transformed from the Fourier +space to G +� +Rd�γ × T +� +by an inverse Fourier transform +g ��γ (xt)� = +1 +(2π)d�γ +� ∞ +−∞ +gF (ξ) exp (2πi⟨xt, ξ⟩) dξ, +(57) +where t = (m + n + r − 1) ε. For convenience, we mark +g ��γ(m+n+r−1)ε +� = F −1 ◦ gF +��γ(m+n+r−1)ε +� . +(58) +Please see Fig. 1 for instances of Part 4. +• Part 5: High-frequency information complement. In +the original Koopman neural operator, we use a convo- +lutional layer to extract high-frequency of g ��γt +� because +convolutional layers can amplify high-frequency compo- +nents according to Fourier analysis [57]. Therefore, we +train a convolutional layer C on the outputs of Part 1 to ex- +tract their high-frequency information. As a complement +of Parts 2-4, the convolutional layer realizes a forward +prediction of high-frequency information +� +gC +� +�γ(j+r−1)ε +� +, . . . , gC +� +�γ(j+m+r−1)ε +��T +=C +� +g +� +�γ jε +� +, . . . , g +� +�γ( j+m)ε +��T , ∀j = 1, . . . , n. +(59) +See Fig. 1 for illustrations. +• Part 6: Inverse observation. +Once two future states, +g ��γ(m+n)ε +� and gC +��γ(m+n)ε +�, are predicted by Parts 2-4 and +Part 5, they are unified in a linear manner +gU +��γ(m+n+r−1)ε +� = g ��γ(m+n+r−1)ε +� + gC +��γ(m+n+r−1)ε +� . +(60) +Given gU +��γ(m+n)ε +�, a non-linear decoder (e.g., a single +non-linear layer with tanh (·) activation function in the +original neural operator) is trained to approximate the in- +verse of observation function +g−1 (·) ≃ g−1 +U (·) +(61) +and derive +�γ(m+n+r−1)ε = Decoder �gU +��γ(m+n+r−1)ε +�� +(62) +as the target state of equation solution in space Rd�γ. Please +see Fig. 1 for illustrations. +7 + +(a) + Ground truth + Ground truth +(b) +Part 2 +Part 3 +Part 1 +Part 2 +Part 3 +Part 4 +Part 6 +Part 1 +Part 4 +Part 6 +9F(V) +gF(Ine) +F· +i ((3(I-u+u)s)6...(3(t-u)s)6) +(j +m)a ++m+r)c +je +G+18 +. +()6 +Truncate* +[Hmxn(n)] (m) +noε +noD +Nt +t+r? +Part 5 +g-1() +Part 5 +[ge(V(j+r-1)e) ... 9e(V(j+m+r-1)8 +()6 +C[g()je) ... g(V(i+m))]Parts 1-6 define the iterative update strategy of Eq. (23). For +any t′ > t ∈ εN, the iterative dynamics is given as +�γt′ += +� +g−1� +F −1 ◦ K +t′−t +ε +◦ F ◦ g ��γ[t−mε,t] +� +�������������������������������������������������������������������� +Parts 1-4 ++ C ◦ g ��γ[t−mε,t] +� +���������������������������� +Part 1 and part 5 +��T (m) , +(63) +in which�γ[t−mε,t] denotes a vector ��γt−mε, . . . ,�γt +�. The loss func- +tion for optimizing Eq. (63) is defined as +L = λp∥�γt′ − γt′∥F + λr +m +� +i=0 +∥g−1 ◦ g ��γt−imε +� − γt−imε∥F, +(64) +where λp, λr ∈ (0, ∞) denotes the weights of prediction and +reconstruction processes in the loss function. +The one-unit architecture of the original Koopman neural op- +erator is visualized in Fig. 1. A multi-unit architecture can be +readily constructed by cascading the copy of Parts 2-5 multiple +times. +3. The Koopman neural operator family +Beyond the original Koopman neural operator (KNO) [31], +we generalize it to four kinds of variants to fit in with different +application demands. +3.1. The compact KNO sub-family: Definition +The compact KNO sub-family includes two compact vari- +ants of KNO realized by multi-layer perceptrons (MLP-KNO) +and convolutional neural networks (CNN-KNO). These vari- +ants are proposed to accurately solve PDEs with small model +sizes. Specifically, they are designed following Eqs. (52-62), +where encoder and decoder are defined as +Encoder = η ◦ E, E ∈ {We, Ce}, +(65) +Decoder = η ◦ D, D ∈ {Wd, Cd}. +(66) +In Eqs. +(65-66), mapping η denotes a non-linear activation +function (e.g., we use tanh (·) in our research), notions We and +Wd are two weight matrices of the corresponding sizes, and Ce +and Cd are two convolutional layers. +Our proposed KoopmanLab module offers user-friendly tools +to customize an MLP-KNO. A customized instance is presented +below. +1 import +koopmanlab +as kp +2 +3 MLP_KNO1D = model.koopman(backbone = "KNO1d", +autoencoder = "MLP", o = o, f = f, r = r, +device = device) +4 +5 MLP_KNO2D = model.koopman(backbone = "KNO2d", +autoencoder = "MLP", o = o, f = f, r = r, +device = device) +6 +7 MLP_KNO1D.compile () +8 MLP_KNO2D.compile () +9 +10 ## Parameter +definitions : +11 # +o: the +dimension +of the +learned +Koopman +operator +12 # +f: the +number of +frequency +modes +below +frequency +truncation +threshold +13 # +r: the +power of the +Koopman +operator +in EQ. +(56) +14 # +device: if CPU or GPU is used +for +computation +Similarly, a CNN-KNO can be customized using the follow- +ing code, where we present 1-dimensional and 2-dimensional +visions. +1 import +koopmanlab +as kp +2 +3 CNN_KNO_1D = model.koopman(backbone = "KNO1d", +autoencoder = "Conv1d", o = o, f = f, r = r, +device = device) +4 +5 CNN_KNO_1D .compile () +6 +7 CNN_KNO_2D = model.koopman(backbone = "KNO2d", +autoencoder = "Conv2d", o = o, f = f, r = r, +device = device) +8 +9 CNN_KNO_1D .compile () +10 CNN_KNO_2D .compile () +11 +12 ## Parameter +definitions : +13 # +o: the +dimension +of the +learned +Koopman +operator +14 # +f: the +number of +frequency +modes +below +frequency +truncation +threshold +15 # +r: the +power of the +Koopman +operator +in EQ. +(56) +16 # +device: if CPU or GPU is used +for +computation +3.2. The compact KNO sub-family: Validation +To validate the proposed compact KNO sub-family in PDE +solving tasks, we design mesh-independent and long-term pre- +diction tasks on representative PDEs (e.g., the 2-dimensional +Navier-Stokes equation [58] and the 1-dimensional Bate- +man–Burgers equation [59]). The numerical data sets of these +two are provided by Ref. [19]. +Specifically, +the incompressible 2-dimensional Navier- +Stokes equation has a vorticity form +∂tγ (xt) + χ (xt) ∇γ (xt) = ν∆γ (xt) + ψ (xt) , +xt ∈ (0, 1)2 × (0, ∞) , +(67) +∇χ (xt) = 0, xt ∈ (0, 1)2 × (0, ∞) , +(68) +γ (x0) = γI, x0 ∈ (0, 1) × {0}, +(69) +in which γ (·) denotes the vorticity, χ (·) measures the velocity, +ψ (·) is a time-independent forcing term. The viscosity coeffi- +cient is ν ∈ {10−3, 10−4} in our research. Given the data with +the highest mesh resolution, one can further generate the data +with the lower resolution by direct down-sampling [19]. The +data with the highest mesh resolution has 213 grids [19]. Our +KoopmanLab module offers a function to load the data of the +incompressible 2-dimensional Navier-Stokes equation. +1 import +koopmanlab +as kp +2 +3 train_loader , test_loader = kp.data. navier_stokes +(path , batch_size = 10, T_in = 10, T_out = +40, type = "1e-3", sub = 1) +8 + +Figure 2: +Experimental validation of the compact KNO sub-family. (a) Results of the mesh-independent experiment on the Bateman–Burgers equation. (b) +Results of the long-term prediction experiment on the 2-dimensional Navier-Stokes equation with a viscosity coefficient ν = 10−3. (b) Results of the same long-term +prediction experiment on the 2-dimensional Navier-Stokes equation with ν = 10−4. (d) The prediction results and errors (RMSE) on the 2-dimensional Navier-Stokes +equation with ν = 10−3. (e) The prediction results and errors (RMSE) on the 2-dimensional Navier-Stokes equation with ν = 10−4. +4 +5 ## Parameter +definitions : +6 # +path: the +file +path of the +downloaded +data +set +7 # +T_in: the +duration +length +of input +data +8 # +T_out: the +duration +length +required +to +predict +9 # +Type: the +viscosity +coefficient +10 # +sub: the down -sampling +scaling +factor. For +instance , a scaling +factor +sub =2 acting on a +2- dimensional +data +with +the +spatial +resolution +64*64 +will +create a down -sampled +space of 32*32. +The +same +factor +action +on a +1- dimensional +data +with +the +spatial +9 + +10-4 +(b) +(a) +KNO(0=32,f=10,r=12,MLP) +KNO(0=128,f=10,r=16,Conv) +0.35 +KNO(0=32,f=10,r=12,MLP) +0.00175 +KNO(0=32,f=16,r=8,MLP) +KNO(0=32,f=16,r=8,MLP) +KNO(o=32,f=10,r=24,Conv) +KNO(0=48,f=10,r=12,MLP) +KNO(0=48,f=10,r=12,MLP) +KNO(o=8,f=10,r=10,Conv) +0.30 +KNO(0=128,f=10,r=16,MLP) +0.00150 +KNO(o=32,f=10,r=12,Conv) +KNO(o=32,f=10,r=12,Conv) +KNO(0=32,f=10,r=24,MLP) +KNO(o=32,f=16,r=8,Conv) +KNO(o=32,f=16,r=8,Conv) +0.25 +KNO(0=8,f=10,r=10,MLP) +0.00125 +KNO(o=48,f=10,r=12,Conv) +KNO(o=48,f=10,r=12,Conv) +FNO(Original, 1 unit) +FNO(Original, 1 unit) +RMSE +E +E 0.201 +00.00100 +MSI +RM +P 0.15 +0.00075 +0.10 +10-5. +0.00050 +0.05 +0.00025 +0.00 +0.00000 +2829 210 +6E E E E E 67 Z EZ Z 61 T ST T 1 6 +12 13 14 15 16 17 18 19 20 +211 +Resolution +Time Step +Time Step +(d) +Initial Condition +CNN-KNO +MLP-KNO +FNO +0.10 +0.05 +0.00 +-0.05 +-0.10 +CNN-KNO (Error) +MLP-KNO (Error) +FNO (Error) +(e) +2 +Initial Condition +CNN-KNO +MLP-KNO +FNO +1.0 +0.5 +0.0 +-0.5 +-1.0 +FNO (Error) +CNN-KNO (Error) +MLP-KNO (Error)Models +Settings +Parameter number +FNO +default settings, one-unit [19] +233897 +MLP-KNO +(o, f, r) = (32, 10, 12) +206538 +MLP-KNO +(o, f, r) = (32, 16, 8) +526026 +MLP-KNO +(o, f, r) = (48, 10, 12) +464170 +CNN-KNO +(o, f, r) = (32, 10, 12) +206538 +CNN-KNO +(o, f, r) = (32, 16, 8) +526026 +CNN-KNO +(o, f, r) = (48, 10, 12) +464170 +Table 1: The parameter numbers of all implemented models in Figs. 2(b-c), which are counted by the tool provided by Ref. [23, 24]. +resolution +1*64 +implies a down -sampled +space +of 1*32 +The 1-dimensional Bateman–Burgers equation is defined as +∂tγ (xt) + ∂x +�γ2 (xt) +2 +� += ν∂xxγ (xt) , xt ∈ (0, 1) × (0, 1] , +xt ∈ (0, 1) × (0, 1] , +(70) +γ (x0) = γI, x0 ∈ (0, 1) × {0}, +(71) +in which γI stands for a periodic initial condition γI +∈ +L2 +periodic [(0, 1) ; R] and parameter ν ∈ (0, ∞) is the viscosity co- +efficient. We set ν = 100 in our research. The data with highest +mesh resolution has 216 grids [19]. To load this data set, one +can use the following function. +1 import +koopmanlab +as kp +2 +3 train_loader , test_loader = kp.data.burgers(path , +batch_size = 64, sub = 32) +4 +5 ## Parameter +definitions : +6 # +path: the +file +path of the +downloaded +data +set +7 # +sub: the down -sampling +scaling +factor. For +instance , a scaling +factor +sub=2 +acting on a +2- dimensional +data +with +the +spatial +resolution +64*64 +will +create a down -sampled +space of 32*32. +The +same +factor +action on a +1- dimensional +data +with +the +spatial +resolution +1*64 +implies a down -sampled +space +of 1*32 +In Fig. +2(a), we validate the mesh-independent property +of the proposed compact KNO sub-family adopting the same +setting used in our earlier work [31]. The mesh-independent +property, as suggested by Refs. [16, 17, 18, 5, 19, 20, 21], +arises from the fact that the neural operator is expected to learn +the solution operator of an entire PDE family rather than be +limited to a concrete parameterized instance. Specifically, we +conduct the experiments on the data of 1-dimensional Bate- +man–Burgers equation associated with different mesh granu- +larity conditions (i.e., spatial resolution of meshes). Different +versions of the compact KNO sub-family are defined by chang- +ing hyper-parameters (e.g., operator size o, frequency mode +number f, and the power of the Koopman operator r = +t′−t +ε +in Eq. (56)). These models are trained by 1000 randomly se- +lected samples with the lowest spatial resolution and conduct +1-second forward prediction on 200 samples associated with +different resolutions. Batch size is set as 64, the learning rate +is initialized as 0.001 and halved every 100 epochs, and the +weights of prediction and reconstruction in Eq. (64)) are set as +� +λp, λr +� += (5, 0.5). As shown in Fig. 2(a), the prediction errors +of all versions of the compact KNO sub-family maintain con- +stantly across different spatial resolutions, suggesting the ca- +pacity of the compact KNO sub-family to be mesh-independent. +Mesh-independence is important for PDE solving because it al- +lows one to train a neural-network-based PDE solver on the +data with low spatial resolution and directly apply the solver on +the data with high spatial resolution, which breaks the trade-off +of accuracy and efficiency in PDE solving. In our earlier work +[31], one can further see a detailed comparison between the +original KNO and FNO [19] in mesh-independent prediction +task, where the original KNO outperforms FNO with a much +smaller model size (e.g., a size of 5×103 for KNO and a size of +2 × 107 for FNO). Other neural operator models, such as graph +neural operator (GNO) [5] and multipole graph neural opera- +tor (MGNO) [60], are no longer considered because they have +been demonstrated as less accurate than FNO as reported by +Ref. [19]. +In Fig. 2(b-e), we validate the compact KNO sub-family +by a long-term prediction task designed on the 2-dimensional +Navier-Stokes equation data sets with viscosity coefficients ν = +10−3 (Fig. 2(b)) and ν = 10−4 (Fig. 2(c)). A down-sampling +scaling factor of 2 is defined to generate the data sets with 212 +grids. For comparison, a one-unit FNO is defined following the +default setting introduced in Ref. [19]. A 40-time-interval pre- +diction task is conducted on the data set with ν = 10−3, where +models are trained on 1000 samples of γ +� +(0, 1)2 × [0, 10) +� +and +tested on 200 samples of γ +� +(0, 1)2 × (10, 50] +� +. +Similarly, a +more challenging 10-time-interval prediction task is conducted +on the data set with ν = 10−4, in which models are trained on +8000 samples of γ +� +(0, 1)2 × [0, 10) +� +and tested on 200 samples +of γ +� +(0, 1)2 × (10, 20] +� +. Fig. 2(b-c) report the prediction per- +formance of all models as the function of increasing prediction +duration length. Fig. 2(d-e) visualize predicted instances and +errors in the cases with ν = 10−3 (Fig. 2(d)) and ν = 10−4 (Fig. +2(e)). All experiment results suggest the optimal potential of +the compact KNO sub-family in characterizing the long-term +evolution of PDE solutions. Combining these results with the +model sizes measured in Table. 1, we suggest that the compact +KNO sub-family realizes a better balance between accuracy and +efficiency because a KNO variant with a smaller model size can +still outperform FNO significantly. +3.3. The ViT-KNO sub-family: Definition +Different from the compact KNO sub-family, the ViT-KNO +sub-family is proposed for dealing with more intricate situa- +10 + +Figure 3: Conceptual illustrations of neural network architectures of the ViT-KNO sub-family. (a) summarizes the computational design of each part. (b) illustrates +an instance of the parallel design (i.e., multi-head design) of the ViT-KNO sub-family, where the depth of each head is 1. +tions (here ViT stands for Vision Transformer [61]). Numerous +applications of PDE solving (e.g., global climate forecasting) +require the solver to be able to capture the underlying patterns +of ultra-large data sets that may be related with certain unknown +PDEs. Meanwhile, there may exist multiple variables of inter- +est that govern by a group of coupled PDEs. To fit in with these +situations, we follow the main idea of the compact KNO sub- +family to develop a kind of transfer-based PDE solver. The +mechanism underlying the proposed ViT-KNO sub-family is +not completely same as Eqs. (52-62) because some mathemati- +cal details are modified to improve model applicability on noisy +real data sets. We suggest the benefits of our modifications +based on an experiment on ERA5, one of the largest data set of +global atmospheric, land, and oceanic climate fields [32]. Nev- +ertheless, more in-depth mathematical analyses of these modi- +fications remain for future studies. +Let us consider a case where there exist v coupled variables, +{γ1, . . . , γh}, defined on domain D × T. The dynamics of these +variables are govern by a group of PDEs with unknown expres- +sions. The objective is to learn the equation solutions of these +latent PDEs such that the dynamics of {γ1, . . . , γh} can be accu- +rately characterized. +The architecture of ViT-KNO sub-family consists of 7 parts. +Below, we present a detailed computational implementation of +each part. +• Part 1: Observation. Similar to the encoder design in +the compact KNO sub-family, an encoder component is +implemented in ViT-KNO to serve as observation function +g (·) and transform φs +t = φs (D × {t}) into g ��γs +t +� for each +s ∈ {1, . . . , h}. Specifically, the encoder is realized by the +token embedding layer in Vision Transformer (ViT) [61]. +Given a joint input +� +φ1 +t , . . . , φh +t +� +∈ Rdφ×h, we first transform +it into a 3-dimensional token tensor Φt by a convolutional +layer Ce +Ce +�� +φ1 +t , . . . , φh +t +�� += g +��Γt +� +∈ Ru×v×l, +(72) +where domain D is reorganized into u × v patches (i.e., to- +kens). The patch is a kind of square and non-overlapping +macro-mesh. If domain D has been already discretized +into multiple meshes, then the size of a patch equals the +number of meshes it covers. Parameter l denotes a cus- +tomized embedding dimension, which is not necessarily +the same as h. The derived tensor g +��Γt +� +denotes the joint +representation of +� +g +� +�γ1 +t +� +, . . . , g +� +�γl +t +�� +. Please see Fig. 3 for +illustrations. +• Part 2: Fourier transform. Similar to adaptive Fourier +neural operator [23, 24, 62], a truncated Fourier transform +is applied on the first two dimensions of g +��Γt +� +to derive +the Fourier series of each embedded variable s ∈ {1, . . . , l} +gs +F (ξ) = χ[0,ω] (ξ) +� +[u]×[v]×{t} +g ��γs (xt)� exp (−2πi⟨xt, ξ⟩) dxt, +(73) +where [u] = {1, . . . , u}. For convenience, we mark +gs +F +��γt +� = F ◦ g ��γs +t +� = +� +gs +F (ξ) |ξ ∈ [0, ∞) +� +, +(74) +gF +��Γt +� += F ◦ g +��Γt +� += +� +g1 +F +��γt +� , . . . , gl +F +��γt +�� +, +(75) +in which s ∈ {1, . . . , l}. +Similar to the compact KNO +sub-family, frequency truncation leads to the loss of high- +frequency information. In the ViT-KNO sub-family, Part +5 is designed for complementing high-frequency informa- +tion. See Fig. 3 for details. +• Part 3: Koopman-operator-associated component. Af- +ter deriving gF +��Γt +� +for every t ∈ εN+, a Koopman- +operator-associated component is designed to function on +the third dimension of every token in gF +��Γt +� +and realize +the iterative dynamics +gF +��Γt+ε +� += KεSgF +��Γt +� += +� +g1 +F +��γt+ε +� , . . . , gl +F +��γt+ε +�� +, +(76) +S = η ◦ W, +(77) +in which Koopman operator Kε is learned by a linear +transform, layer S is constructed by a non-linear activa- +tion function η acting on a linear layer W. Although S is +not a part of the original Koopman neural operator [31], in- +cluding it can efficiently enhance the capacity of this com- +ponent to characterize intricate large-scale data. In Koop- +manLab, the leaky rectified linear unit (Leaky ReLU) [63] +11 + +(a) +Ground truth of multiple variables +(b) +Ground truth of multiple variables +Head 1 +Part 1 +Part 2 +Part 3 +Part 4 +Part 6 +Part 7 +Part 1 +Part 7 +Part 2 +Part 3 +Part 4 +甲 Part 6 +gF(ft+e) +Part 5 +Variable +Token decoding + Token decoding +F-1()- +Token encoding +F(O) +Token +K,SgF(ft+e) +Head 2 +Partition for +田 +l encoding +Part 6 +coupling +parallel design +Part 5 +Part 5 +Head 3 +gc(ft+) = Cg(ft+) +Part 2-* +Part 3 - +Part 4 +田 Part 6 +M +g-1() +g-1() +g() +g() + Part 5is suggested as a default choice of S, which can also re- +duce to the ReLU function as a special case. Please see +Fig. 3 for illustrations of Part 3. +• Part 4: Inverse Fourier transform. Once gF +��Γt+ε +� +is de- +rived in Part 3, the inverse Fourier transform is applied on +the first two dimensions of gF +��Γt+ε +� +to transform gF +��Γt+ε +� +back to the observation space +g ��γs (xt+ε)� = +1 +(2π)d�γ +� ∞ +−∞ +gs +F (ξ) exp (2πi⟨xt+ε, ξ⟩) dξ, +(78) +g +��Γt+ε +� += F −1 ◦ gF +��Γt+ε +� += +� +g +� +�γ1 (xt+ε) +� +, . . . , g +� +�γl (xt+ε) +�� +. +(79) +Please see instances in Fig. 3. +• Part 5: +High-frequency information complement. +Same as the compact KNO sub-family, there is a com- +ponent for complementing high-frequency information in +ViT-KNO. This component is also realized by a convolu- +tional layer C that acts on the outputs of Part 1 to learn +the dynamics of high-frequency information +gC +��Γt+ε +� += Cg +��Γt +� +. +(80) +See Fig. 1 for illustrations. +• Part 6: Variable coupling. Given two predicted states, +g +��Γt+ε +� +and gC +��Γt+ε +� +, by Parts 2-4 and Part 5, we com- +bine them in a linear form +gU +��Γt+ε +� += g +��Γt+ε +� ++ gC +��Γt+ε +� +. +(81) +Because ViT-KNO is designed to learn multi-variate sys- +tems governed by unknown coupled PDEs, we need to +characterize the coupling relation among variables. Be- +cause we lack the a priori knowledge about these underly- +ing PDEs, we suggest to capture the coupling relation by +optimizing a non-linear layer M +gM +��Γt+ε +� += MgU +��Γt+ε +� +. +(82) +Following the idea of adaptive Fourier neural operator [23, +24, 62], we use the Gaussian Error Linear Unit (GELU) as +the activation function in this non-linear layer. Please see +Fig. 3 for illustrations. +• Part 7: Inverse observation. Given gM +��Γt+ε +� +, a decoder +is implemented to function as the inverse of observation +function +� +�γ1 +t+ε, . . . ,�γh +t+ε +� +≃ g−1 � +gM +��Γt+ε +�� += Decoder +� +gM +��Γt+ε +�� +. +(83) +Similar to the compact KNO sub-family, there are two +kinds of decoders included in our proposed KoopmanLab +module +Decoder ∈ {Wd, Cd}, +(84) +in which Wd and Cd denote linear and convolutional lay- +ers, respectively. These two kinds of decoder designs dis- +tinguish between two variants of the ViT-KNO sub-family. +See Fig. 3 for illustrations. +Parts 1-7 define the iterative update strategy of Eq. (23) in +a multi-variate case. For any t ∈ T, the iterative dynamics is +given as +�Γt+ε = g−1 ◦ M ◦ +� +F −1 ◦ KεS ◦ F ◦ g +��Γt +� ++ C ◦ g +��Γt +� � +. +(85) +Multi-step prediction can be realized in an iterative manner. +The loss function for optimizing Eq. (85) is +L = λp +h +� +s=1 +∥�γs +t+ε − γs +t+ε∥F + λr +h +� +s=1 +∥g−1 ◦ g ��γs +t +� − γs +t ∥F, +(86) +where λp, λr ∈ (0, ∞) are the weights of prediction and recon- +struction. +Several computational tricks can be considered in the appli- +cation. First, a LASSO regularization [64] can be included to +improve the robustness and sparsity of Koopman operator Kε +in Eq. (76). This trick has been demonstrated as effective in +adaptive Fourier neural operator [23, 24, 62] and is applicable +to the ViT-KNO sub-family as well. Second, the transformer ar- +chitecture supports a parallel design the ViT-KNO sub-family. +Specifically, the third dimension of the output of Part 1 can be +subdivided into multiple parts (e.g., g +��Γt +� +∈ Ru×v×l is subdi- +vided into k parts such that each part is an element in Ru×v× l +k ). +Then, Parts 2-6 are copied k × j times, where each group of j +copies is organized into a sequence. Each sequence of j copies +is referred to as a head in the transformer, processes a corre- +sponding 1 +k part of g +��Γt +� +∈ Ru×v×l, and shares parameters dur- +ing optimization (see Fig. 3). Computationally, parameters k +and j are referred to as the number and the depth of heads. The +processed outputs of these k parallel heads are unified by Part 7 +to derive the final prediction result. In our proposed Koopman- +Lab, these two tricks are included to improve computational +efficiency. +Our KoopmanLab module supports customizing ViT-KNO +frameworks. Below, we present an instance of ViT-KNO with a +multi-layer perceptron as the decoder +1 import +koopmanlab +as kp +2 +3 ViT_KNO = model. koopman_vit (decoder = "MLP", +resolution =(1440 , +720) , patch_size =(2, 2), +in_chans =20, +out_chans =20, +head_num =20, +embed_dim =768 , +depth = 16, parallel = True , +high_freq = True , device=device) +4 +5 ViT_KNO.compile () +6 +7 ## Parameter +definitions : +8 # +resolution: the +spatial +resolution +of input +data +9 # +patch_size: the +size of each +patch (i.e., +token) +10 # +in_chans: the +number of target +variables +in +the +data +set +12 + +Figure 4: Visualization of three instances of the experiment results on ERA5 data set. (a-c) respectively show the time-dependent predicted results of global wind +speed, U10, and V10 variables on selected moments, accompanied by corresponding ground truths. Please note that wind speed is not originally included in the 20 +climate variables selected from ERA5. It is calculated as wind speed = +� +U102 + V102. +11 # +out_chans: the +number +of +predicted +variables +by ViT -KNO , which is +usually +same as +in_chans +12 # +head_num: the +number of heads +13 # +embed_dim: the +embeding +dimension +denoted +by l +in Eq. (72) +14 # +depth: the +depth of each +head +15 # +parallel: if +parallel +design is +applied +16 # +high_freq: if high -frequency +information +complement +is +applied +17 # +device: if CPU or GPU is used +for +computation +Similarly, a ViT-KNO whose decoder is a convolutional layer +can be defined as the following +13 + +True wend speed +True wend speed +True wend speed +Initial condition +36 hours later +72 hours later +(a) +90°N +N.06 +N.06 +60°N+ +N.09 +N.09 + 28 +30°N +N.0m +30°N +0° +0° +0。 +14 +30°S +30°S- +S.03 +S.09 +60°S +S.09 +S.06 +120°W +i80° +120°W +60°W +0°w +60°E +120°E +180° +0°W +120°W +0°w +120°E +180° +180° +60°W +60°E +120°E +60°W +60°E +Predicted wend speed +Predicted wend speed +36 hours later +72 hours later +N.06 +N.06 +60°N +60°N +N.0m +N.0m +0° +30°S +S.05 +S.09 +S.09 +90°180 +90°80° +180° +120°W +0°w +120°E +120°W +0°w +180° +60°W +60°E +60°E +120°E +60°W +72 hours later +True U10 +True U10 +36 hours later +True U10 +Initial condition +(b) +N.06 +N.06 +N.06 +N.09 +60°N +N.09 +12.5 +30°N +N.O0E +30°N +0° +。0 +0° + 0.0 +30°S +30°S +30°S +12.5 +S.09 +S.09 +S.09 +-25.0 +90°180 +90°180. +60°W +120°W +0°W +60°E +120°E +0°W +120°E +60° +o°w +180° +120°W +60°W +60°E +180° +120°W +60°E +120°E +180° +Predicted U10 +72 hours later +Predicted U10 +36 hours later +N.06 +N.06 +N.09 +60°N +N.Om +N.0m +0° +0° +S.0m +30°S +S.09 +S.09 +90°180 +0°w +120°W +o°W +60°E +180° +120°W +60°W +60°E +180° +60°W +120°E +120°E +True V10 +36 hours later +True V10 +72 hours later +True V10 +Initial condition +(c) +N。06 +N.06 +N.06 +60°N +N.09 +N.09 +25.0 +30°N +N.0 +30°N +12.5 +。0 +0° +S.08 +30°S +30°S +S.09 +60°S +S.09 +90°80 +-25.0 +.05.06 +0°W +120°W +60°W +120°W +120°W +0°W +60°E +0°W +60°E +120°E +180° +60°W +120°E +180° +60°W +60°E +120°E +180° +Predicted V10 +72 hours later +36 hours later +Predicted V10 +N.06 +L N.06 +60°N +N.09 +30°N +N。0m +0° +0° +S.0 +30°S- +S.09 +60°S +90°180° +120°W +o°w +180° +120°W +60°W +0°w +60°E +120°E +60°W +60°E +120°E +180°Figure 5: Time-dependent prediction accuracy (measured by anomaly correlation coefficient, ACC) of ViT-KNO on 20 variables of ERA5 data set. The colored +area denotes the interval of accuracy whose boundaries are fractiles. The dashed line denotes the average accuracy. +1 import +koopmanlab +as kp +2 +3 ViT_KNO = model. koopman_vit (decoder = "Conv2d", +resolution =(1440 , +720) , patch_size =(2, 2), +in_chans =20, +out_chans =20, +head_num =20, +embed_dim =768 , +depth = 16, parallel = True , +high_freq = True , device=device) +4 +5 ViT_KNO.compile () +6 +7 ## Parameter +definitions : +8 # +resolution: the +spatial +resolution +of input +data +9 # +patch_size: the +size of each +patch (i.e., +token) +10 # +in_chans: the +number of +target +variables +in +the +data +set +11 # +out_chans: the +number +of +predicted +variables +by ViT -KNO , which is +usually +same as +in_chans +12 # +head_num: the +number of heads +13 # +embed_dim: the +embeding +dimension +denoted +by l +in Eq. (72) +14 # +depth: the +depth of each +head +15 # +parallel: if +parallel +design is +applied +16 # +high_freq: if high -frequency +information +complement +is +applied +17 # +device: if CPU or GPU is used +for +computation +Please note that there exist some detailed model parameters +that are not covered by the above codes because they are highly +coupled during computation or less important in our theoretical +derivations. Users are suggested to adjust them after loading +the source code of ViT-KNO. +3.4. The ViT-KNO sub-family: Validation +To validate the proposed ViT-KNO sub-family, we imple- +ment a large-scale experiment on ERA5, one of the largest high- +resolution data sets of global-scale multi-variate climate fields +[32]. This data set has been extensively applied in weather fore- +casting tasks (e.g., see FourCastNet [23, 24]), ensuring the re- +producibility and comparability of our results. +Twenty important climate variables are considered in our re- +search, including mean large-scale precipitation (MSLP), rel- +ative humidity with 500 hPa (R500), relative humidity with +850 hPa (R850), surface pressure (SP), 2m temperature (T2M), +temperature with 500 hPa (T500), temperature with 850 hPa +(T850), total column water vapour (TCWV), the 10m u- +component of wind (U10), the u-component of wind with 500 +hPa (U500), the u-component of wind with 850 hPa (U850), +14 + +(b) +(a) +(d) +1.0 +(c) +1.0 +1.07 +0.8 +0.8 +0.8 +0.8 +0.4 +0.4 +0.4 - +0.4 +IIIIIIII +0.2 +0.2 +0.2 +0.2 +R500 +MSLP +R850 +SP +0.0叶 +0.0时 +0.0+ +0.0 +48 +48 +96 +144 +192 +48 +96 +144 +192 +96 +144 +192 +48 +96 +144 +192 +Prediction duration length +Prediction duration length +Prediction duration length +Prediction duration length +(e) 1 +(f) +(g) +(h) +1.0 + +0.8 +0.8 +0.8 +0.8 +IIIIIJ +0.4 +0.4 +0.4 +0.4 +0.2 +0.2 +0.2 +0.2 +TCW +T2M +T500 +T850 +0.0% +0.0叶 +48 +96 +144 +192 +96 +144 +192 +48 +96 +144 +192 +48 +96 +144 +192 +Prediction duration length +Prediction duration length +Prediction duration length +Prediction duration length +(k) 1.0 +0.8 +0.8 +0.8 +0.8 +IIIII +0.4 +0.4 +0.4 +0.4 +0.2 +0.2 +0.2 : +0.2 +U10 +U500 +U850 +U1000 +o.0% +400 +0.0 +0.0 +192 +48 +96 +144 +192 +48 +96 +144 +192 +48 +96 +144 +192 +48 +96 +144 +Prediction duration length +Prediction duration length +Prediction duration length +Prediction duration length +(m) 1.0 t +(n) 1.0 + +0.8 +0.8 +0.8 +0.8 +IIIIIII +0.4 +0.4 +0.4 +0.2 +0.2 +0.2 +0.2 +V1000 +U10 +V500 +V850 +0.0g +%00 +0.0 + +48 +48 +96 +144 +192 +96 +144 +192 +48 +96 +144 +192 +48 +96 +144 +192 +Prediction duration length +Prediction duration length +Prediction duration length +Prediction duration length +(q) +(r) +(s) +1.0- +0.8 +0.8 +0.8 +0.8 +0.4 +0.4 +0.4 +0.4 +0.2 +0.2 +0.2 +0.2 +Z50 +Z500 +Z850 +Z1000 +0.0 +0.0+ +o.0叶 +48 +144 +48 +96 +144 +48 +96 +144 +48 +96 +144 +192 +96 +192 +192 +192 +Prediction duration length +Prediction duration length +Prediction duration length +Prediction duration lengththe u-component of wind with 1000 hPa (U1000), the 10m v- +component of wind (V10), the v-component of wind with 500 +hPa (V500), the v-component of wind with 850 hPa (V850), the +v-component of wind with 1000 hPa (V1000), the geopotential +with 50 hPa (Z50), the geopotential with 500 hPa (Z500), the +geopotential with 850 hPa (Z850), and the geopotential with +1000 hPa (Z1000). +We test a ViT-KNO whose decoder is a multi-layer percep- +tron in a long-term prediction task. Given the samples of initial +conditions, the ViT-KNO is required to predict the future states +of all 20 climate variables after t ∈ {6, 12, 18, . . . , 192} hours. +The training data set includes the samples recorded from 1979 +to 2015. The validation data set includes the samples recorded +during 2016 and 2017. The test data set includes the samples +recorded in 2018. The spatial resolution of all samples is set +as 1440 × 720. All data is pre-processed in a standard manner +following Refs. [32, 23, 24], where a Z-transform is applied +to normalize all variables. The training of our ViT-KNO is im- +plemented in a multi-GPU environment with 128 × 16 GBs in +total. The actual memory cost of training is 1250.56 GBs. The +testing of trained ViT-KNO is implemented in a single 16-GB +GPU environment. +Key model settings are summarized below. The batch size is +set as 128, the learning rate is 5 × 10−4 and updated by a cosine +annealing approach, the patch size is set as 8 × 8, the number of +heads is 8, the depth of heads is 12, the embedded dimension +l in Eq. (72) is 768, the relative weights of prediction and re- +construction in the loss function are λp = 0.9 and λr = 0.1, and +the number of kept low-frequency modes after the fast Fourier +transform is 32. The defined model has 74691840 parameters +in total. All 20 climate variables are learned and predicted to- +gether rather than respectively. Please note that the information +of land-form is not provided to the model. The model is re- +quired to learn climate variables with no additional information. +The maximum number of available epochs for training is set as +300 to explore when the model can converge, which costs about +92.5 hours in our environment. The convergence is observed to +emerge after ≃ 150 epochs. Therefore, users can consider a +150-epoch training in the application, which costs about 2 days +under the same hardware condition. There is no additional trick +applied during training. +In Fig. 4, we visualize several instances of the predicted cli- +mate fields during testing, accompanied by corresponding true +values. High consistency can be seen between these ground +truths and their predicted counterparts derived by ViT-KNO. +Quantitatively, the prediction accuracy of each climate variable +during testing is measured by anomaly correlation coefficient +(ACC) in Fig. 5. According to the same prediction task re- +ported by Refs. [23, 24], the trained ViT-KNO outperforms the +baseline state-of-the-art deep learning models for weather fore- +casting proposed by Ref. [65] significantly. Compared with +the FourCastNet trained with multiple numerical tricks (e.g., +multi-stage training with large memory consumption) [23, 24], +ViT-KNO achieves a similar accuracy during testing. Limited +by computing resources, we are unable to precisely compare +between ViT-KNO and FourCastNet under the same hardware +and training condition yet (the FourCastNet is reported to be +trained on 3808 NVIDIA A100 GPUs with numerous compu- +tational optimization [23]). More detailed comparisons may be +considered in future studies. We suggest that ViT-KNO has +the potential to become a competitive alternative of FourCast- +Net. Moreover, the time cost of a single time of prediction by +ViT-KNO is observed as ≃ 0.02768695354 seconds in a sin- +gle 16-GB GPU. Compared with the classic numerical weather +forecasting systems (e.g., the ECMWF Integrated Forecasting +System) whose prediction inevitably requires a multi-GPU en- +vironment (e.g., more than 1000 NVIDIA Selene nodes where +each node consists of 8 NVIDIA A100 GPUs) [66, 67, 68], ViT- +KNO is orders of magnitude faster in the application (e.g., the +Integrated Forecasting System L91 18 km model is expected to +cost about 9840 node seconds for prediction on a NVIDIA Se- +lene node [68]). Therefore, our ViT-KNO has the potential to +serve as a unit in ensemble weather forecasting frameworks to +realize an efficient prediction of global weather. +4. Conclusion +In this paper, we have presented KoopmanLab, an efficient +module of Koopman neural operator family for solving par- +tial differential equations. The included models in this mod- +ule, such as the compact KNO sub-family and the ViT-KNO +sub-family, are provided with mathematical foundations, com- +putational designs, and validations in solving concrete PDEs or +predicting intricate dynamic system governed by unknown cou- +pled PDEs. All models are suggested as competitive with other +state-of-the-art approaches in corresponding tasks. Compared +with classic numerical and neural-network-based PDE solvers, +the proposed KNO variants can achieve significant acceleration, +more robust mesh-independence, higher generalization capac- +ity on changed conditions, more flexibility in characterizing la- +tent PDEs with unknown forms, and a better balance between +accuracy and efficiency. Therefore, we suggest the potential of +KoopmanLab be applied in diverse down-stream tasks related +with PDE solving. Users can download this module via +1 pip +install +koopmanlab +or +1 git +clone +https :// github.com/Koopman -Laboratory/ +KoopmanLab .git +2 cd +KoopmanLab +3 pip +install +-e . +Several important questions remain for future studies. First, +one may consider more specialized computational optimization +of models in KoopmanLab (e.g., consider multi-stage training +as suggested in Refs. [23, 24] or multi-objective balancing by +Pareto theory [69, 70]). Second, one can explore a more de- +tailed comparison between the ViT-KNO sub-family and Four- +CastNet under the equivalent hardware and training conditions. +Third, one can analyze the errors of our models caused by the +potential continuous spectrum of the Koopman operator or the +absence of ergodic property in real cases. +15 + +Acknowledgements +This project is supported by the Artificial and General Intel- +ligence Research Program of Guo Qiang Research Institute at +Tsinghua University (2020GQG1017) as well as the Tsinghua +University Initiative Scientific Research Program. +References +[1] M. S. Gockenbach, Partial differential equations: analytical and numeri- +cal methods, Vol. 122, Siam, 2005. +[2] H. Tanabe, Functional analytic methods for partial differential equations, +CRC Press, 2017. +[3] L. Debnath, L. Debnath, Nonlinear partial differential equations for sci- +entists and engineers, Springer, 2005. +[4] R. M. Mattheij, S. W. Rienstra, J. T. T. Boonkkamp, Partial differential +equations: modeling, analysis, computation, SIAM, 2005. +[5] Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stu- +art, A. Anandkumar, Neural operator: Graph kernel network for partial +differential equations, arXiv preprint arXiv:2003.03485 (2020). +[6] J. N. Reddy, Introduction to the finite element method, McGraw-Hill Ed- +ucation, 2019. +[7] K. Lipnikov, G. Manzini, M. Shashkov, Mimetic finite difference method, +Journal of Computational Physics 257 (2014) 1163–1227. +[8] R. Eymard, T. Gallou¨et, R. Herbin, Finite volume methods, Handbook of +numerical analysis 7 (2000) 713–1018. +[9] X. Guo, W. Li, F. Iorio, Convolutional neural networks for steady flow +approximation, in: Proceedings of the 22nd ACM SIGKDD international +conference on knowledge discovery and data mining, 2016, pp. 481–490. +[10] Y. Zhu, N. Zabaras, Bayesian deep convolutional encoder–decoder net- +works for surrogate modeling and uncertainty quantification, Journal of +Computational Physics 366 (2018) 415–447. +[11] S. Bhatnagar, Y. Afshar, S. Pan, K. Duraisamy, S. Kaushik, Prediction of +aerodynamic flow fields using convolutional neural networks, Computa- +tional Mechanics 64 (2) (2019) 525–545. +[12] B. Yu, et al., The deep ritz method: a deep learning-based numerical algo- +rithm for solving variational problems, Communications in Mathematics +and Statistics 6 (1) (2018) 1–12. +[13] M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural net- +works: A deep learning framework for solving forward and inverse prob- +lems involving nonlinear partial differential equations, Journal of Com- +putational physics 378 (2019) 686–707. +[14] L. Bar, N. Sochen, Unsupervised deep learning algorithm for pde-based +forward and inverse problems, arXiv preprint arXiv:1904.05417 (2019). +[15] S. Pan, K. Duraisamy, Physics-informed probabilistic learning of linear +embeddings of nonlinear dynamics with guaranteed stability, SIAM Jour- +nal on Applied Dynamical Systems 19 (1) (2020) 480–509. +[16] L. Lu, P. Jin, G. E. Karniadakis, Deeponet: Learning nonlinear operators +for identifying differential equations based on the universal approxima- +tion theorem of operators, arXiv preprint arXiv:1910.03193 (2019). +[17] K. Bhattacharya, B. Hosseini, N. B. Kovachki, A. M. Stuart, Model +reduction and neural networks for parametric pdes, arXiv preprint +arXiv:2005.03180 (2020). +[18] N. H. Nelsen, A. M. Stuart, The random feature model for input-output +maps between banach spaces, SIAM Journal on Scientific Computing +43 (5) (2021) A3212–A3243. +[19] Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stu- +art, A. Anandkumar, Fourier neural operator for parametric partial differ- +ential equations, arXiv preprint arXiv:2010.08895 (2020). +[20] N. Kovachki, S. Lanthaler, S. Mishra, On universal approximation and +error bounds for fourier neural operators, Journal of Machine Learning +Research 22 (2021) Art–No. +[21] Z. Li, D. Z. Huang, B. Liu, A. Anandkumar, Fourier neural operator +with learned deformations for pdes on general geometries, arXiv preprint +arXiv:2207.05209 (2022). +[22] J. Guibas, M. Mardani, Z. Li, A. Tao, A. Anandkumar, B. Catanzaro, Effi- +cient token mixing for transformers via adaptive fourier neural operators, +in: International Conference on Learning Representations, 2021. +[23] J. Pathak, S. Subramanian, P. Harrington, S. Raja, A. Chattopadhyay, +M. Mardani, T. Kurth, D. Hall, Z. Li, K. Azizzadenesheli, et al., Fourcast- +net: A global data-driven high-resolution weather model using adaptive +fourier neural operators, arXiv preprint arXiv:2202.11214 (2022). +[24] T. Kurth, S. Subramanian, P. Harrington, J. Pathak, M. Mardani, D. Hall, +A. Miele, K. Kashinath, A. Anandkumar, Fourcastnet: Accelerating +global high-resolution weather forecasting using adaptive fourier neural +operators, arXiv preprint arXiv:2208.05419 (2022). +[25] E. Tadmor, A review of numerical methods for nonlinear partial differ- +ential equations, Bulletin of the American Mathematical Society 49 (4) +(2012) 507–554. +[26] D. J. Stensrud, Parameterization schemes: keys to understanding numeri- +cal weather prediction models, Cambridge University Press, 2009. +[27] P. Bauer, A. Thorpe, G. Brunet, The quiet revolution of numerical weather +prediction, Nature 525 (7567) (2015) 47–55. +[28] S. L. Brunton, M. Budisic, E. Kaiser, J. N. Kutz, Modern koopman theory +for dynamical systems, SIAM Review 64 (2) (2022) 229–340. +[29] D. Cao, Y. Wang, J. Duan, C. Zhang, X. Zhu, C. Huang, Y. Tong, B. Xu, +J. Bai, J. Tong, et al., Spectral temporal graph neural network for multi- +variate time-series forecasting, Advances in neural information process- +ing systems 33 (2020) 17766–17778. +[30] F. Aminian, E. D. Suarez, M. Aminian, D. T. Walz, Forecasting economic +data with neural networks, Computational Economics 28 (1) (2006) 71– +88. +[31] Anonymous, Koopman neural operator for learning non-linear partial dif- +ferential equations, in: Submitted to The Eleventh International Confer- +ence on Learning Representations, 2023, under review. +URL https://openreview.net/forum?id=kciCTrtbzVl +[32] H. Hersbach, B. Bell, P. Berrisford, S. Hirahara, A. Hor´anyi, J. Mu˜noz- +Sabater, J. Nicolas, C. Peubey, R. Radu, D. Schepers, et al., The era5 +global reanalysis, Quarterly Journal of the Royal Meteorological Society +146 (730) (2020) 1999–2049. +[33] R. Taylor, J. N. Kutz, K. Morgan, B. A. Nelson, Dynamic mode decom- +position for plasma diagnostics and validation, Review of Scientific In- +struments 89 (5) (2018) 053501. +[34] C. W. Rowley, I. Mezi´c, S. Bagheri, P. Schlatter, D. S. Henningson, Spec- +tral analysis of nonlinear flows, Journal of fluid mechanics 641 (2009) +115–127. +[35] I. Abraham, T. D. Murphey, Active learning of dynamics for data-driven +control using koopman operators, IEEE Transactions on Robotics 35 (5) +(2019) 1071–1083. +[36] B. W. Brunton, L. A. Johnson, J. G. Ojemann, J. N. Kutz, Extracting +spatial–temporal coherent patterns in large-scale neural recordings us- +ing dynamic mode decomposition, Journal of neuroscience methods 258 +(2016) 1–15. +[37] B. O. Koopman, Hamiltonian systems and transformation in hilbert space, +Proceedings of the National Academy of Sciences 17 (5) (1931) 315–318. +[38] R. Abraham, J. E. Marsden, T. Ratiu, Manifolds, tensor analysis, and +applications, Vol. 75, Springer Science & Business Media, 2012. +[39] C. Chicone, Y. Latushkin, Evolution semigroups in dynamical systems +and differential equations, no. 70, American Mathematical Soc., 1999. +[40] P. D. Lax, Integrals of nonlinear equations of evolution and solitary +waves, Communications on pure and applied mathematics 21 (5) (1968) +467–490. +[41] J. P. Parker, J. Page, Koopman analysis of isolated fronts and solitons, +SIAM Journal on Applied Dynamical Systems 19 (4) (2020) 2803–2828. +[42] H. Nakao, I. Mezi´c, Spectral analysis of the koopman operator for partial +differential equations, Chaos: An Interdisciplinary Journal of Nonlinear +Science 30 (11) (2020) 113131. +[43] C. Gin, B. Lusch, S. L. Brunton, J. N. Kutz, Deep learning models for +global coordinate transformations that linearise pdes, European Journal +of Applied Mathematics 32 (3) (2021) 515–539. +[44] J. Page, R. R. Kerswell, Koopman analysis of burgers equation, Physical +Review Fluids 3 (7) (2018) 071901. +[45] N. Takeishi, Y. Kawahara, T. Yairi, Learning koopman invariant sub- +spaces for dynamic mode decomposition, Advances in Neural Informa- +tion Processing Systems 30 (2017). +[46] O. Azencot, N. B. Erichson, V. Lin, M. Mahoney, Forecasting sequential +data using consistent koopman autoencoders, in: International Confer- +ence on Machine Learning, PMLR, 2020, pp. 475–485. +[47] S. E. Otto, C. W. Rowley, Linearly recurrent autoencoder networks for +16 + +learning dynamics, SIAM Journal on Applied Dynamical Systems 18 (1) +(2019) 558–593. +[48] D. J. Alford-Lago, C. W. Curtis, A. T. Ihler, O. Issan, Deep learning en- +hanced dynamic mode decomposition, Chaos: An Interdisciplinary Jour- +nal of Nonlinear Science 32 (3) (2022) 033116. +[49] B. Lusch, J. N. Kutz, S. L. Brunton, Deep learning for universal linear +embeddings of nonlinear dynamics, Nature communications 9 (1) (2018) +1–10. +[50] Y. Saad, Numerical methods for large eigenvalue problems: revised edi- +tion, SIAM, 2011. +[51] H. Arbabi, I. Mezic, Ergodic theory, dynamic mode decomposition, and +computation of spectral properties of the koopman operator, SIAM Jour- +nal on Applied Dynamical Systems 16 (4) (2017) 2096–2126. +[52] N. ˇCrnjari´c-ˇZic, S. Ma´ceˇsi´c, I. Mezi´c, Koopman operator spectrum for +random dynamical systems, Journal of Nonlinear Science 30 (5) (2020) +2007–2056. +[53] S. L. Brunton, B. W. Brunton, J. L. Proctor, E. Kaiser, J. N. Kutz, Chaos +as an intermittently forced linear system, Nature communications 8 (1) +(2017) 1–9. +[54] M. Korda, I. Mezi´c, On convergence of extended dynamic mode decom- +position to the koopman operator, Journal of Nonlinear Science 28 (2) +(2018) 687–710. +[55] M. Li, L. Jiang, Reduced-order modeling for koopman operators of +nonautonomous dynamic systems in multiscale media, arXiv preprint +arXiv:2204.13180 (2022). +[56] I. P. Cornfeld, S. V. Fomin, Y. G. Sinai, Ergodic theory, Vol. 245, Springer +Science & Business Media, 2012. +[57] N. Park, S. Kim, How do vision transformers work?, arXiv preprint +arXiv:2202.06709 (2022). +[58] C. Wang, Exact solutions of the steady-state navier-stokes equations, An- +nual Review of Fluid Mechanics 23 (1) (1991) 159–177. +[59] E. R. Benton, G. W. Platzman, A table of solutions of the one-dimensional +burgers equation, Quarterly of Applied Mathematics 30 (2) (1972) 195– +212. +[60] Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, A. Stuart, K. Bhat- +tacharya, A. Anandkumar, Multipole graph neural operator for parametric +partial differential equations, Advances in Neural Information Processing +Systems 33 (2020) 6755–6766. +[61] A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Un- +terthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, et al., An im- +age is worth 16x16 words: Transformers for image recognition at scale, +arXiv preprint arXiv:2010.11929 (2020). +[62] J. Guibas, M. Mardani, Z. Li, A. Tao, A. Anandkumar, B. Catanzaro, +Adaptive fourier neural operators: Efficient token mixers for transform- +ers, arXiv preprint arXiv:2111.13587 (2021). +[63] A. Radford, L. Metz, S. Chintala, Unsupervised representation learning +with deep convolutional generative adversarial networks, arXiv preprint +arXiv:1511.06434 (2015). +[64] R. Tibshirani, Regression shrinkage and selection via the lasso, Journal +of the Royal Statistical Society: Series B (Methodological) 58 (1) (1996) +267–288. +[65] J. A. Weyn, D. R. Durran, R. Caruana, N. Cresswell-Clay, Sub-seasonal +forecasting with a large ensemble of deep-learning weather prediction +models, Journal of Advances in Modeling Earth Systems 13 (7) (2021) +e2021MS002502. +[66] C. D. Roberts, R. Senan, F. Molteni, S. Boussetta, M. Mayer, S. P. Keeley, +Climate model configurations of the ecmwf integrated forecasting system +(ecmwf-ifs cycle 43r1) for highresmip, Geoscientific model development +11 (9) (2018) 3681–3712. +[67] P. Lopez, A lightning parameterization for the ecmwf integrated forecast- +ing system, Monthly Weather Review 144 (9) (2016) 3057–3075. +[68] P. Bauer, T. Quintino, N. Wedi, A. Bonanni, M. Chrust, W. Deconinck, +M. Diamantakis, P. D¨uben, S. English, J. Flemming, et al., The ecmwf +scalability programme: Progress and plans, European Centre for Medium +Range Weather Forecasts, 2020. +[69] O. Sener, V. Koltun, Multi-task learning as multi-objective optimization, +Advances in neural information processing systems 31 (2018). +[70] M. Momma, C. Dong, J. Liu, A multi-objective/multi-task learning +framework induced by pareto stationarity, in: International Conference +on Machine Learning, PMLR, 2022, pp. 15895–15907. +17 + diff --git a/ZNAzT4oBgHgl3EQfKvv0/content/tmp_files/load_file.txt b/ZNAzT4oBgHgl3EQfKvv0/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..079662040deaac6d289506e75dc0f52ada61d985 --- /dev/null +++ b/ZNAzT4oBgHgl3EQfKvv0/content/tmp_files/load_file.txt @@ -0,0 +1,1362 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf,len=1361 +page_content='KoopmanLab: A PyTorch module of Koopman neural operator family for solving partial differential equations Wei Xionga,c, Muyuan Maa, Pei Sunb,∗, Yang Tianb,c,∗ aDepartment of Earth System Science, Tsinghua University, Beijing, 100084, China.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' bDepartment of Psychology & Tsinghua Laboratory of Brain and Intelligence, Tsinghua University, Beijing, 100084, China.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' cLaboratory of Advanced Computing and Storage, Central Research Institute, 2012 Laboratories, Huawei Technologies Co.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Ltd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Beijing, 100084, China.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Abstract Given the increasingly intricate forms of partial differential equations (PDEs) in physics and related fields, computationally solving PDEs without analytic solutions inevitably suffers from the trade-off between accuracy and efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Recent advances in neural operators, a kind of mesh-independent neural-network-based PDE solvers, have suggested the dawn of overcoming this challenge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In this emerging direction, Koopman neural operator (KNO) is a representative demonstration and outperforms other state-of-the- art alternatives in terms of accuracy and efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Here we present KoopmanLab, a self-contained and user-friendly PyTorch module of the Koopman neural operator family for solving partial differential equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Beyond the original version of KNO, we develop multiple new variants of KNO based on different neural network architectures to improve the general applicability of our module.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These variants are validated by mesh-independent and long-term prediction experiments implemented on representative PDEs (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the Navier-Stokes equation and the Bateman–Burgers equation) and ERA5 (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', one of the largest high-resolution data sets of global-scale climate fields).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These demonstrations suggest the potential of KoopmanLab to be considered in diverse applications of partial differential equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Keywords: Partial differential equation;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Koopman neural operator;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Dynamic system PROGRAM SUMMARY Program Title: KoopmanLab CPC Library link to program files: (to be added by Technical Editor) Developer’s repository link: https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='com/ Koopman-Laboratory/KoopmanLab Code Ocean capsule: (to be added by Technical Editor) Licensing provisions: GNU General Public License 3 (GPL) Programming language: Python, PyTorch Nature of problem: Computationally solving partial differential equa- tions (PDEs) by classic frameworks (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', finite element method) faces the trade-off between accuracy and efficiency because these numerical systems are mesh-dependent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Fine mesh granularity ensures high accuracy but inevitably implies high computation complexity and time cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Although different neural-network-based PDE solvers have been proposed to accelerate the processes of solving PDEs and maintain solution accuracy, most pioneer frameworks are not mesh-independent or limited to solving simple equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Solution method: The proposed KoopmanLab provides multiple variants of Koopman neural operator (KNO), a mesh-independent neural-network-based PDE solver with optimal performance in across- mesh-granularity and long-term prediction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The compact variants of KNO (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', implemented on multi-layer perceptrons and convolutional neural networks) can achieve high accuracy PDE solving with small model sizes while the large variants of KNO (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', implemented ∗Corresponding author.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' E-mail address: tiany20@mails.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='tsinghua.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='cn & tyanyang04@gmail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='com (Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='T);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' peisun@tsinghua.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='cn (P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='S) on visual transformers) are more competitive in predicting highly complicated dynamic systems govern by unknown, high-dimensional, and non-linear PDEs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Compared with classic numerical solvers (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', finite element method), the prediction time costs of all variants of KNO are reduced by multiple orders of magnitude.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Introduction 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The rising of partial differential equation solvers Solving partial differential equations (PDEs) essentially re- quires characterizing an appropriate solution operator F that relates Φ = Φ � D;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Rdφ� , a Banach space of inputs (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', initial values), with Γ = Γ � D;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Rdγ� , a Banach space of solutions (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', target values), for a typically time-dependent PDE family de- fined on a bounded open set D ⊂ Rd ∂tγ (xt) = � Lφγ � (xt) + κ (xt) , xt ∈ D × T, (1) γ (xt) = γB, xt ∈ ∂D × T, (2) γ (x0) = γI, x0 ∈ D × {0}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (3) In Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (1-3), set T = [0, ∞) denotes the time domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Notion Lφ is a differential operator characterized by φ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mapping κ (·) is a function that lives in a function space determined by Lφ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mapping γ (·) is the solution of the PDE family that we attempt to obtain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The boundary and initial conditions are denoted by Preprint submitted to Computer Physics Communications January 4, 2023 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='01104v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='LG] 3 Jan 2023 γB and γI, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mathematically, driving an accurate solution operator F : (φ, γB, γI) �→ γ is the key step to obtain the PDE solution γ (·).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' However, even in the case where the boundary and initial conditions are constant (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the solution operator F : (φ, γB, γI) �→ γ reduces to F : φ �→ γ), driving an analytic expression of solution operator F can be highly non- trivial [1, 2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The absence of analytic solutions of various important PDEs in science and engineering naturally calls for the rapid develop- ment of computational solvers, which attempt to approximate a parametric counterpart Fθ ≃ F parameterized by θ to derive so- lution γ (·) [1, 3, 4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To date, the joint efforts of physics, math- ematics, and computer science have given birth to two main- stream families of PDE solvers [5]: (1) The first family of solvers are classic numerical ones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Typical instances of these solvers include finite element (FEM) [6], finite difference (FDM) [7], and finite volume (FVM) [8] methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In general, these methods discretize space and time domains following specific mesh designs and solve parameterized PDEs on meshes by certain itera- tive algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, FEM subdivides the original domain into a set of sub-domains defined by a collection of element equations and recombines these element equa- tions to derive the global solution [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' FDM approximates derivatives as finite differences measured on local values [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' FVM transforms the original problem into a series of surface flux calculations on local volumes [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (2) The second family of solvers are neural-network-based ones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' With a pursuit of accelerating PDE solving and improving the applicability on real data, three kinds of neural-network-based solvers have been proposed: (a) One kind of solvers discretize domains D and T into x and y meshes and approximate a finite-dimensional and mesh-dependent solution operator Fθ by a pa- rameterized neural network between finite Euclidean spaces, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Fθ : Rx × Ry × Θ → Rx × Ry (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', see Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [9, 10, 11]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given an arbitrary input γ (xt), the trained neural network can function as a solution operator to predict γ (xt+τ) = Fθ (γ (xt)) for a certain time difference τ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (b) Another kind of solvers directly parameterize equa- tion solution γ (·) as a neural network, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Fθ : D × T × Θ → R (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', see Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [12, 13, 14, 15]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These solvers are mesh-independent and accurate in learning a given PDE because they can directly trans- form arbitrary domain and parameter setting to target equation solution γ (·).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (c) The last kind of solvers, including neural opera- tors, attempt to parameterize a mesh-dependent and infinite-dimensional solution operator with neural networks, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e, Fθ : Φ × Θ → Γ (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', see Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [16, 17, 18, 5, 19, 20, 21]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These mesh-independent solvers can be flexibly implemented on different dis- cretization schemes and only need to be trained once for a given PDE family.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The equation solution γ (·) of different instances of the PDE family can be gen- erated by a computationally reusable forward pass of the network [5, 19], which can be further ac- celerated by fast Fourier transform [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Repre- sentative demonstrations of this kind of solver are Fourier neural operator [19] and its variants (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', adaptive Fourier neural operator [22] and FourCast- Net [23, 24]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These frameworks not only solve PDEs with known expressions but also be able to predict complex dynamic systems governed by un- known PDEs on real data sets (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', climate system [23, 24]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The limitation of previous partial differential equation solvers Although substantial progress has been accomplished by ex- isting PDE solvers from various perspectives, there remain crit- ical challenges in this booming direction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In practice, the mesh-dependent property of classic numeri- cal solvers has implied an inevitable trade-off between compu- tation accuracy and efficiency, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', fine-grained meshes ensure accuracy yet coarse-grained meshes are favorable for efficiency [19, 25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' However, in many cases, the applications of PDE solv- ing (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', numerical weather forecasting [26, 27]) require timely and accurate computation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To ensure accuracy and speed, ev- ery single time of computation in the downstream applications supported by classic numerical solvers frequently costs large amounts of computing resources.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In cases with limited com- puting power, a significant time delay may occur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Moreover, all numerical solvers require the explicit definitions of target PDEs as a priori knowledge and are less applicable to predict real data generated by unknown PDEs [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' As for neural-network-based solvers, challenges still arise from multiple perspectives, even though these solvers have out- performed the classic numerical ones in prediction efficiency significantly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Type (a) solvers, as we have suggested, are mesh- dependent and lack generalization capacities across different mesh designs [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Type (b) solvers are limited to learning a con- crete instance of the PDE rather than the entire family and, con- sequently, require restarted training given a different instance and can not handle the data with unknown PDEs [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Although type (c) solvers can learn the entire PDE family in a mesh- independent manner [19, 5], they may face challenges in char- acterizing the long-term behaviour of equation solution γ (·).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To understand these challenges,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' let us consider the iterative update strategy of neural operators [5] �γ (xt+ε) = σ � W�γ (xt) + � D×{t} κθ (xt,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' yt,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' φ (xt) ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' φ (yt))�γ (yt) dyt � ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ∀ xt ∈ D × {t},' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (4) in which ε ∈ (0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ∞) denotes time difference,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' notion σ : R → R is an arbitrary element-wise non-linear activation function,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' no- tion W : Rd�γ → Rd�γ stands for a linear layer,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' function κθ : R2(d+dφ) → Rd�γ is a neural network parameterized by θ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' and 2 mapping �γ : D × T → Rd�γ denotes the parameterized coun- terpart of equation solution γ generated by the neural network (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', by embedding) [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (4), the integral term associ- ated with κθ defines an kernel integral operator to parameterize the Green function Jφ : (D × T) × (D × T) → R �γ (xt+ε) = � D×{t} Jφ (xt, yt) η (yt) dyt, ∀ xt ∈ D × {t}, (5) where the Green function is determined by φ as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' One can see a similar form of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (5) in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Computationally, the iteration of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (4) can be significantly accelerated by Fourier transform, which leads to the well-known Fourier neural oper- ator [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' From a dynamic system perspective, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (4) is similar to the iterative dynamics of an infinite-dimensional non-linear dy- namic system of equation solution γt = γ (D × {t}), where each snapshot γ (D × {t}) is generated after function γ acts on all el- ements in set D × {t}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mathematically, the dynamics is defined as γt+ε = γt + � t+ε t ζ (γτ, τ) dτ, ∀t ∈ T, (6) or equivalently ∂tγt = ζ (γt, t) , ∀γt ∈ Rdγ × T, (7) in which ζ : Rdγ × T → Rdγ denotes the associated infinite- dimensional evolution mapping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The challenge faced by type (c) solvers lies in that evolution mapping ζ (·, ·) maybe even more intricate than equation solu- tion γ (·) itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Let us consider the cocycle property of the flow mapping θ associated with ζ (·, ·) according to modern dynamic system theory [28] θt+ε t = θt+ε t+τ ◦ θt+τ t , ∀t ≤ t + τ ≤ t + ε ∈ T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (8) Operator ◦ denotes the composition of mappings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In general, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (8) determines how equation solution γ (·) evolves across adjoining time intervals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In a special case where ζ (·, ·) is time- independent, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', ∂tζ (·, t) ≡ 0, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (8) reduces to the au- tonomous case θt+ε = θε ◦ θt, ∀t, ε ∈ T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (9) Otherwise, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (8) generally corresponds to the non- autonomous case where the underlying mechanisms governing the evolution of γ (·) vary across time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Consequently, a large ε may correspond to a highly non-trivial evolution process of γ (·), making �γ (xt+ε) less predictable during iterative updat- ing and reducing the precision of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (4) significantly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' This phenomenon inevitably impedes the accurate prediction of the long-term dynamics (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', ε → ∞) of diverse non-linear PDE families (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', see those in epidemic prevention [29], economic modelling [30], and weather forecast [23, 24]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To overcome this obstacle, existing models are forced to improve accuracy at the cost of efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Our contributions to partial differential equation solvers In this paper, we build on Koopman neural operator (KNO), one of our latest works [31], to develop an efficient module of PDE solving and overcome the limitation in characterizing the long-term behaviours of complicated PDE families.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' As a study on computational physics programs, our research has the fol- lowing contributions compared with our previous work [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' First, we generalize the original KNO to four kinds of vari- ants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Beyond the original KNO, these differentiated variants of- fer more possibilities for data-specific and task-oriented solver designs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, the compact variants of KNO realized by multi-layer perceptrons and convolutional neural networks can accurately solve PDE with small model sizes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The large vari- ants of KNO implemented on visual transformers can predict highly intricate dynamic systems governed by unknown, high- dimensional, and non-linear PDEs (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', climate system).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Second, we propose KoopmanLab, a PyTorch module of Koopman neural operator family, as a self-contained and user- friendly platform for PDE solving.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' All necessary tools, such as those for data loading, model construction, parameter manipu- lation, output visualization, and performance quantification, are offered in a user-friendly manner to support customized appli- cations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Third, we offer comprehensive validation of the proposed module on representative data sets, including those gener- ated by important PDEs in fluid mechanics (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the Navier- Stokes equation and the Bateman–Burgers equation) or ob- tained by global meteorological recording research (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', atmo- spheric, land, and oceanic climate fields in ERA5 data set) [32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' By measuring accuracy, quantifying efficiency, and comparing all KNO variants with other state-of-the-art alternatives (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Fourier neural operator [19] and FourCastNet [23, 24]), we sug- gest the potential of our module to serve as an ideal choice of PDE solving and dynamic system prediction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The initial version of Koopman neural operator Although the original Koopman neural operator has been proposed in our earlier work [31], here we elaborate on its mechanisms for completeness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' We further present more math- ematical details that are not covered in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [31] to analyze the convergence of the original Koopman neural operator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The original Koopman neural operator: Objective Koopman neural operator (KNO) is proposed to deal with the non-linear, and potentially non-autonomous, dynamic system in Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (6-7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The idea underlying KNO arises from the pursuit to transform the non-linear system in Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (6-7) to a sufficiently simple linear one ∂tg (γt) = Ag (γt) , ∀t ∈ T, (10) where g (·) is an appropriate transform and A is a linear op- erator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In modern dynamic system theory [28], this pursuit may be achieved if we can develop an approach to character- ize the Koopman operator K, an infinite-dimensional linear op- erator governing all possible observations of the dynamic sys- tem of equation solution γ (·), to act on the flow mapping θ and 3 linearizing the dynamics of γ (·) in an appropriate observation space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' This idea has been extensively applied in plasma physics [33], fluid dynamics [34], robot kinetics [35], and neuroscience [36].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mathematically, we need to find a set of observation func- tions (or named as measurement functions) [28] G � Rdγ × T � = {g|g : Rdγ × T → Cdγ} (11) such that a family of Koopman operators can be identified for the autonomous (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Kε : G � Rdγ × T � → G � Rdγ × T � ) or the non-autonomous (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Kt+ε t : G � Rdγ × T � → G � Rdγ × T � ) case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These Koopman operators can function on the observa- tions of γ (·) to update them Kεg (γt) = g (θε (γt)) = g (γt+ε) , ∀t × T, (12) Kt+ε t g (γt) = g � θt+ε t (γt) � = g (γt+ε) , ∀t ≤ t + ε ∈ T, (13) where Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (12-13) correspond to the autonomous and non- autonomous cases, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The updating is implemented in a linear manner, which can be illustrated by taking the non- autonomous case as an example ∂tg (γt) = lim ε→0 Kt+ε t g (γt) − g (γt) ε .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (14) Apart from the linear system of g (γt) in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (14), one may also consider the Lie operator (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the Lie derivative of g (·) along the vector field γ (·)), which is generator operator of such a Koopman operator [37, 38, 39] Ltg = lim t+ε→t Kt+ε t g (γt) − g (γt) t + ε − t .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (15) Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (15) defines a linear system of g (γt) as well ∂tg (γt) = lim t+ε→t Kt+ε t g (γt) − g (γt) ε = Ltg (γt) , (16) which can also be considered in the application.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To understand the linearization of g (γt) by the Koopman op- erator from the perspective of PDE solving, let us consider the Lax pair (M, N) of an integrable version of Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (1-3) [40] M = Dn x + αγ (xt) I, α ∈ C, (17) Mψ (xt) = λψ (xt) , λ ∈ C, (18) ∂tψ (xt) = Nψ (xt) , (19) where Dn x denotes the n-th total derivative operator and I is an identity operator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (18) denotes an eigenvalue problem at moment t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A relation between linear operators M and N can be identified if we calculate the time derivative of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (18) (∂tM + MN − NM) ψ (xt) = ∂tλψ (xt) , (20) which directly leads to ∂tM + [M, N] = 0, (21) where [M, N] = MN − NM denotes the commutator of op- erators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Combining Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (17-21) with Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (16), we can readily see the close relation between N and Kt+ε t ψ (D × {t}) = g (γt) ⇒ N = lim t+ε→t Kt+ε t g (γt) − g (γt) ε , (22) which holds in the autonomous case as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In sum, the lin- earization of g (γt) is intrinsically related to the Lax pair and the inverse scattering transform of integrable PDEs [40].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Note that similar ideas have been comprehensively explored in math- ematics and physics [41, 42, 43, 44].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Once we find a way to derive the Koopman operator, we can reformulate Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (4) as �γt+ε = g−1 � Kt+ε t g ��γt �� , ∀t ∈ T, (23) where �γt = �γ (D × {t}).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Certainly, an infinite-dimensional lin- ear operator is not operable in practice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To enable neural net- works to learn a potential Koopman operator, we need to con- sider � K ∈ Rr×r, a finite matrix, as a counterpart of K that acts on K = span �� G � , a finite invariant sub-space spanned by � G = {g1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , gr} ⊂ G � Rdγ × T � � Kgi = ⟨[ν1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , νr] , �g1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , gr �⟩, ∀gi ∈ � G, (24) where [ν1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , νr] ∈ Rr and ⟨·, ·⟩ denotes the inner product.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mathematically, any finite set of eigenfunctions of the Koop- man operator K can span a finite invariant sub-space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The original Koopman neural operator: Mathematics There exist numerous previous works that pursue to char- acterize the Koopman operator by machine-learning-based ap- proaches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Some approaches are highly practical but limited to the autonomous case (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the case in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (12)) [45, 46, 47, 48].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Other approaches are more general in application but critically depend on a priori knowledge about the eigenvalue spectrum (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g, the numbers of real and complex eigenvalues) of Koopman operator to deal with the continuous spectrum problem [49].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In practice, a balance should be reached between mathe- matical completeness and computational practicality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' An ideal Koopman-operator-based PDE solver should fit in with both au- tonomous and non-autonomous cases and limit the dependence of a priori knowledge as much as possible (even though these restraints inevitably reduce mathematical completeness).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To ex- plore such a balance, we introduce the Koopman neural opera- tor (KNO), a flexible approach, in our previous work [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The formalization of KNO begins with the Krylov sequence [50] of the observable defined by a unit time step ε ∈ [0, ∞], which is used in the Krylov subspace approach for comput- ing the eigenvalues of large matrices [50].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' One can see its application in Koopman-operator-related algorithms such as the Hankel-DMD [51], sHankel-DMD [52], and HAVOK [53].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, the Krylov sequence is given as Rn = �g (γ0) , g (γε) , g (γ2ε) , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , g (γnε)� , (25) 4 which is generated by K and g (γ0) Rn = � g (γ0) , Kε 0g (γ0) , K2ε ε Kε 0g (γ0) , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , Knε (n−1)ε · · · Kε 0g (γ0) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (26) Computationally, the Krylov sequence can be sampled by a Hankel matrix of observations Hm×n = ������������ g (γ0) g (γε) · · g (γnε) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' g �γ(m−1)ε � g (γmε) · · g �γ(m+n−1)ε � ������������ , (27) where m ∈ N+ denotes the dimension of delay-embedding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (27), each column is a sampled result that approximates a function in the Krylov subspace.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' If the Koopman operator has a discrete spectrum (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', has eigenvalues), there exists an invariant subspace K of the Koop- man operator, which can be spanned by the Krylov subspace K = span (Rn) ≃ span �H(m,n) � (28) as long as n ≥ dim (K) − 1 (here dim (·) denotes the dimension- ality).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' This property suggests the possibility of approximating the actual Koopman operator to K by � Kt+ε t : G � Rdγ × T � → K, a finite Koopman operator restricted to K for any t ∈ T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Math- ematically, matrix � K is required to satisfy the Galerkin projec- tion relation ⟨ � Kt+ε t h (γt) , g (γiε)⟩ = ⟨Kt+ε t h (γt) , g (γiε)⟩, ∀i = 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , m, (29) where h (·) ∈ G � Rdγ × T � is an arbitrary function [54, 55].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' If the target Koopman operator is bounded and H(m,n) spans its invariant subspace, the approximation can be realized by lim m→∞ � G(Rdγ ×T) ∥ � Kt+ε t h (γt) − Kt+ε t h (γt) ∥Fdµ = 0, ∀h (·) ∈ G � Rdγ × T � , (30) where µ is a measure on G � Rdγ × T � and ∥ · ∥F denotes the Frobenius norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Once a restricted Koopman operator is de- rived, we can obtain the following iterative dynamics Hm×n (k + 1) = � K(k+1)ε kε Hm×n (k) , ∀k = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , n, (31) in which Hm×n (k) is the k-th column of Hm×n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' As for the case where the Koopman operator has a continuous spectrum (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', has no eigenvalue), there is no finite invariant subspace to support computational approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Such an ill- posed situation remains for future exploration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The restricted Koopman operator � K can be learned efficiently if it corresponds to autonomous system, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', � Kt+ε t = � Kε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' How- ever, an online optimization will be necessary if it corresponds to non-autonomous system, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', � Kt+ε t is time-varying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lim- ited by computing resources or data size, expensive online op- timization may not always be available during PDE solving.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Therefore, we propose a compromised approach to realize off- line training under the ergodic assumption [51, 56] of the dy- namic system of γt, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', γt ultimately visits every possible sys- tem states as t → ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Under this assumption, the proportion of retention time of γt at a certain system state is equivalent to the probability of this state in the space, making the time-averaging equivalent to the actual expectation at the limit of infinite time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Based on this property, we can define an expectation of the re- stricted Koopman operator associated with ε Kε = lim t→∞ 1 t � [0,t) g (γτ)−1 g (γτ+ε) dτ, (32) ≃ argmin P∈R n � k=1 ∥Hm×n (k + 1) − PHm×n (k) ∥F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (33) For a fixed time difference ε, the expected Koopman operator Kε : G � Rdγ × T � → K is a time-average of � Kt+ε t that can be learned during offline optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The original Koopman neural operator: Convergence Given an ideal setting of m → ∞, we can ensure the conver- gence of the eigenvalues and eigenfunctions of � K to those of K under the assumption of ergodic property.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similar conclu- sions can be seen in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [54].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To understand this convergence, we need to indicate several important properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' First, as we have mentioned, there exists an equivalence relation between time-averaging and the real expectation as the time approaches to infinity (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the Birkhoff ergodic theorem [51, 56]) lim m→∞ 1 m m−1 � i=0 g (γi) = � K gdµ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (34) Second, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (34) directly implies that lim m→∞ 1 m⟨Hm×n (i) , Hm×n ( j)⟩ = � K Hm×n (i) �Hm×n ( j)�∗ dµ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (35) = ⟨Hm×n (i) , Hm×n ( j)⟩K, (36) where ∗ denotes the complex conjugate and ⟨·, ·⟩K stands for the inner product of functions in K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given the learned Koopman operator Kε, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (36) coincides with the definition of the actual Gramian matrix V associated with the inner product space K Vi,j = ⟨K i−1 ε Rn, K j−1 ε Rn⟩K, (37) = ⟨Hm×n (i) , Hm×n ( j)⟩K, (38) where we mark K i−1 ε Rn = � K i−1 ε g (γ0) , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , K i−1 ε g (γnε) � for convenience.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (38) is derived from the fact that Hm×n serves as the sampling of Rn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Meanwhile, the left side of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (35) co- incides with the empirical Gramian matrix � V associated with matrix Hm×n � Vi,j = 1 m⟨Hm×n (i) , Hm×n ( j)⟩.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (39) 5 Our formal proof can be developed based on these two prop- erties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Let us consider the first r < n element of Rn Rr = � g (γ0) , Kεg (γ0) , K 2 εg (γ0) , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , K r−1 ε g (γ0) � , (40) which defines a possible basis of K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Theoretically, the learned Koopman operator restricted to K can be represented by a companion matrix C = ���������������������� 0 0 · · 0 c0 1 0 · · 0 c1 0 1 · · 0 c2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 0 0 · · 1 cr−1 ���������������������� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (41) The last column of C denotes the coordinate of K r εg (γ0) defined by the basis, which should be calculated as C (r) = V−1 ��������������������������� ⟨g (γ0) , K r εg (γ0)⟩K ⟨Kεg (γ0) , K r εg (γ0)⟩K ⟨K 2 εg (γ0) , K r εg (γ0)⟩K .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ⟨K r−1 ε g (γ0) , K r εg (γ0)⟩K ��������������������������� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (42) Empirically, the learned Koopman operator restricted to span �H(m,n) � can be also represented by a companion matrix, whose last column can be calculated as � C (r) = 1 m � V−1 ��������������������������� ⟨Hm×n (1) , K r εHm×n (1)⟩ ⟨KεHm×n (1) , K r εHm×n (1)⟩ ⟨K 2 εHm×n (1) , K r εHm×n (1)⟩ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ⟨K r−1 ε g (γ0) , K r εg (γ0)⟩K ��������������������������� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (43) It is trivial to prove lim m→∞ � V−1 = � lim m→∞ � V �−1 = V−1 (44) applying Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (36) and Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (38-39).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similarly, we can know lim m→∞ 1 m ��������������������������� ⟨Hm×n (1) , K r εHm×n (1)⟩ ⟨KεHm×n (1) , K r εHm×n (1)⟩ ⟨K 2 εHm×n (1) , K r εHm×n (1)⟩ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ⟨K r−1 ε g (γ0) , K r εg (γ0)⟩K ��������������������������� = ��������������������������� ⟨g (γ0) , K r εg (γ0)⟩K ⟨Kεg (γ0) , K r εg (γ0)⟩K ⟨K 2 εg (γ0) , K r εg (γ0)⟩K .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ⟨K r−1 ε g (γ0) , K r εg (γ0)⟩K ��������������������������� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (45) Therefore, we can derive lim m→∞ � Cij = Cij, ∀i, j ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , r}2 (46) based on Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (44-45), implying that lim m→∞ � M∈PMk �� C � M = � M∈PMk(C) M, ∀k ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , r}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (47) In Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (47), notion PMk (·) denotes the set of all the k-order principal minors of the corresponding matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Now, let us con- sider the characteristic polynomials of � C and C P� C (z) = zr + r� i=1 (−1)i ������������ � M∈PMi �� C � M ������������ zr−i, (48) PC (z) = zr + r� i=1 (−1)i �������� � M∈PMi(C) M �������� zr−i, (49) whose distance at the limit of m → ∞ can be measured as lim m→∞ ∥P� C (z) − PC (z) ∥ = lim m→∞ max 1≤i≤r ������ (−1)i ������������ � M∈PMi �� C � M − � M∈PMi(C) M ������������ ������, (50) =0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (51) Because the roots of a given polynomial evolve continuously as the function of the coefficients, we know that � C and C share the same set of eigenvalues at the limit of m → ∞ since their characteristic polynomials converge to the same.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Moreover, the convergence of � C to C and the convergence of the eigenvalues � C to those of C eventually imply the convergence of the eigen- functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The original Koopman neural operator: Computation In Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [31], we have proposed an architecture to implement the original Koopman neural operator on neural networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The details of architecture designs are presented below: Part 1: Observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' An encoder (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', a single non- linear layer with tanh (·) activation function in the original Koopman neural operator) serves as observation function g (·) to transform φt = φ (D × {t}), an arbitrary input of the PDE family (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', φt can be directly set as γt), into g ��γt � ∈ G � Rd�γ × T � g ��γt � = Encoder (φt) , ∀t ∈ T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (52) Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 2: Fourier transform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similar to the Fourier neural operator [19], the original Koopman neural operator also utilizes the Fourier transform during the iterative update of the Green function in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given g ��γt �, we derive the Fourier transform gF (·) = F ◦ g (·), where we truncate the Fourier series at ω, a maximum frequency gF (ξ) = χ[0,ω] (ξ) � D×{t} g ��γ (xt)� exp (−2πi⟨xt, ξ⟩) dxt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (53) Note that χ· (·) denotes the indicator function (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', χA (a) = 1 if a ∈ A, otherwise χA (a) = 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Computationally, the 6 Figure 1: Conceptual illustrations of neural network architectures of the original Koopman neural operator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (a) summarizes key mathematical transform in each part, where r is the prediction length.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (b) visualizes a prediction instance on the 2-dimensional Navier-Stokes equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' above transform is implemented by fast Fourier transform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For convenience, we mark gF ��γt � = F ◦ g ��γt � = {gF (ξ) |ξ ∈ [0, ∞)} (54) as the transformed result of �γt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Different from Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [19], our main motivation for using the truncated Fourier trans- form is to extract the low-frequency information (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', main system components) of the represented equation so- lution g ��γt �.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Certainly, frequency truncation inevitably causes the loss of high-frequency information (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', high- frequency perturbations or edges).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In the original Koop- man neural operator, Part 5 is designed to complement the lost information associated with high-frequency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' See Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for more details.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 3: Hankel representation and offline Koopman operator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Once we have derived gF ��γt � for every t ∈ εN+, a Hankel matrix � Hm×n of gF ��γt � will be generated follow- ing m ∈ N, a dimension of delay-embedding (note that n ∈ N is the number of all accessible samples) � Hm×n = ������������ gF ��γ0 � gF ��γε � · · gF ��γnε � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' gF ��γ(m−1)ε � gF ��γmε � · · gF ��γ(m+n−1)ε � ������������ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (55) We train a o × o linear layer to learn a neural network rep- resentation of Koopman operator Kε : G � Rd�γ × T � → �K following Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (32-33), where �K is spanned by � Hm×n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The learned Kε can be used to predict the future state of gF ��γ(m+n−1)ε � as gF ��γ(m+n+r−1)ε � = � K r ε � Hm×n (n) �T (m) , r ∈ N+ (56) where notion T denotes the transpose of a matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 4: Inverse Fourier transform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' After gF ��γ(m+n)ε � is predicted in Part 3, it is transformed from the Fourier space to G � Rd�γ × T � by an inverse Fourier transform g ��γ (xt)� = 1 (2π)d�γ � ∞ −∞ gF (ξ) exp (2πi⟨xt, ξ⟩) dξ, (57) where t = (m + n + r − 1) ε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For convenience, we mark g ��γ(m+n+r−1)ε � = F −1 ◦ gF ��γ(m+n+r−1)ε � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (58) Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for instances of Part 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 5: High-frequency information complement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In the original Koopman neural operator, we use a convo- lutional layer to extract high-frequency of g ��γt � because convolutional layers can amplify high-frequency compo- nents according to Fourier analysis [57].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Therefore, we train a convolutional layer C on the outputs of Part 1 to ex- tract their high-frequency information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' As a complement of Parts 2-4, the convolutional layer realizes a forward prediction of high-frequency information � gC � �γ(j+r−1)ε � , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , gC � �γ(j+m+r−1)ε ��T =C � g � �γ jε � , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , g � �γ( j+m)ε ��T , ∀j = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (59) See Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 6: Inverse observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Once two future states, g ��γ(m+n)ε � and gC ��γ(m+n)ε �, are predicted by Parts 2-4 and Part 5, they are unified in a linear manner gU ��γ(m+n+r−1)ε � = g ��γ(m+n+r−1)ε � + gC ��γ(m+n+r−1)ε � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (60) Given gU ��γ(m+n)ε �, a non-linear decoder (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', a single non-linear layer with tanh (·) activation function in the original neural operator) is trained to approximate the in- verse of observation function g−1 (·) ≃ g−1 U (·) (61) and derive �γ(m+n+r−1)ε = Decoder �gU ��γ(m+n+r−1)ε �� (62) as the target state of equation solution in space Rd�γ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 7 (a) Ground truth Ground truth (b) Part 2 Part 3 Part 1 Part 2 Part 3 Part 4 Part 6 Part 1 Part 4 Part 6 9F(V) gF(Ine) F· i ((3(I-u+u)s)6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='(3(t-u)s)6) (j +m)a +m+r)c je G+18 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ()6 Truncate* [Hmxn(n)] (m) noε noD Nt t+r?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 5 g-1() Part 5 [ge(V(j+r-1)e) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 9e(V(j+m+r-1)8 ()6 C[g()je) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' g(V(i+m))]Parts 1-6 define the iterative update strategy of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (23).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For any t′ > t ∈ εN, the iterative dynamics is given as �γt′ = � g−1� F −1 ◦ K t′−t ε F ◦ g ��γ[t−mε,t] � �������������������������������������������������������������������� Parts 1-4 + C ◦ g ��γ[t−mε,t] � ���������������������������� Part 1 and part 5 ��T (m) , (63) in which�γ[t−mε,t] denotes a vector ��γt−mε, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ,�γt �.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The loss func- tion for optimizing Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (63) is defined as L = λp∥�γt′ − γt′∥F + λr m � i=0 ∥g−1 ◦ g ��γt−imε � − γt−imε∥F, (64) where λp, λr ∈ (0, ∞) denotes the weights of prediction and reconstruction processes in the loss function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The one-unit architecture of the original Koopman neural op- erator is visualized in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A multi-unit architecture can be readily constructed by cascading the copy of Parts 2-5 multiple times.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The Koopman neural operator family Beyond the original Koopman neural operator (KNO) [31], we generalize it to four kinds of variants to fit in with different application demands.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The compact KNO sub-family: Definition The compact KNO sub-family includes two compact vari- ants of KNO realized by multi-layer perceptrons (MLP-KNO) and convolutional neural networks (CNN-KNO).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These vari- ants are proposed to accurately solve PDEs with small model sizes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, they are designed following Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (52-62), where encoder and decoder are defined as Encoder = η ◦ E, E ∈ {We, Ce}, (65) Decoder = η ◦ D, D ∈ {Wd, Cd}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (66) In Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (65-66), mapping η denotes a non-linear activation function (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', we use tanh (·) in our research), notions We and Wd are two weight matrices of the corresponding sizes, and Ce and Cd are two convolutional layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Our proposed KoopmanLab module offers user-friendly tools to customize an MLP-KNO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A customized instance is presented below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 import koopmanlab as kp 2 3 MLP_KNO1D = model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='koopman(backbone = "KNO1d", autoencoder = "MLP", o = o, f = f, r = r, device = device) 4 5 MLP_KNO2D = model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='koopman(backbone = "KNO2d", autoencoder = "MLP", o = o, f = f, r = r, device = device) 6 7 MLP_KNO1D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 8 MLP_KNO2D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 9 10 ## Parameter definitions : 11 # o: the dimension of the learned Koopman operator 12 # f: the number of frequency modes below frequency truncation threshold 13 # r: the power of the Koopman operator in EQ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (56) 14 # device: if CPU or GPU is used for computation Similarly, a CNN-KNO can be customized using the follow- ing code, where we present 1-dimensional and 2-dimensional visions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 import koopmanlab as kp 2 3 CNN_KNO_1D = model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='koopman(backbone = "KNO1d", autoencoder = "Conv1d", o = o, f = f, r = r, device = device) 4 5 CNN_KNO_1D .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 6 7 CNN_KNO_2D = model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='koopman(backbone = "KNO2d", autoencoder = "Conv2d", o = o, f = f, r = r, device = device) 8 9 CNN_KNO_1D .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 10 CNN_KNO_2D .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 11 12 ## Parameter definitions : 13 # o: the dimension of the learned Koopman operator 14 # f: the number of frequency modes below frequency truncation threshold 15 # r: the power of the Koopman operator in EQ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (56) 16 # device: if CPU or GPU is used for computation 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The compact KNO sub-family: Validation To validate the proposed compact KNO sub-family in PDE solving tasks, we design mesh-independent and long-term pre- diction tasks on representative PDEs (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the 2-dimensional Navier-Stokes equation [58] and the 1-dimensional Bate- man–Burgers equation [59]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The numerical data sets of these two are provided by Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, the incompressible 2-dimensional Navier- Stokes equation has a vorticity form ∂tγ (xt) + χ (xt) ∇γ (xt) = ν∆γ (xt) + ψ (xt) , xt ∈ (0, 1)2 × (0, ∞) , (67) ∇χ (xt) = 0, xt ∈ (0, 1)2 × (0, ∞) , (68) γ (x0) = γI, x0 ∈ (0, 1) × {0}, (69) in which γ (·) denotes the vorticity, χ (·) measures the velocity, ψ (·) is a time-independent forcing term.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The viscosity coeffi- cient is ν ∈ {10−3, 10−4} in our research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given the data with the highest mesh resolution, one can further generate the data with the lower resolution by direct down-sampling [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The data with the highest mesh resolution has 213 grids [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Our KoopmanLab module offers a function to load the data of the incompressible 2-dimensional Navier-Stokes equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 import koopmanlab as kp 2 3 train_loader , test_loader = kp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' navier_stokes (path , batch_size = 10, T_in = 10, T_out = 40, type = "1e-3", sub = 1) 8 Figure 2: Experimental validation of the compact KNO sub-family.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (a) Results of the mesh-independent experiment on the Bateman–Burgers equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (b) Results of the long-term prediction experiment on the 2-dimensional Navier-Stokes equation with a viscosity coefficient ν = 10−3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (b) Results of the same long-term prediction experiment on the 2-dimensional Navier-Stokes equation with ν = 10−4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (d) The prediction results and errors (RMSE) on the 2-dimensional Navier-Stokes equation with ν = 10−3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (e) The prediction results and errors (RMSE) on the 2-dimensional Navier-Stokes equation with ν = 10−4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 4 5 ## Parameter definitions : 6 # path: the file path of the downloaded data set 7 # T_in: the duration length of input data 8 # T_out: the duration length required to predict 9 # Type: the viscosity coefficient 10 # sub: the down -sampling scaling factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For instance , a scaling factor sub =2 acting on a 2- dimensional data with the spatial resolution 64*64 will create a down -sampled space of 32*32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The same factor action on a 1- dimensional data with the spatial 9 10-4 (b) (a) KNO(0=32,f=10,r=12,MLP) KNO(0=128,f=10,r=16,Conv) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='35 KNO(0=32,f=10,r=12,MLP) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00175 KNO(0=32,f=16,r=8,MLP) KNO(0=32,f=16,r=8,MLP) KNO(o=32,f=10,r=24,Conv) KNO(0=48,f=10,r=12,MLP) KNO(0=48,f=10,r=12,MLP) KNO(o=8,f=10,r=10,Conv) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='30 KNO(0=128,f=10,r=16,MLP) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00150 KNO(o=32,f=10,r=12,Conv) KNO(o=32,f=10,r=12,Conv) KNO(0=32,f=10,r=24,MLP) KNO(o=32,f=16,r=8,Conv) KNO(o=32,f=16,r=8,Conv) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='25 KNO(0=8,f=10,r=10,MLP) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00125 KNO(o=48,f=10,r=12,Conv) KNO(o=48,f=10,r=12,Conv) FNO(Original, 1 unit) FNO(Original, 1 unit) RMSE E E 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='201 00.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00100 MSI RM P 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00075 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='10 10-5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00050 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00025 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00000 2829 210 6E E E E E 67 Z EZ Z 61 T ST T 1 6 12 13 14 15 16 17 18 19 20 211 Resolution Time Step Time Step (d) Initial Condition CNN-KNO MLP-KNO FNO 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='10 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='10 CNN-KNO (Error) MLP-KNO (Error) FNO (Error) (e) 2 Initial Condition CNN-KNO MLP-KNO FNO 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 FNO (Error) CNN-KNO (Error) MLP-KNO (Error)Models Settings Parameter number FNO default settings, one-unit [19] 233897 MLP-KNO (o, f, r) = (32, 10, 12) 206538 MLP-KNO (o, f, r) = (32, 16, 8) 526026 MLP-KNO (o, f, r) = (48, 10, 12) 464170 CNN-KNO (o, f, r) = (32, 10, 12) 206538 CNN-KNO (o, f, r) = (32, 16, 8) 526026 CNN-KNO (o, f, r) = (48, 10, 12) 464170 Table 1: The parameter numbers of all implemented models in Figs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(b-c), which are counted by the tool provided by Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [23, 24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' resolution 1*64 implies a down -sampled space of 1*32 The 1-dimensional Bateman–Burgers equation is defined as ∂tγ (xt) + ∂x �γ2 (xt) 2 � = ν∂xxγ (xt) , xt ∈ (0, 1) × (0, 1] , xt ∈ (0, 1) × (0, 1] , (70) γ (x0) = γI, x0 ∈ (0, 1) × {0}, (71) in which γI stands for a periodic initial condition γI ∈ L2 periodic [(0, 1) ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' R] and parameter ν ∈ (0, ∞) is the viscosity co- efficient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' We set ν = 100 in our research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The data with highest mesh resolution has 216 grids [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To load this data set, one can use the following function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 import koopmanlab as kp 2 3 train_loader , test_loader = kp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='burgers(path , batch_size = 64, sub = 32) 4 5 ## Parameter definitions : 6 # path: the file path of the downloaded data set 7 # sub: the down -sampling scaling factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For instance , a scaling factor sub=2 acting on a 2- dimensional data with the spatial resolution 64*64 will create a down -sampled space of 32*32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The same factor action on a 1- dimensional data with the spatial resolution 1*64 implies a down -sampled space of 1*32 In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(a), we validate the mesh-independent property of the proposed compact KNO sub-family adopting the same setting used in our earlier work [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The mesh-independent property, as suggested by Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [16, 17, 18, 5, 19, 20, 21], arises from the fact that the neural operator is expected to learn the solution operator of an entire PDE family rather than be limited to a concrete parameterized instance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, we conduct the experiments on the data of 1-dimensional Bate- man–Burgers equation associated with different mesh granu- larity conditions (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', spatial resolution of meshes).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Different versions of the compact KNO sub-family are defined by chang- ing hyper-parameters (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', operator size o, frequency mode number f, and the power of the Koopman operator r = t′−t ε in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (56)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These models are trained by 1000 randomly se- lected samples with the lowest spatial resolution and conduct 1-second forward prediction on 200 samples associated with different resolutions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Batch size is set as 64, the learning rate is initialized as 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='001 and halved every 100 epochs, and the weights of prediction and reconstruction in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (64)) are set as � λp, λr � = (5, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' As shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(a), the prediction errors of all versions of the compact KNO sub-family maintain con- stantly across different spatial resolutions, suggesting the ca- pacity of the compact KNO sub-family to be mesh-independent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mesh-independence is important for PDE solving because it al- lows one to train a neural-network-based PDE solver on the data with low spatial resolution and directly apply the solver on the data with high spatial resolution, which breaks the trade-off of accuracy and efficiency in PDE solving.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In our earlier work [31], one can further see a detailed comparison between the original KNO and FNO [19] in mesh-independent prediction task, where the original KNO outperforms FNO with a much smaller model size (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', a size of 5×103 for KNO and a size of 2 × 107 for FNO).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Other neural operator models, such as graph neural operator (GNO) [5] and multipole graph neural opera- tor (MGNO) [60], are no longer considered because they have been demonstrated as less accurate than FNO as reported by Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(b-e), we validate the compact KNO sub-family by a long-term prediction task designed on the 2-dimensional Navier-Stokes equation data sets with viscosity coefficients ν = 10−3 (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(b)) and ν = 10−4 (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(c)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A down-sampling scaling factor of 2 is defined to generate the data sets with 212 grids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For comparison, a one-unit FNO is defined following the default setting introduced in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A 40-time-interval pre- diction task is conducted on the data set with ν = 10−3, where models are trained on 1000 samples of γ � (0, 1)2 × [0, 10) � and tested on 200 samples of γ � (0, 1)2 × (10, 50] � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similarly, a more challenging 10-time-interval prediction task is conducted on the data set with ν = 10−4, in which models are trained on 8000 samples of γ � (0, 1)2 × [0, 10) � and tested on 200 samples of γ � (0, 1)2 × (10, 20] � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(b-c) report the prediction per- formance of all models as the function of increasing prediction duration length.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(d-e) visualize predicted instances and errors in the cases with ν = 10−3 (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(d)) and ν = 10−4 (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 2(e)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' All experiment results suggest the optimal potential of the compact KNO sub-family in characterizing the long-term evolution of PDE solutions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Combining these results with the model sizes measured in Table.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1, we suggest that the compact KNO sub-family realizes a better balance between accuracy and efficiency because a KNO variant with a smaller model size can still outperform FNO significantly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The ViT-KNO sub-family: Definition Different from the compact KNO sub-family, the ViT-KNO sub-family is proposed for dealing with more intricate situa- 10 Figure 3: Conceptual illustrations of neural network architectures of the ViT-KNO sub-family.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (a) summarizes the computational design of each part.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (b) illustrates an instance of the parallel design (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', multi-head design) of the ViT-KNO sub-family, where the depth of each head is 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' tions (here ViT stands for Vision Transformer [61]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Numerous applications of PDE solving (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', global climate forecasting) require the solver to be able to capture the underlying patterns of ultra-large data sets that may be related with certain unknown PDEs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Meanwhile, there may exist multiple variables of inter- est that govern by a group of coupled PDEs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' To fit in with these situations, we follow the main idea of the compact KNO sub- family to develop a kind of transfer-based PDE solver.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The mechanism underlying the proposed ViT-KNO sub-family is not completely same as Eqs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (52-62) because some mathemati- cal details are modified to improve model applicability on noisy real data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' We suggest the benefits of our modifications based on an experiment on ERA5, one of the largest data set of global atmospheric, land, and oceanic climate fields [32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Nev- ertheless, more in-depth mathematical analyses of these modi- fications remain for future studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Let us consider a case where there exist v coupled variables, {γ1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , γh}, defined on domain D × T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The dynamics of these variables are govern by a group of PDEs with unknown expres- sions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The objective is to learn the equation solutions of these latent PDEs such that the dynamics of {γ1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , γh} can be accu- rately characterized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The architecture of ViT-KNO sub-family consists of 7 parts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Below, we present a detailed computational implementation of each part.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 1: Observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similar to the encoder design in the compact KNO sub-family, an encoder component is implemented in ViT-KNO to serve as observation function g (·) and transform φs t = φs (D × {t}) into g ��γs t � for each s ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , h}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, the encoder is realized by the token embedding layer in Vision Transformer (ViT) [61].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given a joint input � φ1 t , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , φh t � ∈ Rdφ×h, we first transform it into a 3-dimensional token tensor Φt by a convolutional layer Ce Ce �� φ1 t , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , φh t �� = g ��Γt � ∈ Ru×v×l, (72) where domain D is reorganized into u × v patches (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', to- kens).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The patch is a kind of square and non-overlapping macro-mesh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' If domain D has been already discretized into multiple meshes, then the size of a patch equals the number of meshes it covers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Parameter l denotes a cus- tomized embedding dimension, which is not necessarily the same as h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The derived tensor g ��Γt � denotes the joint representation of � g � �γ1 t � , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , g � �γl t �� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 2: Fourier transform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similar to adaptive Fourier neural operator [23, 24, 62], a truncated Fourier transform is applied on the first two dimensions of g ��Γt � to derive the Fourier series of each embedded variable s ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , l} gs F (ξ) = χ[0,ω] (ξ) � [u]×[v]×{t} g ��γs (xt)� exp (−2πi⟨xt, ξ⟩) dxt, (73) where [u] = {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , u}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For convenience, we mark gs F ��γt � = F ◦ g ��γs t � = � gs F (ξ) |ξ ∈ [0, ∞) � , (74) gF ��Γt � = F ◦ g ��Γt � = � g1 F ��γt � , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , gl F ��γt �� , (75) in which s ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , l}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Similar to the compact KNO sub-family, frequency truncation leads to the loss of high- frequency information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In the ViT-KNO sub-family, Part 5 is designed for complementing high-frequency informa- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' See Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3 for details.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 3: Koopman-operator-associated component.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Af- ter deriving gF ��Γt � for every t ∈ εN+, a Koopman- operator-associated component is designed to function on the third dimension of every token in gF ��Γt � and realize the iterative dynamics gF ��Γt+ε � = KεSgF ��Γt � = � g1 F ��γt+ε � , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , gl F ��γt+ε �� , (76) S = η ◦ W, (77) in which Koopman operator Kε is learned by a linear transform, layer S is constructed by a non-linear activa- tion function η acting on a linear layer W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Although S is not a part of the original Koopman neural operator [31], in- cluding it can efficiently enhance the capacity of this com- ponent to characterize intricate large-scale data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In Koop- manLab,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the leaky rectified linear unit (Leaky ReLU) [63] 11 (a) Ground truth of multiple variables (b) Ground truth of multiple variables Head 1 Part 1 Part 2 Part 3 Part 4 Part 6 Part 7 Part 1 Part 7 Part 2 Part 3 Part 4 甲 Part 6 gF(ft+e) Part 5 Variable Token decoding Token decoding F-1()- Token encoding F(O) Token K,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='SgF(ft+e) Head 2 Partition for 田 l encoding Part 6 coupling parallel design Part 5 Part 5 Head 3 gc(ft+) = Cg(ft+) Part 2-* Part 3 - Part 4 田 Part 6 M g-1() g-1() g() g() Part 5is suggested as a default choice of S,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' which can also re- duce to the ReLU function as a special case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3 for illustrations of Part 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 4: Inverse Fourier transform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Once gF ��Γt+ε � is de- rived in Part 3, the inverse Fourier transform is applied on the first two dimensions of gF ��Γt+ε � to transform gF ��Γt+ε � back to the observation space g ��γs (xt+ε)� = 1 (2π)d�γ � ∞ −∞ gs F (ξ) exp (2πi⟨xt+ε, ξ⟩) dξ, (78) g ��Γt+ε � = F −1 ◦ gF ��Γt+ε � = � g � �γ1 (xt+ε) � , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , g � �γl (xt+ε) �� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (79) Please see instances in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 5: High-frequency information complement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Same as the compact KNO sub-family, there is a com- ponent for complementing high-frequency information in ViT-KNO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' This component is also realized by a convolu- tional layer C that acts on the outputs of Part 1 to learn the dynamics of high-frequency information gC ��Γt+ε � = Cg ��Γt � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (80) See Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 6: Variable coupling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given two predicted states, g ��Γt+ε � and gC ��Γt+ε � , by Parts 2-4 and Part 5, we com- bine them in a linear form gU ��Γt+ε � = g ��Γt+ε � + gC ��Γt+ε � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (81) Because ViT-KNO is designed to learn multi-variate sys- tems governed by unknown coupled PDEs, we need to characterize the coupling relation among variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Be- cause we lack the a priori knowledge about these underly- ing PDEs, we suggest to capture the coupling relation by optimizing a non-linear layer M gM ��Γt+ε � = MgU ��Γt+ε � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (82) Following the idea of adaptive Fourier neural operator [23, 24, 62], we use the Gaussian Error Linear Unit (GELU) as the activation function in this non-linear layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Part 7: Inverse observation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given gM ��Γt+ε � , a decoder is implemented to function as the inverse of observation function � �γ1 t+ε, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ,�γh t+ε � ≃ g−1 � gM ��Γt+ε �� = Decoder � gM ��Γt+ε �� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (83) Similar to the compact KNO sub-family, there are two kinds of decoders included in our proposed KoopmanLab module Decoder ∈ {Wd, Cd}, (84) in which Wd and Cd denote linear and convolutional lay- ers, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' These two kinds of decoder designs dis- tinguish between two variants of the ViT-KNO sub-family.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' See Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3 for illustrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Parts 1-7 define the iterative update strategy of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (23) in a multi-variate case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' For any t ∈ T, the iterative dynamics is given as �Γt+ε = g−1 ◦ M ◦ � F −1 ◦ KεS ◦ F ◦ g ��Γt � + C ◦ g ��Γt � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (85) Multi-step prediction can be realized in an iterative manner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The loss function for optimizing Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (85) is L = λp h � s=1 ∥�γs t+ε − γs t+ε∥F + λr h � s=1 ∥g−1 ◦ g ��γs t � − γs t ∥F, (86) where λp, λr ∈ (0, ∞) are the weights of prediction and recon- struction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Several computational tricks can be considered in the appli- cation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' First, a LASSO regularization [64] can be included to improve the robustness and sparsity of Koopman operator Kε in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (76).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' This trick has been demonstrated as effective in adaptive Fourier neural operator [23, 24, 62] and is applicable to the ViT-KNO sub-family as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Second, the transformer ar- chitecture supports a parallel design the ViT-KNO sub-family.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Specifically, the third dimension of the output of Part 1 can be subdivided into multiple parts (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', g ��Γt � ∈ Ru×v×l is subdi- vided into k parts such that each part is an element in Ru×v× l k ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Then, Parts 2-6 are copied k × j times, where each group of j copies is organized into a sequence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Each sequence of j copies is referred to as a head in the transformer, processes a corre- sponding 1 k part of g ��Γt � ∈ Ru×v×l, and shares parameters dur- ing optimization (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Computationally, parameters k and j are referred to as the number and the depth of heads.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The processed outputs of these k parallel heads are unified by Part 7 to derive the final prediction result.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In our proposed Koopman- Lab, these two tricks are included to improve computational efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Our KoopmanLab module supports customizing ViT-KNO frameworks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Below, we present an instance of ViT-KNO with a multi-layer perceptron as the decoder 1 import koopmanlab as kp 2 3 ViT_KNO = model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' koopman_vit (decoder = "MLP", resolution =(1440 , 720) , patch_size =(2, 2), in_chans =20, out_chans =20, head_num =20, embed_dim =768 , depth = 16, parallel = True , high_freq = True , device=device) 4 5 ViT_KNO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 6 7 ## Parameter definitions : 8 # resolution: the spatial resolution of input data 9 # patch_size: the size of each patch (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', token) 10 # in_chans: the number of target variables in the data set 12 Figure 4: Visualization of three instances of the experiment results on ERA5 data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (a-c) respectively show the time-dependent predicted results of global wind speed, U10, and V10 variables on selected moments, accompanied by corresponding ground truths.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please note that wind speed is not originally included in the 20 climate variables selected from ERA5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' It is calculated as wind speed = � U102 + V102.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 11 # out_chans: the number of predicted variables by ViT -KNO , which is usually same as in_chans 12 # head_num: the number of heads 13 # embed_dim: the embeding dimension denoted by l in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (72) 14 # depth: the depth of each head 15 # parallel: if parallel design is applied 16 # high_freq: if high -frequency information complement is applied 17 # device: if CPU or GPU is used for computation Similarly, a ViT-KNO whose decoder is a convolutional layer can be defined as the following 13 True wend speed True wend speed True wend speed Initial condition 36 hours later 72 hours later (a) 90°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 60°N+ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 28 30°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0m 30°N 0° 0° 0。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 14 30°S 30°S- S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='03 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 60°S S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 120°W i80° 120°W 60°W 0°w 60°E 120°E 180° 0°W 120°W 0°w 120°E 180° 180° 60°W 60°E 120°E 60°W 60°E Predicted wend speed Predicted wend speed 36 hours later 72 hours later N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 60°N 60°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0m N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0m 0° 30°S S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 90°180 90°80° 180° 120°W 0°w 120°E 120°W 0°w 180° 60°W 60°E 60°E 120°E 60°W 72 hours later True U10 True U10 36 hours later True U10 Initial condition (b) N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 60°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5 30°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='O0E 30°N 0° 。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0° 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 30°S 30°S 30°S 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 90°180 90°180.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 60°W 120°W 0°W 60°E 120°E 0°W 120°E 60° o°w 180° 120°W 60°W 60°E 180° 120°W 60°E 120°E 180° Predicted U10 72 hours later Predicted U10 36 hours later N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 60°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='Om N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0m 0° 0° S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0m 30°S S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 90°180 0°w 120°W o°W 60°E 180° 120°W 60°W 60°E 180° 60°W 120°E 120°E True V10 36 hours later True V10 72 hours later True V10 Initial condition (c) N。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 60°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 30°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 30°N 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5 。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0° S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='08 30°S 30°S S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 60°S S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 90°80 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 0°W 120°W 60°W 120°W 120°W 0°W 60°E 0°W 60°E 120°E 180° 60°W 120°E 180° 60°W 60°E 120°E 180° Predicted V10 72 hours later 36 hours later Predicted V10 N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 L N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06 60°N N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 30°N N。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0m 0° 0° S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 30°S- S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='09 60°S 90°180° 120°W o°w 180° 120°W 60°W 0°w 60°E 120°E 60°W 60°E 120°E 180°Figure 5: Time-dependent prediction accuracy (measured by anomaly correlation coefficient, ACC) of ViT-KNO on 20 variables of ERA5 data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The colored area denotes the interval of accuracy whose boundaries are fractiles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The dashed line denotes the average accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 1 import koopmanlab as kp 2 3 ViT_KNO = model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' koopman_vit (decoder = "Conv2d", resolution =(1440 , 720) , patch_size =(2, 2), in_chans =20, out_chans =20, head_num =20, embed_dim =768 , depth = 16, parallel = True , high_freq = True , device=device) 4 5 ViT_KNO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='compile () 6 7 ## Parameter definitions : 8 # resolution: the spatial resolution of input data 9 # patch_size: the size of each patch (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', token) 10 # in_chans: the number of target variables in the data set 11 # out_chans: the number of predicted variables by ViT -KNO , which is usually same as in_chans 12 # head_num: the number of heads 13 # embed_dim: the embeding dimension denoted by l in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (72) 14 # depth: the depth of each head 15 # parallel: if parallel design is applied 16 # high_freq: if high -frequency information complement is applied 17 # device: if CPU or GPU is used for computation Please note that there exist some detailed model parameters that are not covered by the above codes because they are highly coupled during computation or less important in our theoretical derivations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Users are suggested to adjust them after loading the source code of ViT-KNO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The ViT-KNO sub-family: Validation To validate the proposed ViT-KNO sub-family, we imple- ment a large-scale experiment on ERA5, one of the largest high- resolution data sets of global-scale multi-variate climate fields [32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' This data set has been extensively applied in weather fore- casting tasks (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', see FourCastNet [23, 24]), ensuring the re- producibility and comparability of our results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Twenty important climate variables are considered in our re- search, including mean large-scale precipitation (MSLP), rel- ative humidity with 500 hPa (R500), relative humidity with 850 hPa (R850), surface pressure (SP), 2m temperature (T2M), temperature with 500 hPa (T500), temperature with 850 hPa (T850), total column water vapour (TCWV), the 10m u- component of wind (U10), the u-component of wind with 500 hPa (U500), the u-component of wind with 850 hPa (U850), 14 (b) (a) (d) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 (c) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='07 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 - 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 IIIIIIII 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 R500 MSLP R850 SP 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0叶 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0时 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0+ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 48 48 96 144 192 48 96 144 192 96 144 192 48 96 144 192 Prediction duration length Prediction duration length Prediction duration length Prediction duration length (e) 1 (f) (g) (h) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 IIIIIJ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 TCW T2M T500 T850 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0% 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0叶 48 96 144 192 96 144 192 48 96 144 192 48 96 144 192 Prediction duration length Prediction duration length Prediction duration length Prediction duration length (k) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 IIIII 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 : 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 U10 U500 U850 U1000 o.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0% 400 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 192 48 96 144 192 48 96 144 192 48 96 144 192 48 96 144 Prediction duration length Prediction duration length Prediction duration length Prediction duration length (m) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 t (n) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 IIIIIII 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 V1000 U10 V500 V850 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0g %00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 + 48 48 96 144 192 96 144 192 48 96 144 192 48 96 144 192 Prediction duration length Prediction duration length Prediction duration length Prediction duration length (q) (r) (s) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0- 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='2 Z50 Z500 Z850 Z1000 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0+ o.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='0叶 48 144 48 96 144 48 96 144 48 96 144 192 96 192 192 192 Prediction duration length Prediction duration length Prediction duration length Prediction duration lengththe u-component of wind with 1000 hPa (U1000),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the 10m v- component of wind (V10),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the v-component of wind with 500 hPa (V500),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the v-component of wind with 850 hPa (V850),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the v-component of wind with 1000 hPa (V1000),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the geopotential with 50 hPa (Z50),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the geopotential with 500 hPa (Z500),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' the geopotential with 850 hPa (Z850),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' and the geopotential with 1000 hPa (Z1000).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' We test a ViT-KNO whose decoder is a multi-layer percep- tron in a long-term prediction task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Given the samples of initial conditions, the ViT-KNO is required to predict the future states of all 20 climate variables after t ∈ {6, 12, 18, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' , 192} hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The training data set includes the samples recorded from 1979 to 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The validation data set includes the samples recorded during 2016 and 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The test data set includes the samples recorded in 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The spatial resolution of all samples is set as 1440 × 720.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' All data is pre-processed in a standard manner following Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [32, 23, 24], where a Z-transform is applied to normalize all variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The training of our ViT-KNO is im- plemented in a multi-GPU environment with 128 × 16 GBs in total.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The actual memory cost of training is 1250.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='56 GBs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The testing of trained ViT-KNO is implemented in a single 16-GB GPU environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Key model settings are summarized below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The batch size is set as 128, the learning rate is 5 × 10−4 and updated by a cosine annealing approach, the patch size is set as 8 × 8, the number of heads is 8, the depth of heads is 12, the embedded dimension l in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' (72) is 768, the relative weights of prediction and re- construction in the loss function are λp = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='9 and λr = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='1, and the number of kept low-frequency modes after the fast Fourier transform is 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The defined model has 74691840 parameters in total.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' All 20 climate variables are learned and predicted to- gether rather than respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Please note that the information of land-form is not provided to the model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The model is re- quired to learn climate variables with no additional information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The maximum number of available epochs for training is set as 300 to explore when the model can converge, which costs about 92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='5 hours in our environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The convergence is observed to emerge after ≃ 150 epochs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Therefore, users can consider a 150-epoch training in the application, which costs about 2 days under the same hardware condition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' There is no additional trick applied during training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 4, we visualize several instances of the predicted cli- mate fields during testing, accompanied by corresponding true values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' High consistency can be seen between these ground truths and their predicted counterparts derived by ViT-KNO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Quantitatively, the prediction accuracy of each climate variable during testing is measured by anomaly correlation coefficient (ACC) in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' According to the same prediction task re- ported by Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [23, 24], the trained ViT-KNO outperforms the baseline state-of-the-art deep learning models for weather fore- casting proposed by Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [65] significantly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Compared with the FourCastNet trained with multiple numerical tricks (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', multi-stage training with large memory consumption) [23, 24], ViT-KNO achieves a similar accuracy during testing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Limited by computing resources, we are unable to precisely compare between ViT-KNO and FourCastNet under the same hardware and training condition yet (the FourCastNet is reported to be trained on 3808 NVIDIA A100 GPUs with numerous compu- tational optimization [23]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' More detailed comparisons may be considered in future studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' We suggest that ViT-KNO has the potential to become a competitive alternative of FourCast- Net.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Moreover, the time cost of a single time of prediction by ViT-KNO is observed as ≃ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='02768695354 seconds in a sin- gle 16-GB GPU.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Compared with the classic numerical weather forecasting systems (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the ECMWF Integrated Forecasting System) whose prediction inevitably requires a multi-GPU en- vironment (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', more than 1000 NVIDIA Selene nodes where each node consists of 8 NVIDIA A100 GPUs) [66, 67, 68], ViT- KNO is orders of magnitude faster in the application (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', the Integrated Forecasting System L91 18 km model is expected to cost about 9840 node seconds for prediction on a NVIDIA Se- lene node [68]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Therefore, our ViT-KNO has the potential to serve as a unit in ensemble weather forecasting frameworks to realize an efficient prediction of global weather.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Conclusion In this paper, we have presented KoopmanLab, an efficient module of Koopman neural operator family for solving par- tial differential equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' The included models in this mod- ule, such as the compact KNO sub-family and the ViT-KNO sub-family, are provided with mathematical foundations, com- putational designs, and validations in solving concrete PDEs or predicting intricate dynamic system governed by unknown cou- pled PDEs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' All models are suggested as competitive with other state-of-the-art approaches in corresponding tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Compared with classic numerical and neural-network-based PDE solvers, the proposed KNO variants can achieve significant acceleration, more robust mesh-independence, higher generalization capac- ity on changed conditions, more flexibility in characterizing la- tent PDEs with unknown forms, and a better balance between accuracy and efficiency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Therefore, we suggest the potential of KoopmanLab be applied in diverse down-stream tasks related with PDE solving.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Users can download this module via 1 pip install koopmanlab or 1 git clone https :// github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='com/Koopman -Laboratory/ KoopmanLab .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='git 2 cd KoopmanLab 3 pip install e .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Several important questions remain for future studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' First, one may consider more specialized computational optimization of models in KoopmanLab (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', consider multi-stage training as suggested in Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [23, 24] or multi-objective balancing by Pareto theory [69, 70]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Second, one can explore a more de- tailed comparison between the ViT-KNO sub-family and Four- CastNet under the equivalent hardware and training conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Third, one can analyze the errors of our models caused by the potential continuous spectrum of the Koopman operator or the absence of ergodic property in real cases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 15 Acknowledgements This project is supported by the Artificial and General Intel- ligence Research Program of Guo Qiang Research Institute at Tsinghua University (2020GQG1017) as well as the Tsinghua University Initiative Scientific Research Program.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' References [1] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Gockenbach, Partial differential equations: analytical and numeri- cal methods, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 122, Siam, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [2] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tanabe, Functional analytic methods for partial differential equations, CRC Press, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [3] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Debnath, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Debnath, Nonlinear partial differential equations for sci- entists and engineers, Springer, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [4] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mattheij, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Rienstra, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Boonkkamp, Partial differential equations: modeling, analysis, computation, SIAM, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [5] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kovachki, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Azizzadenesheli, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Liu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bhattacharya, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Stu- art, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, Neural operator: Graph kernel network for partial differential equations, arXiv preprint arXiv:2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='03485 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [6] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Reddy, Introduction to the finite element method, McGraw-Hill Ed- ucation, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [7] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lipnikov, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Manzini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Shashkov, Mimetic finite difference method, Journal of Computational Physics 257 (2014) 1163–1227.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [8] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Eymard, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Gallou¨et, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Herbin, Finite volume methods, Handbook of numerical analysis 7 (2000) 713–1018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [9] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Guo, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Iorio, Convolutional neural networks for steady flow approximation, in: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, 2016, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 481–490.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [10] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Zhu, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Zabaras, Bayesian deep convolutional encoder–decoder net- works for surrogate modeling and uncertainty quantification, Journal of Computational Physics 366 (2018) 415–447.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [11] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bhatnagar, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Afshar, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Pan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Duraisamy, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kaushik, Prediction of aerodynamic flow fields using convolutional neural networks, Computa- tional Mechanics 64 (2) (2019) 525–545.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [12] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Yu, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', The deep ritz method: a deep learning-based numerical algo- rithm for solving variational problems, Communications in Mathematics and Statistics 6 (1) (2018) 1–12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [13] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Raissi, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Perdikaris, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Karniadakis, Physics-informed neural net- works: A deep learning framework for solving forward and inverse prob- lems involving nonlinear partial differential equations, Journal of Com- putational physics 378 (2019) 686–707.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [14] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bar, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Sochen, Unsupervised deep learning algorithm for pde-based forward and inverse problems, arXiv preprint arXiv:1904.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05417 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [15] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Pan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Duraisamy, Physics-informed probabilistic learning of linear embeddings of nonlinear dynamics with guaranteed stability, SIAM Jour- nal on Applied Dynamical Systems 19 (1) (2020) 480–509.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [16] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lu, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Jin, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Karniadakis, Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approxima- tion theorem of operators, arXiv preprint arXiv:1910.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='03193 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [17] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bhattacharya, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Hosseini, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kovachki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Stuart, Model reduction and neural networks for parametric pdes, arXiv preprint arXiv:2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='03180 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [18] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Nelsen, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Stuart, The random feature model for input-output maps between banach spaces, SIAM Journal on Scientific Computing 43 (5) (2021) A3212–A3243.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [19] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kovachki, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Azizzadenesheli, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Liu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bhattacharya, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Stu- art, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, Fourier neural operator for parametric partial differ- ential equations, arXiv preprint arXiv:2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='08895 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [20] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kovachki, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lanthaler, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mishra, On universal approximation and error bounds for fourier neural operators, Journal of Machine Learning Research 22 (2021) Art–No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [21] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Huang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Liu, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, Fourier neural operator with learned deformations for pdes on general geometries, arXiv preprint arXiv:2207.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05209 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [22] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Guibas, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mardani, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tao, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Catanzaro, Effi- cient token mixing for transformers via adaptive fourier neural operators, in: International Conference on Learning Representations, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [23] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Pathak, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Subramanian, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Harrington, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Raja, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Chattopadhyay, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mardani, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kurth, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Hall, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Azizzadenesheli, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Fourcast- net: A global data-driven high-resolution weather model using adaptive fourier neural operators, arXiv preprint arXiv:2202.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='11214 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [24] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kurth, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Subramanian, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Harrington, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Pathak, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mardani, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Hall, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Miele, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kashinath, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, Fourcastnet: Accelerating global high-resolution weather forecasting using adaptive fourier neural operators, arXiv preprint arXiv:2208.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='05419 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [25] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tadmor, A review of numerical methods for nonlinear partial differ- ential equations, Bulletin of the American Mathematical Society 49 (4) (2012) 507–554.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [26] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Stensrud, Parameterization schemes: keys to understanding numeri- cal weather prediction models, Cambridge University Press, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [27] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bauer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Thorpe, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunet, The quiet revolution of numerical weather prediction, Nature 525 (7567) (2015) 47–55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [28] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunton, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Budisic, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kaiser, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kutz, Modern koopman theory for dynamical systems, SIAM Review 64 (2) (2022) 229–340.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [29] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Cao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Wang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Duan, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Zhang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Zhu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Huang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tong, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Xu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bai, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tong, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', Spectral temporal graph neural network for multi- variate time-series forecasting, Advances in neural information process- ing systems 33 (2020) 17766–17778.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [30] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Aminian, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Suarez, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Aminian, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Walz, Forecasting economic data with neural networks, Computational Economics 28 (1) (2006) 71– 88.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [31] Anonymous, Koopman neural operator for learning non-linear partial dif- ferential equations, in: Submitted to The Eleventh International Confer- ence on Learning Representations, 2023, under review.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' URL https://openreview.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='net/forum?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='id=kciCTrtbzVl [32] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Hersbach, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bell, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Berrisford, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Hirahara, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Hor´anyi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mu˜noz- Sabater, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Nicolas, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Peubey, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Radu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Schepers, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', The era5 global reanalysis, Quarterly Journal of the Royal Meteorological Society 146 (730) (2020) 1999–2049.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [33] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Taylor, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kutz, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Morgan, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Nelson, Dynamic mode decom- position for plasma diagnostics and validation, Review of Scientific In- struments 89 (5) (2018) 053501.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [34] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Rowley, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mezi´c, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bagheri, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Schlatter, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Henningson, Spec- tral analysis of nonlinear flows, Journal of fluid mechanics 641 (2009) 115–127.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [35] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Abraham, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Murphey, Active learning of dynamics for data-driven control using koopman operators, IEEE Transactions on Robotics 35 (5) (2019) 1071–1083.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [36] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunton, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Johnson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Ojemann, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kutz, Extracting spatial–temporal coherent patterns in large-scale neural recordings us- ing dynamic mode decomposition, Journal of neuroscience methods 258 (2016) 1–15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [37] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Koopman, Hamiltonian systems and transformation in hilbert space, Proceedings of the National Academy of Sciences 17 (5) (1931) 315–318.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [38] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Abraham, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Marsden, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Ratiu, Manifolds, tensor analysis, and applications, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 75, Springer Science & Business Media, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [39] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Chicone, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Latushkin, Evolution semigroups in dynamical systems and differential equations, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 70, American Mathematical Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', 1999.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [40] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lax, Integrals of nonlinear equations of evolution and solitary waves, Communications on pure and applied mathematics 21 (5) (1968) 467–490.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [41] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Parker, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Page, Koopman analysis of isolated fronts and solitons, SIAM Journal on Applied Dynamical Systems 19 (4) (2020) 2803–2828.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [42] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Nakao, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mezi´c, Spectral analysis of the koopman operator for partial differential equations, Chaos: An Interdisciplinary Journal of Nonlinear Science 30 (11) (2020) 113131.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [43] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Gin, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lusch, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunton, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kutz, Deep learning models for global coordinate transformations that linearise pdes, European Journal of Applied Mathematics 32 (3) (2021) 515–539.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [44] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Page, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kerswell, Koopman analysis of burgers equation, Physical Review Fluids 3 (7) (2018) 071901.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [45] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Takeishi, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kawahara, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Yairi, Learning koopman invariant sub- spaces for dynamic mode decomposition, Advances in Neural Informa- tion Processing Systems 30 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [46] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Azencot, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Erichson, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lin, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mahoney, Forecasting sequential data using consistent koopman autoencoders, in: International Confer- ence on Machine Learning, PMLR, 2020, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 475–485.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [47] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Otto, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Rowley, Linearly recurrent autoencoder networks for 16 learning dynamics, SIAM Journal on Applied Dynamical Systems 18 (1) (2019) 558–593.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [48] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Alford-Lago, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Curtis, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Ihler, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Issan, Deep learning en- hanced dynamic mode decomposition, Chaos: An Interdisciplinary Jour- nal of Nonlinear Science 32 (3) (2022) 033116.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [49] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lusch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kutz, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunton, Deep learning for universal linear embeddings of nonlinear dynamics, Nature communications 9 (1) (2018) 1–10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [50] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Saad, Numerical methods for large eigenvalue problems: revised edi- tion, SIAM, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [51] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Arbabi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mezic, Ergodic theory, dynamic mode decomposition, and computation of spectral properties of the koopman operator, SIAM Jour- nal on Applied Dynamical Systems 16 (4) (2017) 2096–2126.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [52] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' ˇCrnjari´c-ˇZic, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Ma´ceˇsi´c, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mezi´c, Koopman operator spectrum for random dynamical systems, Journal of Nonlinear Science 30 (5) (2020) 2007–2056.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [53] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunton, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Brunton, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Proctor, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kaiser, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kutz, Chaos as an intermittently forced linear system, Nature communications 8 (1) (2017) 1–9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [54] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Korda, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mezi´c, On convergence of extended dynamic mode decom- position to the koopman operator, Journal of Nonlinear Science 28 (2) (2018) 687–710.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [55] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Jiang, Reduced-order modeling for koopman operators of nonautonomous dynamic systems in multiscale media, arXiv preprint arXiv:2204.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='13180 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [56] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Cornfeld, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Fomin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Sinai, Ergodic theory, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 245, Springer Science & Business Media, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [57] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Park, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kim, How do vision transformers work?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', arXiv preprint arXiv:2202.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06709 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [58] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Wang, Exact solutions of the steady-state navier-stokes equations, An- nual Review of Fluid Mechanics 23 (1) (1991) 159–177.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [59] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Benton, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Platzman, A table of solutions of the one-dimensional burgers equation, Quarterly of Applied Mathematics 30 (2) (1972) 195– 212.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [60] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kovachki, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Azizzadenesheli, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Liu, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Stuart, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bhat- tacharya, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, Multipole graph neural operator for parametric partial differential equations, Advances in Neural Information Processing Systems 33 (2020) 6755–6766.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [61] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Dosovitskiy, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Beyer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Kolesnikov, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Weissenborn, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Zhai, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Un- terthiner, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Dehghani, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Minderer, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Heigold, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Gelly, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', An im- age is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv:2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='11929 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [62] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Guibas, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mardani, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Li, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tao, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Anandkumar, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Catanzaro, Adaptive fourier neural operators: Efficient token mixers for transform- ers, arXiv preprint arXiv:2111.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='13587 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [63] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Radford, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Metz, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Chintala, Unsupervised representation learning with deep convolutional generative adversarial networks, arXiv preprint arXiv:1511.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content='06434 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [64] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Tibshirani, Regression shrinkage and selection via the lasso, Journal of the Royal Statistical Society: Series B (Methodological) 58 (1) (1996) 267–288.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [65] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Weyn, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Durran, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Caruana, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Cresswell-Clay, Sub-seasonal forecasting with a large ensemble of deep-learning weather prediction models, Journal of Advances in Modeling Earth Systems 13 (7) (2021) e2021MS002502.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [66] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Roberts, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Senan, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Molteni, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Boussetta, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Mayer, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Keeley, Climate model configurations of the ecmwf integrated forecasting system (ecmwf-ifs cycle 43r1) for highresmip, Geoscientific model development 11 (9) (2018) 3681–3712.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [67] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Lopez, A lightning parameterization for the ecmwf integrated forecast- ing system, Monthly Weather Review 144 (9) (2016) 3057–3075.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [68] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bauer, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Quintino, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Wedi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Bonanni, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Chrust, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Deconinck, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Diamantakis, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' D¨uben, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' English, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Flemming, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=', The ecmwf scalability programme: Progress and plans, European Centre for Medium Range Weather Forecasts, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [69] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Sener, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Koltun, Multi-task learning as multi-objective optimization, Advances in neural information processing systems 31 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' [70] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Momma, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Dong, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' Liu, A multi-objective/multi-task learning framework induced by pareto stationarity, in: International Conference on Machine Learning, PMLR, 2022, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 15895–15907.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} +page_content=' 17' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ZNAzT4oBgHgl3EQfKvv0/content/2301.01104v1.pdf'} diff --git a/_tAzT4oBgHgl3EQfS_vY/vector_store/index.faiss b/_tAzT4oBgHgl3EQfS_vY/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..f2d1cd1a2a73bb489e4ee016133e800098236b75 --- /dev/null +++ b/_tAzT4oBgHgl3EQfS_vY/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2dc4fdbf23ae79d553de8190c7185fe7893915dcffb8900c4b6ca12e022adfe0 +size 4653101 diff --git a/b9E0T4oBgHgl3EQf4wKo/content/tmp_files/2301.02743v1.pdf.txt b/b9E0T4oBgHgl3EQf4wKo/content/tmp_files/2301.02743v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..b760c49d20a98dac50ed56e27920e7f7789c7c61 --- /dev/null +++ b/b9E0T4oBgHgl3EQf4wKo/content/tmp_files/2301.02743v1.pdf.txt @@ -0,0 +1,582 @@ +1 + +Direct-Writing Atom-by-Atom +Ondrej Dyck1, Andrew R. Lupini1, Stephen Jesse1 +1 Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN + +Abstract +Direct-write processes enable the alteration or deposition of materials in a continuous, directable, +sequential fashion. In this work we demonstrate an electron beam direct-write process in an +aberration-corrected scanning transmission electron microscope. This process has several +fundamental differences from conventional electron beam induced deposition techniques, where +the electron beam dissociates precursor gases into chemically reactive products that bond to a +substrate. Here, we use elemental tin (Sn) as a precursor and employ a different mechanism to +facilitate deposition. The atomic-sized electron beam is used to generate chemically reactive point +defects at desired locations in a graphene substrate. Temperature control of the sample is used to +enable the precursor atoms to migrate across the surface and bond to the defect sites thereby +enabling atom-by-atom direct-writing. + +2 + +Introduction +The scanning transmission electron microscope (STEM) has a long legacy as an indispensable +materials characterization platform. In recent years this picture has begun to expand to include not +only characterization but also manipulation.1–3 From this perspective the STEM can be viewed as +a fabrication platform capable of addressing individual atoms with the highly focused electron +beam (e-beam) and altering materials atom-by-atom. Such capabilities hearken back to Feynman’s +famous quote about putting every atom where we want.4 One would like to be able to build +structures from the atom up, designed for their electronic, optical, chemical, or structural properties. +Quantum qubits, quantum sensors, and perhaps even full computing circuits could be tailored with +atomic specificity to elicit the emergent quantum properties of the materials. Creating structures +with atomic specificity would throw open the doors for experiments developing fundamental +understanding of physics, chemistry, and materials science at the atomic level. Realizing such a +vision, however, is a more difficult task. +Observations of e-beam induced sample alterations date back to some of the earliest days of +electron microscopy. Every electron microscopist is familiar with e-beam damage5–7 and +hydrocarbon deposition. Mitigation of these unintentional processes has been a key enabling factor +for the (S)TEM to excel at materials characterization. Lower accelerating voltages,8,9 enhanced by +aberration correction,10 have been effective at reducing knock-on damage. With unintended +processes at least partially under control, we can begin to isolate variables to examine their effects +on the specimen with a view toward reproducible atomistic transformations. +To simplify matters further, here we work with 2D materials where, instead of looking at projected +columns of atoms (as in a 3D crystal), each atom can be uniquely identified in the image. The most +robust and well-studied 2D material is graphene and many studies have been carried out to find +ways to clean and image contamination-free graphene.11–24 With these advancements, which act to +mitigate major alterations to the specimen, minor alterations at the atomic scale can be observed +and studied. Some examples of these atomic scale dynamics are shown in Figure 1(a)-(d): the +spinning of a Si molecular rotor,25 the dynamic rearrangement of a Si6 cluster embedded in a +graphene nanopore,26 the movement of a 3-fold coordinated Si dopant in graphene, and vacancy +diffusion in graphene.27 In these examples the major focus was to observe and understand the e- + +3 + +beam driven effects. The transition to top-down control over atomic movement stemmed from the +realization that the position of the e-beam could influence the direction of motion of the 3-fold +coordinated Si atoms. The first demonstration of precisely directed movement of a single atom +using an e-beam is shown in Figure 1(e) where a 3-fold coordinated Si atom was moved seven steps +through a graphene lattice.28 This result was independently replicated shortly afterward29 and +extended to longer distances,30,31 shown in Figure 1(f) and (g). The first e-beam assembly of a +primitive structure, a Si dimer, is shown in Figure 1(h).30 Controlled movement of a Si atom through +the wall of a single walled carbon nanotube is shown in Figure 1(i)32 and intentional rotation of a Si +trimer, as well as the reversible conversion to a tetrimer is shown in Figure 1(j).30 + +Figure 1 Select examples of observed and controlled atomic dynamics in graphene. (a) Observation of Si trimer +rotation. (b) Observation of reversible transformation of a Si6 cluster embedded in a graphene nanopore. (c) +Observation of movement of 3-fold coordinated Si dopant in graphene. (d) Observation of vacancy diffusion in +graphene. (e) Controlled movement of 3-fold coordinated Si dopant in graphene. (f) and (g) Controlled movement of +3-fold coordinated Si atom over an extended distance. (h) Controlled assembly of a Si dimer in graphene. (i) Controlled +movement of Si atom in a single walled carbon nanotube. (j) Controlled rotation of a Si trimer and conversion to +tetrimer. (a) Adapted from reference [25]. Reuse is permitted per the United States Department of Energy public access +plan http://energy.gov/downloads/doe-public-access-plan. (b) Adapted with permission from reference [26]. (c) +Reproduced with permission from reference [33] https://creativecommons.org/licenses/by/3.0/. (d) Reproduced with +permission from reference [27] https://creativecommons.org/licenses/by/3.0/. (e) Adapted with permission from +reference [34]. (f) Reproduced with permission from reference [30]. (g) Reproduced with permission from reference [31] +https://pubs.acs.org/page/policy/authorchoice_ccby_termsofuse.html. (h) Adapted with permission from reference [30]. +(i) Reproduced with permission from reference [32] https://creativecommons.org/licenses/by/4.0/. (j) Adapted with +permission from reference [30]. + +Observed atomic dynamics +Controlledatomicdynamics +a +b +e +50 +100 +150 +200 +h +Time (s) +d +Rotation +Rotation +Trimer to tetrimer +Tetrimerto trimer4 + +While there is a certain elegance to the ability to move a single atom through the graphene lattice, +it is not obvious that positioned Si atoms in graphene have applications aside from the development +of these experimental techniques and unraveling the fundamental physics of beam-matter +interactions. Likewise, the Si atom movement typically terminates with the ejection of an +additional carbon atom converting it into the much less mobile 4-fold coordination.31 This +conversion occurs unpredictably, meaning that the distance a 3-fold coordinated Si atom can be +moved is also unpredictable. First principles simulations of the assembly of the dimer shown in +Figure 1(h) indicate that this assembly process is energetically favorable, which facilitates bringing +atoms together. However, this energetic trend to agglomerate further complicates the predictability +of atom movement in the presence of other dopants. Thus, the tailoring of atomic structures in this +way is inherently limited. +Investigations have been undertaken to extend the types of dopant atoms than can be put in +graphene. The main, serendipitous factor that facilitated the studies shown in Figure 1 was that Si +dopant atoms are commonly found in graphene. Moving from serendipity to predictability, it was +shown that the Si atoms in the surface contamination could be introduced into the graphene lattice +by deliberately ejecting C atoms with the e-beam to create attachment sites for the Si atoms.29,30 +Dopant Si atoms could then be reliably produced using the e-beam as a manipulation tool. This +technique was extended and applied generally to many elements where the source materials were +first introduced onto the graphene surface.35–37 Common to each of these demonstrations was the +limitation that dopant insertion was only accomplished within a few nanometers from the parent +nanoparticle and required a two-step procedure where the e-beam was used both for creating +defects in the graphene substrate as well as separating source atoms from the parent nanoparticles. +These and other closely related studies21,28,32,38–46 suggest it might be possible to find appropriate +conditions under which a continuous atom-by-atom direct-writing procedure may be realized. A +direct-write mechanism would enable the creation of longer-range atomic structures without +requiring the step-by-step movement of single atoms through the lattice. Here, we aim to +demonstrate this possibility. +For an atom-by-atom direct-writing procedure to be of practical use we must address several +challenges. Conceptually, a direct-write procedure binds a material to a substrate through some +local chemical or physical influence. This conceptualization glosses over the atomistic details of + +5 + +the process in favor of the more abstract representation that more easily describes the intended +goal. In a similar way, common descriptions of direct-write deposition such as electron beam +induced deposition (EBID) do not spell out the details of the chemical attachment of every atom +to the substrate. However, as we scale down a writing process, eventually we must confront the +atomistic nature of the world and begin to address the fact that “binding a material to a substrate” +is a vague statement. If we are to write with individual atoms, we must be specific about what we +mean. Here, we mean that atoms, differing elementally from the substrate, form strong chemical +bonds to defect sites introduced in the substrate through a localized influence. With this +formulation we can begin to specify the requirements for writing at the atomic scale. We must have +source materials (atoms) that are delivered to the region of interest. We must have a highly +localized influence on the sample that is able to create chemically reactive sites with which the +source atoms can bond. These bonds need to be strong enough to prevent spontaneous thermal +diffusion over long time spans. The definition of “long”, clearly, depends on the application but +here we envision stability for days, ideally years, at ambient temperatures. Critically, we must have +a clean surface on which to write, in order to provide a spatially uniform environment that enhances +predictable control over the process. +Methods +To satisfy these constraints we use a suspended single layer graphene substrate, operate our +microscope above the knock-on threshold for graphene, and thermally deliver Sn atoms to the +damaged regions in situ. Atmospheric pressure chemical vapor deposition (AP-CVD) was used to +grow graphene on Cu foil.47 Poly(methyl methacrylate) (PMMA) was spin coated over the +graphene for protection. A wet transfer29 was used to transfer the graphene to a ProtochipsTM heater +chip substrate and Sn was evaporated onto the graphene surface in a vacuum chamber using a Sn +coated W filament. The graphene was pre-cleaned using a rapid thermal ramp to 1200 °C23 and the +Sn evaporation was carried out while the sample temperature was maintained at 500 °C. +A Nion UltraSTEM 200 operated at an accelerating voltage of 100 kV and nominal beam current +of 17 pA was used to examine the sample. Prior to examination, the sample, holder cartridge, and +magazine were baked in vacuum for eight hours at 160 °C. After introduction into the pole piece, + +6 + +the sample was again heated to 1200 °C to remove residual contaminants and then cooled to 300 °C +for further examination. +Experimental +Sample overview +Figure 2(a) and (b) show two high angle annular dark field (HAADF) images of the graphene and +Sn nanoparticles at 300 °C. Much of the sample was multilayer and the Sn nanoparticles attached +predominantly at step edges. We note the presence of single Sn atoms decorating the step edges +and various other locations. Figure 2(c) shows an example EELS spectrum acquired from one of the +nanoparticles confirming that they are made of Sn. + +Figure 2 Sample overview illustrating Sn nanoparticles attached to graphene. (a), (b) representative HAADF +images of Sn nanoparticles attached to the graphene surface particularly at step edges. (c) Sn M4,5 EELS edge acquired +on one of the nanoparticles confirming they are composed of Sn. + +Beam Dragging +Following the “beam dragging” strategy laid out in previous work,30,35–37 Sn atoms were inserted +into the graphene lattice with the substrate maintained at 300 °C. The e-beam was positioned on +the source Sn particle and manually moved from the particle to the graphene, acting to both sputter +Sn atoms from the parent nanoparticle and generate point defects in the graphene lattice for the Sn +atoms to occupy. This procedure is depicted schematically in Figure 3(a)-(d). In the initial + +a +b +c +Sn M4.5 EELS Edge +20000 +Single Layer +17500 +.u. +15000 +12500 +Snnanoparticles +Sn nanoparticles +Counts +10000 +7500 +Single Layer +5000 +2500 +Bilayer +Bilayer +450500550600650 +700 +Energy loss [eV] +5nm +Trilayer +Trilayer7 + +configuration, Figure 3(a), the e-beam is positioned on a nanoparticle of source material sitting on +the graphene substrate. The e-beam is then moved across the substrate, depicted in Figure 3(b). This +procedure generates vacancies in the graphene through direct ejection of lattice carbon atoms. The +e-beam is then moved back to the source material and used to sputter atoms from the parent +nanoparticle, Figure 3(c). The freed atoms can diffuse away from the nanoparticle across the +graphene surface. There is some probability for vacancy diffusion and lattice restructuring; a +detailed study and dynamic calculations are given in reference 48. Figure 3(c) illustrates an example +where two vacancies have merged forming a reconstructed divacancy. This structure does not have +any dangling carbon bonds, which suggests that it (and other similar defect structures) will be less +attractive for a diffusing adatom. Figure 3(d) depicts the capture of a diffusing atom by the remaining +vacancy. +The generation of vacancies, sputtering of source atoms, and diffusion of vacancies and atoms are +all stochastic processes, which means they are not guaranteed to occur. Typically, the e-beam must +be repeatedly moved back and forth and the sample state checked via imaging after some time. +Figure 3(e)-(g) shows a summary of the result of this procedure performed experimentally. Sn atoms +are ejected from the parent nanoparticle and incorporated into the graphene lattice. As was also +true in prior publications,29,30,35–37 this process works across the short distance of a few nanometers +and requires repeated attempts at defect creation and sputtering. This example was carried out with +the substrate held at 300 °C and gives a reference point for the behavior of the graphene/Sn system +which was qualitatively like the behavior of other elements at room temperature.35–37 + +8 + +Figure 3 Example of dopant insertion into single layer graphene. (a) Schematic of e-beam positioned on source +material. (b) e-beam is manually dragged across the substrate creating vacancies. (c) E-beam is repositioned on the +source material and atoms are sputtered off. (d) adatoms can bond to the created vacancy sites. (e) Initial sample state +where a Sn nanoparticle is attached to a step edge overlapping the single layer region. (f) “beam dragging” was +performed in the locations indicated by the arrows resulting in Sn dopant insertion along the beam trajectory. (g) final +state of the sample. + +-beam +Source +Material +d +Captured +Dopant +Vacancies +Ejected +Diffusing +Carbon +Adatoms +Reconstructed +Divacancy +e +Hhm +4nm9 + +It is worth emphasizing that the e-beam in this and prior examples of dopant insertion is performing +two distinct functions: 1) the generation of vacancies/defects in the graphene substrate to which +the source atoms can chemically bond (Figure 3(b)), and 2) the ejection/sputtering of source atoms +from the parent nanoparticle (Figure 3(c) and (d)). This means that the rate of supply of source +atoms, freed from the parent nanoparticle, is dependent on e-beam current and position. More +source atoms could likely be supplied by increasing the e-beam current, however, this also +increases the defect generation rate. Because the defects are generated when the e-beam is +positioned off the nanoparticle, the supply of source atoms is not continuously maintained. It is +likely that these constraints are the main reason why dopant insertion with this method has only +been accomplished within a few nanometers of the parent nanoparticle and why multiple attempts +must be made before dopant insertion is accomplished. +Direct Writing +To enhance our understanding of the processes described above quantitative modeling and +measurement of the availability of source atoms as a function of e-beam current or position, for +example, could be performed. However, the crux of the dilemma is that this process is reliant on +the e-beam for both sputtering source material and the creation of defects. In contrast, e-beam +induced deposition (EBID) does not rely on the e-beam to generate the flow of precursor material, +and ideally the same principle should apply here. We seek appropriate sample conditions where +the source material can be supplied to the e-beam location without the need for e-beam sputtering. +For the system being examined here, the most relevant parameter that can be easily controlled is +the substrate temperature. +To this end, the temperature of the sample was increased in steps to 900 °C, stopping momentarily +to verify the continued presence of the Sn nanoparticles at 600, 700, and 800 °C. At 900 °C the +sample was examined more closely. Figure 4(a) shows an image of a step edge harboring many Sn +atoms. Figure 4(b) shows the same location, acquired five minutes later. The spontaneous +dispersion of Sn atoms along the step edge and the formation of multiple Sn nanoparticles suggests +that the Sn atoms are highly mobile. +One potential interpretation is that Sn atoms are rapidly diffusing across the graphene surface (too +rapidly to image) but are stabilized at the step edge and become imageable. This interpretation + +10 + +suggests that a vacancy created anywhere in the graphene should result in the attachment of a +randomly diffusing Sn atom. To explore this hypothesis, we used the e-beam to scan a small (~0.5 +nm square) area over a random patch of the graphene. After a few seconds, instead of observing +dopant insertion, a hole was created in the graphene indicating that the supply of Sn atoms in that +location is lower than is present at the step edge. The spatial distribution of the supply of Sn atoms +deserves further theoretical investigation, however our goal here was to empirically find conditions +favorable for direct-writing. +Since there seemed to be an ample supply of Sn atoms at the step edge, a different strategy was +employed. A small (~0.5 nm) subscan box was defined and positioned at the step edge, which +provides a small, rapidly updating image of the irradiated location for real-time monitoring. As the +step edge harboring Sn atoms was irradiated, Sn atoms were observed to attach to the underlying +graphene. The subscan box was then manually repositioned to expose more of the pristine +graphene. In this way, Sn atoms were drawn from the step edge and attached to vacancies/defects +in a continuous manner with a single pass. This procedure is illustrated visually in the schematic +in Figure 4(c). +Figure 4(d) shows a HAADF image of the resulting Sn dopant chain drawn in the graphene lattice. +Figure 4(e) shows a magnified view displayed as a smoothed surface plot and artificially colored. +The subscan box was moved as indicated by the arrow. After image acquisition, the process was +continued, as shown in. Figure 4(f). This second line required about seven minutes to complete. +Figure 4(g) shows a magnified surface plot of (f). + +11 + +Figure 4 Demonstrating atom-by-atom direct write using the e-beam at 900 °C. (a) Initial state of a bilayer +graphene step edge at 900 °C. (b) The same region shown in (a) acquired five minutes later. Additional Sn +nanoparticles attached to the step edge indicate a high degree of Sn mobility along the edge, arrowed. Images shown +in the (d)-(g) were acquired near the dashed region on the left. (c) Schematic illustration of the direct writing procedure. +(d) HAADF image of a line of Sn dopant atoms directly written in graphene. (e) Enlarged and labeled surface plot of +the image in (d). (f) The direct write process was continued, starting where the previous writing was halted. (g) Surface +plot of the image shown in (f). Images in (a) and (b) were artificially colored using the “Fire” look up table in Imagej. +Images (e) and (g) were artificially colored and smoothed to reduce noise. +It should be noted that we observe some Sn dopants that attach to the graphene surface outside the +direct-write region. The most likely interpretation is that thermally induced vacancy diffusion +enables vacancy migration from the irradiated region into the pristine graphene48 after which Sn +atoms can bond to the vacancy site. This evidence of Sn atoms diffusing across the graphene +surface combined with the result of forming holes when direct deposition was attempted at random + +Direct WritingProcedure +SnAtomsPopulatingStep Edge +Sn Atoms +Sn Atoms +a +Bilayer +onStepEdge +inGraphene +Graphene +5hm +Irradiated +Sn Aggregating into Nanoparticles +Region +b +Single Layer +Graphene +ProgressionThroughTime +f +5 nm +e +g +Sn Atoms +on Step Edge +Sn Atoms +ManuallyPositioned +in Graphene +Subscan Box +Single Layer +Progression Through Time +Graphene12 + +locations on the graphene suggests that the ratio of the defect generation rate to the adatom +diffusivity (or concentration) plays a significant role in determining the outcome. These +observations point toward fruitful avenues of inquiry for future experiments. +Clearly there are a variety of factors that need to be further understood and controlled to optimize +atom-by-atom direct writing. Previous work in this context highlights the role of mobile carbon +adatoms that spontaneously heal vacancies and larger structural defects.49,50 At the same time, +rapidly diffusing vacancies can markedly change the behavior of graphene at high temperatures.48 +E-beam patterning of corrals24 can influence the influx of hydrocarbon contamination or provide +a trap for long-range vacancy diffusion. The dynamics of these factors are clearly dependent on +the sample temperature, and here we have demonstrated that temperature can also be used to +influence the supply of source atoms. We have shown that a step-edge on the graphene and a +deliberately created dopant chain provide local channels for the migration of source atoms and can +enable continuous direct-writing of nanoscale features. + +Conclusion +We have demonstrated an e-beam directed, continuous atom-by-atom direct-write process on +suspended single layer graphene. This process relies upon the repeated ejection of C atoms from +the substrate lattice as well as a continuous supply of source atoms. The generation of defect sites +is governed by the e-beam accelerating voltage and beam current, here fixed at 100 kV and 17 pA. +The incorporation of source atoms into the defect sites is governed by the rate of supply of source +atoms to the generated vacancies. Substrate temperature, diffusion rates, and the local energetic +landscape play significant roles in this process. We find that sample features, such as the graphene +step edge, or artificially engineered features, such as the direct-write lines of atoms implanted into +the graphene lattice, both serve a role in delivering source atoms to the defect sites. While the role +of temperature might generally be expected, the role of local defects suggests that one could +deliberately tailor the local environment to be favorable for the direct-write operation. +These findings indicate generally that e-beam atom-by-atom direct-write processes are possible +provided that the source material can be supplied at the requisite rate. This suggests that a search + +13 + +for conditions which enable a direct-write process at arbitrary sample locations would be a fruitful +future direction of inquiry. Similar processes may enable atom-by-atom direct-writing on various +other 2D materials or possibly inside 3D materials. +Acknowledgment +The authors would like to gratefully acknowledge the assistance of Zachary Gosser in the Sn +evaporation process. +This work was supported by the U.S. Department of Energy, Office of Science, Basic Energy +Sciences, Materials Sciences and Engineering Division (O.D. A.R.L., S.J.), and was performed at +the Center for Nanophase Materials Sciences (CNMS), a U.S. Department of Energy, Office of +Science User Facility. +References +(1) +Dyck, O.; Ziatdinov, M.; Lingerfelt, D. B.; Unocic, R. R.; Hudak, B. M.; Lupini, A. R.; Jesse, S.; Kalinin, +S. V. Atom-by-Atom Fabrication with Electron Beams. Nat. Rev. Mater. 2019, 4 (7), 497–507. +https://doi.org/10.1038/s41578-019-0118-z. +(2) +Kalinin, S. V.; Borisevich, A.; Jesse, S. Fire up the Atom Forge. Nature 2016, 539 (7630), 485–487. +(3) +Dyck, O.; Jesse, S.; Kalinin, S. V. A Self-Driving Microscope and the Atomic Forge. MRS Bull. 2019, 44 +(9), 669–670. https://doi.org/10.1557/mrs.2019.211. +(4) +Feynman, R. P. There’s Plenty of Room at the Bottom. Eng. Sci. 1960, 23 (5), 22–36. +(5) +Stenn, K.; Bahr, G. F. Specimen Damage Caused by the Beam of the Transmission Electron Microscope, a +Correlative Reconsideration. J. Ultrastruct. Res. 1970, 31 (5), 526–550. https://doi.org/10.1016/S0022- +5320(70)90167-X. +(6) +Nan Jiang. Electron Beam Damage in Oxides: A Review. Rep. Prog. Phys. 2016, 79 (1), 016501. +(7) +Egerton, R. F.; Wang, F.; Crozier, P. A. Beam-Induced Damage to Thin Specimens in an Intense Electron +Probe. Microsc. Microanal. 2005, 12 (1), 65–71. https://doi.org/10.1017/S1431927606060065. +(8) +Krivanek, O. L.; Dellby, N.; Murfitt, M. F.; Chisholm, M. F.; Pennycook, T. J.; Suenaga, K.; Nicolosi, V. +Gentle STEM: ADF Imaging and EELS at Low Primary Energies. Ultramicroscopy 2010, 110 (8), 935–945. +https://doi.org/10.1016/j.ultramic.2010.02.007. +(9) +Ondrej L. Krivanek; Wu Zhou; Matthew F. Chisholm; Juan Carlos Idrobo; Tracy C. Lovejoy; Quentin M. +Ramasse; Niklas Dellby. Gentle STEM of Single Atoms: Low KeV Imaging and Analysis at Ultimate Detection +Limits. In Low Voltage Electron Microscopy; S. Brooks, D. C. Bell, N. Erdman, Eds.; 2012. +(10) +Hawkes, P. W. Advances in Imaging and Electron Physics: Aberration-Corrected Electron Microscopy; +Elsevier Science, 2009. +(11) +A. M. Goossens; V. E. Calado; A. Barreiro; K. Watanabe; T. Taniguchi; L. M. K. Vandersypen. Mechanical +Cleaning of Graphene. Appl. Phys. Lett. 2012, 100 (7), 073110. https://doi.org/10.1063/1.3685504. +(12) +C. A. Joiner; T. Roy; Z. R. Hesabi; B. Chakrabarti; E. M. Vogel. Cleaning Graphene with a Titanium +Sacrificial Layer. Appl. Phys. Lett. 2014, 104 (22), 223109. https://doi.org/10.1063/1.4881886. +(13) +Gerardo Algara-Siller; Ossi Lehtinen; Andrey Turchanin; Ute Kaiser. Dry-Cleaning of Graphene. Appl. +Phys. Lett. 2014, 104 (15), 153115. https://doi.org/10.1063/1.4871997. +(14) +J. Moser; A. Barreiro; A. Bachtold. Current-Induced Cleaning of Graphene. Appl. Phys. Lett. 2007, 91 (16), +163513. https://doi.org/10.1063/1.2789673. +(15) +Jean-Nicolas Longchamp; Conrad Escher; Hans-Werner Fink. Ultraclean Freestanding Graphene by + +14 + +Platinum-Metal Catalysis. J. Vac. Sci. Technol. B 2013, 31 (2), 020605. https://doi.org/10.1116/1.4793746. +(16) +Liang, X.; Sperling, B. A.; Calizo, I.; Cheng, G.; Hacker, C. A.; Zhang, Q.; Obeng, Y.; Yan, K.; Peng, H.; +Li, Q.; Zhu, X.; Yuan, H.; Hight Walker, A. R.; Liu, Z.; Peng, L.; Richter, C. A. Toward Clean and Crackless +Transfer of Graphene. ACS Nano 2011, 5 (11), 9144–9153. https://doi.org/10.1021/nn203377t. +(17) +Lin, Y.-C.; Jin, C.; Lee, J.-C.; Jen, S.-F.; Suenaga, K.; Chiu, P.-W. Clean Transfer of Graphene for Isolation +and Suspension. ACS Nano 2011, 5 (3), 2362–2368. https://doi.org/10.1021/nn200105j. +(18) +Lin, Y.-C.; Lu, C.-C.; Yeh, C.-H.; Jin, C.; Suenaga, K.; Chiu, P.-W. Graphene Annealing: How Clean Can It +Be? Nano Lett. 2012, 12 (1), 414–419. https://doi.org/10.1021/nl203733r. +(19) +Schweizer, P.; Dolle, C.; Dasler, D.; Abellán, G.; Hauke, F.; Hirsch, A.; Spiecker, E. Mechanical Cleaning +of Graphene Using in Situ Electron Microscopy. Nat. Commun. 2020, 11 (1), 1743. https://doi.org/10.1038/s41467- +020-15255-3. +(20) +Tripathi, M.; Mittelberger, A.; Mustonen, K.; Mangler, C.; Kotakoski, J.; Meyer, J. C.; Susi, T. Cleaning +Graphene: Comparing Heat Treatments in Air and in Vacuum. Phys. Status Solidi RRL – Rapid Res. Lett. 2017, 11 +(8), 1700124-n/a. https://doi.org/10.1002/pssr.201700124. +(21) +W. F. van Dorp; X. Zhang; B. L. Feringa; J. B. Wagner; T. W. Hansen; J. Th M De Hosson. Nanometer- +Scale Lithography on Microscopically Clean Graphene. Nanotechnology 2011, 22 (50), 505303. +(22) +Xie, W.; Weng, L.-T.; Ng, K. M.; Chan, C. K.; Chan, C.-M. Clean Graphene Surface through High +Temperature Annealing. Carbon 2015, 94, 740–748. http://dx.doi.org/10.1016/j.carbon.2015.07.046. +(23) +Dyck, O.; Kim, S.; Kalinin, S. V.; Jesse, S. Mitigating E-Beam-Induced Hydrocarbon Deposition on +Graphene for Atomic-Scale Scanning Transmission Electron Microscopy Studies. J. Vac. Sci. Technol. B +Nanotechnol. Microelectron. Mater. Process. Meas. Phenom. 2018, 36 (1), 011801. +https://doi.org/10.1116/1.5003034. +(24) +Dyck, O.; Lupini, A. R.; Rack, P. D.; Fowlkes, J.; Jesse, S. Controlling Hydrocarbon Transport and Electron +Beam Induced Deposition on Single Layer Graphene: Toward Atomic Scale Synthesis in the Scanning Transmission +Electron Microscope. Nano Sel. 2022, 3 (3), 643–654. https://doi.org/10.1002/nano.202100188. +(25) +Yang, Z.; Yin, L.; Lee, J.; Ren, W.; Cheng, H.-M.; Ye, H.; Pantelides, S. T.; Pennycook, S. J.; Chisholm, M. +F. Direct Observation of Atomic Dynamics and Silicon Doping at a Topological Defect in Graphene. Angew. Chem. +2014, 126 (34), 9054–9058. https://doi.org/10.1002/ange.201403382. +(26) +Lee, J.; Zhou, W.; Pennycook, S. J.; Idrobo, J.-C.; Pantelides, S. T. Direct Visualization of Reversible +Dynamics in a Si6 Cluster Embedded in a Graphene Pore. 2013, 4, 1650. https://doi.org/10.1038/ncomms2671 +https://www.nature.com/articles/ncomms2671#supplementary-information. +(27) +Kotakoski, J.; Mangler, C.; Meyer, J. C. Imaging Atomic-Level Random Walk of a Point Defect in +Graphene. Nat. Commun. 2014, 5, 3991. https://doi.org/10.1038/ncomms4991 +https://www.nature.com/articles/ncomms4991#supplementary-information. +(28) +Susi, T.; Kepaptsoglou, D.; Lin, Y.-C.; Ramasse, Q. M.; Meyer, J. C.; Suenaga, K.; Kotakoski, J. Towards +Atomically Precise Manipulation of 2D Nanostructures in the Electron Microscope. 2D Mater. 2017, 4 (4), 042004. +(29) +Dyck, O.; Kim, S.; Kalinin, S. V.; Jesse, S. Placing Single Atoms in Graphene with a Scanning +Transmission Electron Microscope. Appl. Phys. Lett. 2017, 111 (11), 113104. https://doi.org/10.1063/1.4998599. +(30) +Dyck, O.; Kim, S.; Jimenez-Izal, E.; Alexandrova, A. N.; Kalinin, S. V.; Jesse, S. Building Structures Atom +by Atom via Electron Beam Manipulation. Small 2018, 14 (38), 1801771. +https://doi.org/doi:10.1002/smll.201801771. +(31) +Tripathi, M.; Mittelberger, A.; Pike, N. A.; Mangler, C.; Meyer, J. C.; Verstraete, M. J.; Kotakoski, J.; Susi, +T. Electron-Beam Manipulation of Silicon Dopants in Graphene. Nano Lett. 2018, 18 (8), 5319–5323. +https://doi.org/10.1021/acs.nanolett.8b02406. +(32) +Mustonen, K.; Markevich, A.; Tripathi, M.; Inani, H.; Ding, E.-X.; Hussain, A.; Mangler, C.; Kauppinen, E. +I.; Kotakoski, J.; Susi, T. Electron-Beam Manipulation of Silicon Impurities in Single-Walled Carbon Nanotubes. +Adv. Funct. Mater. 2019, 29 (52), 1901327. https://doi.org/10.1002/adfm.201901327. +(33) +Susi, T.; Kotakoski, J.; Kepaptsoglou, D.; Mangler, C.; Lovejoy, T. C.; Krivanek, O. L.; Zan, R.; Bangert, +U.; Ayala, P.; Meyer, J. C.; Ramasse, Q. Silicon-Carbon Bond Inversions Driven by 60-KeV Electrons in Graphene. +Phys. Rev. Lett. 2014, 113 (11), 115501. +(34) +Susi, T.; Meyer, J. C.; Kotakoski, J. Manipulating Low-Dimensional Materials down to the Level of Single +Atoms with Electron Irradiation. Ultramicroscopy 2017, 180, 163–172. +http://dx.doi.org/10.1016/j.ultramic.2017.03.005. +(35) +Dyck, O.; Zhang, C.; Rack, P. D.; Fowlkes, J. D.; Sumpter, B.; Lupini, A. R.; Kalinin, S. V.; Jesse, S. +Electron-Beam Introduction of Heteroatomic Pt–Si Structures in Graphene. Carbon 2020, 161, 750–757. +https://doi.org/10.1016/j.carbon.2020.01.042. + +15 + +(36) +Dyck, O.; Yoon, M.; Zhang, L.; Lupini, A. R.; Swett, J. L.; Jesse, S. Doping of Cr in Graphene Using +Electron Beam Manipulation for Functional Defect Engineering. ACS Appl. Nano Mater. 2020, 3 (11), 10855– +10863. https://doi.org/10.1021/acsanm.0c02118. +(37) +Dyck, O.; Zhang, L.; Yoon, M.; Swett, J. L.; Hensley, D.; Zhang, C.; Rack, P. D.; Fowlkes, J. D.; Lupini, A. +R.; Jesse, S. Doping Transition-Metal Atoms in Graphene for Atomic-Scale Tailoring of Electronic, Magnetic, and +Quantum Topological Properties. Carbon 2021, 173, 205–214. https://doi.org/10.1016/j.carbon.2020.11.015. +(38) +Susi, T.; Kotakoski, J.; Arenal, R.; Kurasch, S.; Jiang, H.; Skakalova, V.; Stephan, O.; Krasheninnikov, A. +V.; Kauppinen, E. I.; Kaiser, U.; Meyer, J. C. Atomistic Description of Electron Beam Damage in Nitrogen-Doped +Graphene and Single-Walled Carbon Nanotubes. ACS Nano 2012, 6 (10), 8837–8846. +https://doi.org/10.1021/nn303944f. +(39) +Susi, T.; Hardcastle, T. P.; Hofsäss, H.; Mittelberger, A.; Pennycook, T. J.; Mangler, C.; Drummond- +Brydson, R.; Scott, A. J.; Meyer, J. C.; Kotakoski, J. Single-Atom Spectroscopy of Phosphorus Dopants Implanted +into Graphene. 2D Mater. 2017, 4 (2), 021013. +(40) +Chirita Mihaila, A. I.; Susi, T.; Kotakoski, J. Influence of Temperature on the Displacement Threshold +Energy in Graphene. Sci. Rep. 2019, 9 (1), 12981. https://doi.org/10.1038/s41598-019-49565-4. +(41) +Elibol, K.; Mangler, C.; O’Regan, D. D.; Mustonen, K.; Eder, D.; Meyer, J. C.; Kotakoski, J.; Hobbs, R. G.; +Susi, T.; Bayer, B. C. Single Indium Atoms and Few-Atom Indium Clusters Anchored onto Graphene via Silicon +Heteroatoms. ACS Nano 2021, 15 (9), 14373–14383. https://doi.org/10.1021/acsnano.1c03535. +(42) +Inani, H.; Mustonen, K.; Markevich, A.; Ding, E.-X.; Tripathi, M.; Hussain, A.; Mangler, C.; Kauppinen, E. +I.; Susi, T.; Kotakoski, J. Silicon Substitution in Nanotubes and Graphene via Intermittent Vacancies. J. Phys. Chem. +C 2019, 123 (20), 13136–13140. https://doi.org/10.1021/acs.jpcc.9b01894. +(43) +van Dorp, W. F.; van Someren, B.; Hagen, C. W.; Kruit, P.; Crozier, P. A. Approaching the Resolution Limit +of Nanometer-Scale Electron Beam-Induced Deposition. Nano Lett. 2005, 5 (7), 1303–1307. +https://doi.org/10.1021/nl050522i. +(44) +van Dorp, W. F.; Zhang, X.; Feringa, B. L.; Hansen, T. W.; Wagner, J. B.; De Hosson, J. T. M. Molecule-by- +Molecule Writing Using a Focused Electron Beam. ACS Nano 2012, 6 (11), 10076–10081. +https://doi.org/10.1021/nn303793w. +(45) +van Dorp, W. F.; Lazić, I.; Beyer, A.; Gölzhäuser, A.; Wagner, J. B.; Hansen, T. W.; Hagen, C. W. Ultrahigh +Resolution Focused Electron Beam Induced Processing: The Effect of Substrate Thickness. Nanotechnology 2011, +22 (11), 115303. https://doi.org/10.1088/0957-4484/22/11/115303. +(46) +Jannik C. Meyer; C. O. Girit; M. F. Crommie; A. Zettl. Hydrocarbon Lithography on Graphene +Membranes. Appl. Phys. Lett. 2008, 92 (12), 123110. https://doi.org/10.1063/1.2901147. +(47) +Vlassiouk, I.; Fulvio, P.; Meyer, H.; Lavrik, N.; Dai, S.; Datskos, P.; Smirnov, S. Large Scale Atmospheric +Pressure Chemical Vapor Deposition of Graphene. Carbon 2013, 54 (Supplement C), 58–67. +https://doi.org/10.1016/j.carbon.2012.11.003. +(48) +Dyck, O.; Yeom, S.; Dillender, S.; Lupini, A. R.; Yoon, M.; Jesse, S. The Role of Temperature on Defect +Diffusion and Nanoscale Patterning in Graphene. Carbon 2023, 201, 212–221. +https://doi.org/10.1016/j.carbon.2022.09.006. +(49) +Postl, A.; Hilgert, P. P. P.; Markevich, A.; Madsen, J.; Mustonen, K.; Kotakoski, J.; Susi, T. Indirect +Measurement of the Carbon Adatom Migration Barrier on Graphene. Carbon 2022, 196, 596–601. +https://doi.org/10.1016/j.carbon.2022.05.039. +(50) +Zan, R.; Ramasse, Q. M.; Bangert, U.; Novoselov, K. S. Graphene Reknits Its Holes. Nano Lett. 2012, 12 +(8), 3936–3940. https://doi.org/10.1021/nl300985q. + + diff --git a/b9E0T4oBgHgl3EQf4wKo/content/tmp_files/load_file.txt b/b9E0T4oBgHgl3EQf4wKo/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..f873d031eeb8993e1819f23a63bdb2556a42dfb0 --- /dev/null +++ b/b9E0T4oBgHgl3EQf4wKo/content/tmp_files/load_file.txt @@ -0,0 +1,1272 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf,len=1271 +page_content='1 Direct-Writing Atom-by-Atom Ondrej Dyck1, Andrew R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini1, Stephen Jesse1 1 Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN Abstract Direct-write processes enable the alteration or deposition of materials in a continuous, directable, sequential fashion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In this work we demonstrate an electron beam direct-write process in an aberration-corrected scanning transmission electron microscope.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This process has several fundamental differences from conventional electron beam induced deposition techniques, where the electron beam dissociates precursor gases into chemically reactive products that bond to a substrate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Here, we use elemental tin (Sn) as a precursor and employ a different mechanism to facilitate deposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The atomic-sized electron beam is used to generate chemically reactive point defects at desired locations in a graphene substrate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Temperature control of the sample is used to enable the precursor atoms to migrate across the surface and bond to the defect sites thereby enabling atom-by-atom direct-writing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2 Introduction The scanning transmission electron microscope (STEM) has a long legacy as an indispensable materials characterization platform.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In recent years this picture has begun to expand to include not only characterization but also manipulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1–3 From this perspective the STEM can be viewed as a fabrication platform capable of addressing individual atoms with the highly focused electron beam (e-beam) and altering materials atom-by-atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Such capabilities hearken back to Feynman’s famous quote about putting every atom where we want.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='4 One would like to be able to build structures from the atom up, designed for their electronic, optical, chemical, or structural properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Quantum qubits, quantum sensors, and perhaps even full computing circuits could be tailored with atomic specificity to elicit the emergent quantum properties of the materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Creating structures with atomic specificity would throw open the doors for experiments developing fundamental understanding of physics, chemistry, and materials science at the atomic level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Realizing such a vision, however, is a more difficult task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Observations of e-beam induced sample alterations date back to some of the earliest days of electron microscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Every electron microscopist is familiar with e-beam damage5–7 and hydrocarbon deposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mitigation of these unintentional processes has been a key enabling factor for the (S)TEM to excel at materials characterization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lower accelerating voltages,8,9 enhanced by aberration correction,10 have been effective at reducing knock-on damage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' With unintended processes at least partially under control, we can begin to isolate variables to examine their effects on the specimen with a view toward reproducible atomistic transformations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' To simplify matters further, here we work with 2D materials where, instead of looking at projected columns of atoms (as in a 3D crystal), each atom can be uniquely identified in the image.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The most robust and well-studied 2D material is graphene and many studies have been carried out to find ways to clean and image contamination-free graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='11–24 With these advancements, which act to mitigate major alterations to the specimen, minor alterations at the atomic scale can be observed and studied.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Some examples of these atomic scale dynamics are shown in Figure 1(a)-(d): the spinning of a Si molecular rotor,25 the dynamic rearrangement of a Si6 cluster embedded in a graphene nanopore,26 the movement of a 3-fold coordinated Si dopant in graphene, and vacancy diffusion in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='27 In these examples the major focus was to observe and understand the e- 3 beam driven effects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The transition to top-down control over atomic movement stemmed from the realization that the position of the e-beam could influence the direction of motion of the 3-fold coordinated Si atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The first demonstration of precisely directed movement of a single atom using an e-beam is shown in Figure 1(e) where a 3-fold coordinated Si atom was moved seven steps through a graphene lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='28 This result was independently replicated shortly afterward29 and extended to longer distances,30,31 shown in Figure 1(f) and (g).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The first e-beam assembly of a primitive structure, a Si dimer, is shown in Figure 1(h).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='30 Controlled movement of a Si atom through the wall of a single walled carbon nanotube is shown in Figure 1(i)32 and intentional rotation of a Si trimer, as well as the reversible conversion to a tetrimer is shown in Figure 1(j).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='30 Figure 1 Select examples of observed and controlled atomic dynamics in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (a) Observation of Si trimer rotation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (b) Observation of reversible transformation of a Si6 cluster embedded in a graphene nanopore.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (c) Observation of movement of 3-fold coordinated Si dopant in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (d) Observation of vacancy diffusion in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (e) Controlled movement of 3-fold coordinated Si dopant in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (f) and (g) Controlled movement of 3-fold coordinated Si atom over an extended distance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (h) Controlled assembly of a Si dimer in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (i) Controlled movement of Si atom in a single walled carbon nanotube.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (j) Controlled rotation of a Si trimer and conversion to tetrimer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (a) Adapted from reference [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Reuse is permitted per the United States Department of Energy public access plan http://energy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='gov/downloads/doe-public-access-plan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (b) Adapted with permission from reference [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (c) Reproduced with permission from reference [33] https://creativecommons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/licenses/by/3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='0/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (d) Reproduced with permission from reference [27] https://creativecommons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/licenses/by/3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='0/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (e) Adapted with permission from reference [34].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (f) Reproduced with permission from reference [30].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (g) Reproduced with permission from reference [31] https://pubs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/page/policy/authorchoice_ccby_termsofuse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='html.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (h) Adapted with permission from reference [30].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (i) Reproduced with permission from reference [32] https://creativecommons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/licenses/by/4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='0/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (j) Adapted with permission from reference [30].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Observed atomic dynamics Controlledatomicdynamics a b e 50 100 150 200 h Time (s) d Rotation Rotation Trimer to tetrimer Tetrimerto trimer4 While there is a certain elegance to the ability to move a single atom through the graphene lattice, it is not obvious that positioned Si atoms in graphene have applications aside from the development of these experimental techniques and unraveling the fundamental physics of beam-matter interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Likewise, the Si atom movement typically terminates with the ejection of an additional carbon atom converting it into the much less mobile 4-fold coordination.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='31 This conversion occurs unpredictably, meaning that the distance a 3-fold coordinated Si atom can be moved is also unpredictable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' First principles simulations of the assembly of the dimer shown in Figure 1(h) indicate that this assembly process is energetically favorable, which facilitates bringing atoms together.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' However, this energetic trend to agglomerate further complicates the predictability of atom movement in the presence of other dopants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Thus, the tailoring of atomic structures in this way is inherently limited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Investigations have been undertaken to extend the types of dopant atoms than can be put in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The main, serendipitous factor that facilitated the studies shown in Figure 1 was that Si dopant atoms are commonly found in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Moving from serendipity to predictability, it was shown that the Si atoms in the surface contamination could be introduced into the graphene lattice by deliberately ejecting C atoms with the e-beam to create attachment sites for the Si atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='29,30 Dopant Si atoms could then be reliably produced using the e-beam as a manipulation tool.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This technique was extended and applied generally to many elements where the source materials were first introduced onto the graphene surface.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='35–37 Common to each of these demonstrations was the limitation that dopant insertion was only accomplished within a few nanometers from the parent nanoparticle and required a two-step procedure where the e-beam was used both for creating defects in the graphene substrate as well as separating source atoms from the parent nanoparticles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' These and other closely related studies21,28,32,38–46 suggest it might be possible to find appropriate conditions under which a continuous atom-by-atom direct-writing procedure may be realized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A direct-write mechanism would enable the creation of longer-range atomic structures without requiring the step-by-step movement of single atoms through the lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Here, we aim to demonstrate this possibility.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' For an atom-by-atom direct-writing procedure to be of practical use we must address several challenges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Conceptually, a direct-write procedure binds a material to a substrate through some local chemical or physical influence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This conceptualization glosses over the atomistic details of 5 the process in favor of the more abstract representation that more easily describes the intended goal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In a similar way, common descriptions of direct-write deposition such as electron beam induced deposition (EBID) do not spell out the details of the chemical attachment of every atom to the substrate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' However, as we scale down a writing process, eventually we must confront the atomistic nature of the world and begin to address the fact that “binding a material to a substrate” is a vague statement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' If we are to write with individual atoms, we must be specific about what we mean.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Here, we mean that atoms, differing elementally from the substrate, form strong chemical bonds to defect sites introduced in the substrate through a localized influence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' With this formulation we can begin to specify the requirements for writing at the atomic scale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' We must have source materials (atoms) that are delivered to the region of interest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' We must have a highly localized influence on the sample that is able to create chemically reactive sites with which the source atoms can bond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' These bonds need to be strong enough to prevent spontaneous thermal diffusion over long time spans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The definition of “long”, clearly, depends on the application but here we envision stability for days, ideally years, at ambient temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Critically, we must have a clean surface on which to write, in order to provide a spatially uniform environment that enhances predictable control over the process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Methods To satisfy these constraints we use a suspended single layer graphene substrate, operate our microscope above the knock-on threshold for graphene, and thermally deliver Sn atoms to the damaged regions in situ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Atmospheric pressure chemical vapor deposition (AP-CVD) was used to grow graphene on Cu foil.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='47 Poly(methyl methacrylate) (PMMA) was spin coated over the graphene for protection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A wet transfer29 was used to transfer the graphene to a ProtochipsTM heater chip substrate and Sn was evaporated onto the graphene surface in a vacuum chamber using a Sn coated W filament.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The graphene was pre-cleaned using a rapid thermal ramp to 1200 °C23 and the Sn evaporation was carried out while the sample temperature was maintained at 500 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A Nion UltraSTEM 200 operated at an accelerating voltage of 100 kV and nominal beam current of 17 pA was used to examine the sample.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Prior to examination, the sample, holder cartridge, and magazine were baked in vacuum for eight hours at 160 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' After introduction into the pole piece, 6 the sample was again heated to 1200 °C to remove residual contaminants and then cooled to 300 °C for further examination.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Experimental Sample overview Figure 2(a) and (b) show two high angle annular dark field (HAADF) images of the graphene and Sn nanoparticles at 300 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Much of the sample was multilayer and the Sn nanoparticles attached predominantly at step edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' We note the presence of single Sn atoms decorating the step edges and various other locations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 2(c) shows an example EELS spectrum acquired from one of the nanoparticles confirming that they are made of Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 2 Sample overview illustrating Sn nanoparticles attached to graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (a), (b) representative HAADF images of Sn nanoparticles attached to the graphene surface particularly at step edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (c) Sn M4,5 EELS edge acquired on one of the nanoparticles confirming they are composed of Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Beam Dragging Following the “beam dragging” strategy laid out in previous work,30,35–37 Sn atoms were inserted into the graphene lattice with the substrate maintained at 300 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The e-beam was positioned on the source Sn particle and manually moved from the particle to the graphene, acting to both sputter Sn atoms from the parent nanoparticle and generate point defects in the graphene lattice for the Sn atoms to occupy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This procedure is depicted schematically in Figure 3(a)-(d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In the initial a b c Sn M4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='5 EELS Edge 20000 Single Layer 17500 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 15000 12500 Snnanoparticles Sn nanoparticles Counts 10000 7500 Single Layer 5000 2500 Bilayer Bilayer 450500550600650 700 Energy loss [eV] 5nm Trilayer Trilayer7 configuration, Figure 3(a), the e-beam is positioned on a nanoparticle of source material sitting on the graphene substrate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The e-beam is then moved across the substrate, depicted in Figure 3(b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This procedure generates vacancies in the graphene through direct ejection of lattice carbon atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The e-beam is then moved back to the source material and used to sputter atoms from the parent nanoparticle, Figure 3(c).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The freed atoms can diffuse away from the nanoparticle across the graphene surface.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' There is some probability for vacancy diffusion and lattice restructuring;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' a detailed study and dynamic calculations are given in reference 48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 3(c) illustrates an example where two vacancies have merged forming a reconstructed divacancy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This structure does not have any dangling carbon bonds, which suggests that it (and other similar defect structures) will be less attractive for a diffusing adatom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 3(d) depicts the capture of a diffusing atom by the remaining vacancy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The generation of vacancies, sputtering of source atoms, and diffusion of vacancies and atoms are all stochastic processes, which means they are not guaranteed to occur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Typically, the e-beam must be repeatedly moved back and forth and the sample state checked via imaging after some time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 3(e)-(g) shows a summary of the result of this procedure performed experimentally.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sn atoms are ejected from the parent nanoparticle and incorporated into the graphene lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' As was also true in prior publications,29,30,35–37 this process works across the short distance of a few nanometers and requires repeated attempts at defect creation and sputtering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This example was carried out with the substrate held at 300 °C and gives a reference point for the behavior of the graphene/Sn system which was qualitatively like the behavior of other elements at room temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='35–37 8 Figure 3 Example of dopant insertion into single layer graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (a) Schematic of e-beam positioned on source material.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (b) e-beam is manually dragged across the substrate creating vacancies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (c) E-beam is repositioned on the source material and atoms are sputtered off.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (d) adatoms can bond to the created vacancy sites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (e) Initial sample state where a Sn nanoparticle is attached to a step edge overlapping the single layer region.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (f) “beam dragging” was performed in the locations indicated by the arrows resulting in Sn dopant insertion along the beam trajectory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (g) final state of the sample.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' beam Source Material d Captured Dopant Vacancies Ejected Diffusing Carbon Adatoms Reconstructed Divacancy e Hhm 4nm9 It is worth emphasizing that the e-beam in this and prior examples of dopant insertion is performing two distinct functions: 1) the generation of vacancies/defects in the graphene substrate to which the source atoms can chemically bond (Figure 3(b)), and 2) the ejection/sputtering of source atoms from the parent nanoparticle (Figure 3(c) and (d)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This means that the rate of supply of source atoms, freed from the parent nanoparticle, is dependent on e-beam current and position.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' More source atoms could likely be supplied by increasing the e-beam current, however, this also increases the defect generation rate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Because the defects are generated when the e-beam is positioned off the nanoparticle, the supply of source atoms is not continuously maintained.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' It is likely that these constraints are the main reason why dopant insertion with this method has only been accomplished within a few nanometers of the parent nanoparticle and why multiple attempts must be made before dopant insertion is accomplished.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Direct Writing To enhance our understanding of the processes described above quantitative modeling and measurement of the availability of source atoms as a function of e-beam current or position, for example, could be performed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' However, the crux of the dilemma is that this process is reliant on the e-beam for both sputtering source material and the creation of defects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In contrast, e-beam induced deposition (EBID) does not rely on the e-beam to generate the flow of precursor material, and ideally the same principle should apply here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' We seek appropriate sample conditions where the source material can be supplied to the e-beam location without the need for e-beam sputtering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' For the system being examined here, the most relevant parameter that can be easily controlled is the substrate temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' To this end, the temperature of the sample was increased in steps to 900 °C, stopping momentarily to verify the continued presence of the Sn nanoparticles at 600, 700, and 800 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' At 900 °C the sample was examined more closely.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 4(a) shows an image of a step edge harboring many Sn atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 4(b) shows the same location, acquired five minutes later.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The spontaneous dispersion of Sn atoms along the step edge and the formation of multiple Sn nanoparticles suggests that the Sn atoms are highly mobile.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' One potential interpretation is that Sn atoms are rapidly diffusing across the graphene surface (too rapidly to image) but are stabilized at the step edge and become imageable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This interpretation 10 suggests that a vacancy created anywhere in the graphene should result in the attachment of a randomly diffusing Sn atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' To explore this hypothesis, we used the e-beam to scan a small (~0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='5 nm square) area over a random patch of the graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' After a few seconds, instead of observing dopant insertion, a hole was created in the graphene indicating that the supply of Sn atoms in that location is lower than is present at the step edge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The spatial distribution of the supply of Sn atoms deserves further theoretical investigation, however our goal here was to empirically find conditions favorable for direct-writing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Since there seemed to be an ample supply of Sn atoms at the step edge, a different strategy was employed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A small (~0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='5 nm) subscan box was defined and positioned at the step edge, which provides a small, rapidly updating image of the irradiated location for real-time monitoring.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' As the step edge harboring Sn atoms was irradiated, Sn atoms were observed to attach to the underlying graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The subscan box was then manually repositioned to expose more of the pristine graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In this way, Sn atoms were drawn from the step edge and attached to vacancies/defects in a continuous manner with a single pass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This procedure is illustrated visually in the schematic in Figure 4(c).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 4(d) shows a HAADF image of the resulting Sn dopant chain drawn in the graphene lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 4(e) shows a magnified view displayed as a smoothed surface plot and artificially colored.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The subscan box was moved as indicated by the arrow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' After image acquisition, the process was continued, as shown in.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 4(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This second line required about seven minutes to complete.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Figure 4(g) shows a magnified surface plot of (f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 11 Figure 4 Demonstrating atom-by-atom direct write using the e-beam at 900 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (a) Initial state of a bilayer graphene step edge at 900 °C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (b) The same region shown in (a) acquired five minutes later.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Additional Sn nanoparticles attached to the step edge indicate a high degree of Sn mobility along the edge, arrowed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Images shown in the (d)-(g) were acquired near the dashed region on the left.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (c) Schematic illustration of the direct writing procedure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (d) HAADF image of a line of Sn dopant atoms directly written in graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (e) Enlarged and labeled surface plot of the image in (d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (f) The direct write process was continued, starting where the previous writing was halted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (g) Surface plot of the image shown in (f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Images in (a) and (b) were artificially colored using the “Fire” look up table in Imagej.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Images (e) and (g) were artificially colored and smoothed to reduce noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' It should be noted that we observe some Sn dopants that attach to the graphene surface outside the direct-write region.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The most likely interpretation is that thermally induced vacancy diffusion enables vacancy migration from the irradiated region into the pristine graphene48 after which Sn atoms can bond to the vacancy site.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This evidence of Sn atoms diffusing across the graphene ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='surface combined with the result of forming holes when direct deposition was attempted at random ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Direct WritingProcedure ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='SnAtomsPopulatingStep Edge ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Sn Atoms ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Sn Atoms ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='a ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Bilayer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='onStepEdge ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='inGraphene ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Graphene ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='5hm ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Irradiated ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Sn Aggregating into Nanoparticles ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Region ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='b ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Single Layer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Graphene ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='ProgressionThroughTime ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='f ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='5 nm ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='e ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='g ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Sn Atoms ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='on Step Edge ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Sn Atoms ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='ManuallyPositioned ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='in Graphene ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Subscan Box ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Single Layer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Progression Through Time ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='Graphene12 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='locations on the graphene suggests that the ratio of the defect generation rate to the adatom ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='diffusivity (or concentration) plays a significant role in determining the outcome.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' These observations point toward fruitful avenues of inquiry for future experiments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Clearly there are a variety of factors that need to be further understood and controlled to optimize atom-by-atom direct writing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Previous work in this context highlights the role of mobile carbon adatoms that spontaneously heal vacancies and larger structural defects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='49,50 At the same time, rapidly diffusing vacancies can markedly change the behavior of graphene at high temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='48 E-beam patterning of corrals24 can influence the influx of hydrocarbon contamination or provide a trap for long-range vacancy diffusion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The dynamics of these factors are clearly dependent on the sample temperature, and here we have demonstrated that temperature can also be used to influence the supply of source atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' We have shown that a step-edge on the graphene and a deliberately created dopant chain provide local channels for the migration of source atoms and can enable continuous direct-writing of nanoscale features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Conclusion We have demonstrated an e-beam directed, continuous atom-by-atom direct-write process on suspended single layer graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This process relies upon the repeated ejection of C atoms from the substrate lattice as well as a continuous supply of source atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The generation of defect sites is governed by the e-beam accelerating voltage and beam current, here fixed at 100 kV and 17 pA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The incorporation of source atoms into the defect sites is governed by the rate of supply of source atoms to the generated vacancies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Substrate temperature, diffusion rates, and the local energetic landscape play significant roles in this process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' We find that sample features, such as the graphene step edge, or artificially engineered features, such as the direct-write lines of atoms implanted into the graphene lattice, both serve a role in delivering source atoms to the defect sites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' While the role of temperature might generally be expected, the role of local defects suggests that one could deliberately tailor the local environment to be favorable for the direct-write operation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' These findings indicate generally that e-beam atom-by-atom direct-write processes are possible provided that the source material can be supplied at the requisite rate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This suggests that a search 13 for conditions which enable a direct-write process at arbitrary sample locations would be a fruitful future direction of inquiry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Similar processes may enable atom-by-atom direct-writing on various other 2D materials or possibly inside 3D materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Acknowledgment The authors would like to gratefully acknowledge the assistance of Zachary Gosser in the Sn evaporation process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' This work was supported by the U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division (O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=', S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ), and was performed at the Center for Nanophase Materials Sciences (CNMS), a U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Department of Energy, Office of Science User Facility.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' References (1) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ziatdinov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lingerfelt, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Unocic, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hudak, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Atom-by-Atom Fabrication with Electron Beams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2019, 4 (7), 497–507.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1038/s41578-019-0118-z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (2) Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Borisevich, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Fire up the Atom Forge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nature 2016, 539 (7630), 485–487.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (3) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A Self-Driving Microscope and the Atomic Forge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' MRS Bull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2019, 44 (9), 669–670.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1557/mrs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='211.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (4) Feynman, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' There’s Plenty of Room at the Bottom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Eng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 1960, 23 (5), 22–36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (5) Stenn, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Bahr, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Specimen Damage Caused by the Beam of the Transmission Electron Microscope, a Correlative Reconsideration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ultrastruct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 1970, 31 (5), 526–550.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/S0022- 5320(70)90167-X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (6) Nan Jiang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Electron Beam Damage in Oxides: A Review.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Prog.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2016, 79 (1), 016501.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (7) Egerton, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Wang, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Crozier, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Beam-Induced Damage to Thin Specimens in an Intense Electron Probe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Microsc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Microanal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2005, 12 (1), 65–71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1017/S1431927606060065.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (8) Krivanek, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Dellby, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Murfitt, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chisholm, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pennycook, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Suenaga, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nicolosi, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Gentle STEM: ADF Imaging and EELS at Low Primary Energies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ultramicroscopy 2010, 110 (8), 935–945.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='ultramic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='02.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (9) Ondrej L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Krivanek;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Wu Zhou;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Matthew F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chisholm;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Juan Carlos Idrobo;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Tracy C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lovejoy;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Quentin M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ramasse;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Niklas Dellby.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Gentle STEM of Single Atoms: Low KeV Imaging and Analysis at Ultimate Detection Limits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' In Low Voltage Electron Microscopy;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Brooks, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Bell, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Erdman, Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (10) Hawkes, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Advances in Imaging and Electron Physics: Aberration-Corrected Electron Microscopy;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Elsevier Science, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (11) A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Goossens;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Calado;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Barreiro;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Watanabe;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Taniguchi;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Vandersypen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mechanical Cleaning of Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2012, 100 (7), 073110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='3685504.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (12) C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Joiner;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Roy;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hesabi;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chakrabarti;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Vogel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Cleaning Graphene with a Titanium Sacrificial Layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2014, 104 (22), 223109.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='4881886.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (13) Gerardo Algara-Siller;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ossi Lehtinen;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Andrey Turchanin;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ute Kaiser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Dry-Cleaning of Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2014, 104 (15), 153115.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='4871997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (14) J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Moser;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Barreiro;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Bachtold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Current-Induced Cleaning of Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2007, 91 (16), 163513.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2789673.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (15) Jean-Nicolas Longchamp;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Conrad Escher;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hans-Werner Fink.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ultraclean Freestanding Graphene by 14 Platinum-Metal Catalysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Vac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Technol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B 2013, 31 (2), 020605.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1116/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='4793746.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (16) Liang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sperling, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Calizo, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Cheng, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hacker, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Obeng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Peng, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Li, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yuan, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hight Walker, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Peng, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Richter, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Toward Clean and Crackless Transfer of Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ACS Nano 2011, 5 (11), 9144–9153.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nn203377t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (17) Lin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lee, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Suenaga, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chiu, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Clean Transfer of Graphene for Isolation and Suspension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ACS Nano 2011, 5 (3), 2362–2368.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nn200105j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (18) Lin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yeh, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Suenaga, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chiu, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Graphene Annealing: How Clean Can It Be?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nano Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2012, 12 (1), 414–419.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nl203733r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (19) Schweizer, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Dolle, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Dasler, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Abellán, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hauke, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hirsch, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Spiecker, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mechanical Cleaning of Graphene Using in Situ Electron Microscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2020, 11 (1), 1743.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1038/s41467- 020-15255-3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (20) Tripathi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mittelberger, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mustonen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Cleaning Graphene: Comparing Heat Treatments in Air and in Vacuum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Status Solidi RRL – Rapid Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2017, 11 (8), 1700124-n/a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1002/pssr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='201700124.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (21) W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' van Dorp;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Feringa;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Wagner;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hansen;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Th M De Hosson.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nanometer- Scale Lithography on Microscopically Clean Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nanotechnology 2011, 22 (50), 505303.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (22) Xie, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Weng, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ng, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chan, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chan, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Clean Graphene Surface through High Temperature Annealing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Carbon 2015, 94, 740–748.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' http://dx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='carbon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='07.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='046.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (23) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mitigating E-Beam-Induced Hydrocarbon Deposition on Graphene for Atomic-Scale Scanning Transmission Electron Microscopy Studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Vac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Technol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B Nanotechnol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Microelectron.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phenom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2018, 36 (1), 011801.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1116/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='5003034.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (24) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rack, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Fowlkes, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Controlling Hydrocarbon Transport and Electron Beam Induced Deposition on Single Layer Graphene: Toward Atomic Scale Synthesis in the Scanning Transmission Electron Microscope.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nano Sel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2022, 3 (3), 643–654.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1002/nano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='202100188.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (25) Yang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yin, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lee, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ren, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Cheng, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ye, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pantelides, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pennycook, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chisholm, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Direct Observation of Atomic Dynamics and Silicon Doping at a Topological Defect in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2014, 126 (34), 9054–9058.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1002/ange.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='201403382.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (26) Lee, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhou, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pennycook, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Idrobo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pantelides, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Direct Visualization of Reversible Dynamics in a Si6 Cluster Embedded in a Graphene Pore.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2013, 4, 1650.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1038/ncomms2671 https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='nature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='com/articles/ncomms2671#supplementary-information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (27) Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Imaging Atomic-Level Random Walk of a Point Defect in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2014, 5, 3991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1038/ncomms4991 https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='nature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='com/articles/ncomms4991#supplementary-information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (28) Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kepaptsoglou, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ramasse, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Suenaga, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Towards Atomically Precise Manipulation of 2D Nanostructures in the Electron Microscope.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2D Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2017, 4 (4), 042004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (29) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Placing Single Atoms in Graphene with a Scanning Transmission Electron Microscope.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2017, 111 (11), 113104.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='4998599.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (30) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jimenez-Izal, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Alexandrova, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Building Structures Atom by Atom via Electron Beam Manipulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Small 2018, 14 (38), 1801771.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/doi:10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1002/smll.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='201801771.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (31) Tripathi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mittelberger, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pike, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Verstraete, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Electron-Beam Manipulation of Silicon Dopants in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nano Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2018, 18 (8), 5319–5323.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='nanolett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='8b02406.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (32) Mustonen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Markevich, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Tripathi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Inani, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ding, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hussain, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kauppinen, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Electron-Beam Manipulation of Silicon Impurities in Single-Walled Carbon Nanotubes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Funct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2019, 29 (52), 1901327.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1002/adfm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='201901327.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (33) Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kepaptsoglou, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lovejoy, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Krivanek, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zan, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Bangert, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ayala, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ramasse, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Silicon-Carbon Bond Inversions Driven by 60-KeV Electrons in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2014, 113 (11), 115501.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (34) Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Manipulating Low-Dimensional Materials down to the Level of Single Atoms with Electron Irradiation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ultramicroscopy 2017, 180, 163–172.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' http://dx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='ultramic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='03.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (35) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rack, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Fowlkes, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sumpter, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kalinin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Electron-Beam Introduction of Heteroatomic Pt–Si Structures in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Carbon 2020, 161, 750–757.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='carbon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='042.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 15 (36) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yoon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Swett, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Doping of Cr in Graphene Using Electron Beam Manipulation for Functional Defect Engineering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ACS Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nano Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2020, 3 (11), 10855– 10863.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/acsanm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='0c02118.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (37) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yoon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Swett, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hensley, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rack, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Fowlkes, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Doping Transition-Metal Atoms in Graphene for Atomic-Scale Tailoring of Electronic, Magnetic, and Quantum Topological Properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Carbon 2021, 173, 205–214.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='carbon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (38) Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Arenal, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kurasch, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jiang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Skakalova, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Stephan, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Krasheninnikov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kauppinen, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kaiser, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Atomistic Description of Electron Beam Damage in Nitrogen-Doped Graphene and Single-Walled Carbon Nanotubes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ACS Nano 2012, 6 (10), 8837–8846.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nn303944f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (39) Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hardcastle, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hofsäss, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mittelberger, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Pennycook, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Drummond- Brydson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Scott, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Single-Atom Spectroscopy of Phosphorus Dopants Implanted into Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2D Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2017, 4 (2), 021013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (40) Chirita Mihaila, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Influence of Temperature on the Displacement Threshold Energy in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2019, 9 (1), 12981.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1038/s41598-019-49565-4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (41) Elibol, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' O’Regan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mustonen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Eder, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hobbs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Bayer, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Single Indium Atoms and Few-Atom Indium Clusters Anchored onto Graphene via Silicon Heteroatoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ACS Nano 2021, 15 (9), 14373–14383.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/acsnano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1c03535.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (42) Inani, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mustonen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Markevich, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ding, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='-X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Tripathi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hussain, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mangler, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kauppinen, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Silicon Substitution in Nanotubes and Graphene via Intermittent Vacancies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C 2019, 123 (20), 13136–13140.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='jpcc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='9b01894.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (43) van Dorp, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' van Someren, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hagen, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kruit, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Crozier, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Approaching the Resolution Limit of Nanometer-Scale Electron Beam-Induced Deposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nano Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2005, 5 (7), 1303–1307.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nl050522i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (44) van Dorp, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zhang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Feringa, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hansen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Wagner, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' De Hosson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Molecule-by- Molecule Writing Using a Focused Electron Beam.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' ACS Nano 2012, 6 (11), 10076–10081.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nn303793w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (45) van Dorp, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lazić, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Beyer, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Gölzhäuser, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Wagner, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hansen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hagen, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ultrahigh Resolution Focused Electron Beam Induced Processing: The Effect of Substrate Thickness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nanotechnology 2011, 22 (11), 115303.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1088/0957-4484/22/11/115303.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (46) Jannik C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Girit;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Crommie;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Zettl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hydrocarbon Lithography on Graphene Membranes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2008, 92 (12), 123110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2901147.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (47) Vlassiouk, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Fulvio, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Meyer, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lavrik, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Dai, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Datskos, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Smirnov, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Large Scale Atmospheric Pressure Chemical Vapor Deposition of Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Carbon 2013, 54 (Supplement C), 58–67.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='carbon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (48) Dyck, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yeom, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Dillender, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Lupini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Yoon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Jesse, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' The Role of Temperature on Defect Diffusion and Nanoscale Patterning in Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Carbon 2023, 201, 212–221.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='carbon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='09.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (49) Postl, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Hilgert, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Markevich, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Madsen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Mustonen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Kotakoski, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Susi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Indirect Measurement of the Carbon Adatom Migration Barrier on Graphene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Carbon 2022, 196, 596–601.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='carbon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='039.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' (50) Zan, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Ramasse, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Bangert, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=';' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Novoselov, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Graphene Reknits Its Holes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' Nano Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' 2012, 12 (8), 3936–3940.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} +page_content='1021/nl300985q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9E0T4oBgHgl3EQf4wKo/content/2301.02743v1.pdf'} diff --git a/etE4T4oBgHgl3EQfqA2v/content/tmp_files/2301.05197v1.pdf.txt b/etE4T4oBgHgl3EQfqA2v/content/tmp_files/2301.05197v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..697bb1aa211bc41845461142500cb60979ca326a --- /dev/null +++ b/etE4T4oBgHgl3EQfqA2v/content/tmp_files/2301.05197v1.pdf.txt @@ -0,0 +1,1259 @@ +arXiv:2301.05197v1 [hep-ph] 12 Jan 2023 +Stop comparing resummation methods +Johan L¨ofgren1, ∗ +1Department of Physics and Astronomy, Uppsala University, Box 516, SE-751 20 Uppsala, Sweden +I argue that perturbative studies of phase transitions should always establish a consistent power- +counting through a hierarchy of scales, as it allows one to derive a correct resummation method. +Hence any confusion over which resummation method to use is misplaced once an approximation +scheme is settled on. Furthermore, adopting a consistent power-counting scheme and performing a +strict perturbative expansion dissolves many other common problems of such studies: gauge depen- +dence, strong renormalization scale dependence, the Goldstone boson catastrophe, IR divergences, +imaginary potentials, mirages (illusory barriers), perturbative breakdown, and linear terms. I il- +lustrate the strength of power-counting through various examples, and by revisiting highlights of +Arnold & Espinosa’s pioneering paper [1]. +IT HAS BEEN 30 YEARS +30 years ago, Peter Arnold and Olivier Espinosa demon- +strated how to treat first order phase transitions pertur- +batively in a groundbreaking paper [1]. Their paper is +widely cited in the phenomenological thermal field the- +ory community, where similar methods are applied to ex- +tensions of the Standard Model. In particular to study +plausible consequences of a first order electroweak phase +transition—such as electroweak baryogenesis and gravi- +tational wave signals. +In this work I revisit the arguments of [1]. In particu- +lar, I show that adhering to a strict power-counting by- +passes many current issues such as gauge dependence and +infrared divergences. Doing so provides a direct link be- +tween [1] and modern effective field theories [2–4]. +The overarching lesson is a familiar story: consistent +resummations are implemented by integrating out heavy +modes with respect to the physics you are interested in. +Establishing a hierarchy of scales will enable a derivation +of the correct resummation method to use, and hence +I argue that it is time we stop comparing resummation +methods. Instead we should start all our perturbative +studies with a power-counting analysis, and make sure +we are expanding in a small parameter. +For simplicity I focus on equilibrium quantities such as +the critical temperature, but proper perturbative expan- +sions are equally important when calculating dynamical +quantities such as the bubble nucleation rate [5–8]. +I also demonstrate the use of these techniques for a +class of models with radiative symmetry breaking. +SEE THROUGH MIRAGES +Arnold & Espinosa point out how easy it is to misin- +terpret perturbative calculations of phase transitions; I +will repeat part of their demonstration here. Consider a +∗ johan.lofgren@physics.uu.se +real-scalar theory: +Vcl(φ) = −1 +2ν2φ2 + 1 +4!λφ4, +(1) +with ν2 ≥ 0. Though this potential exhibits spontaneous +symmetry breaking, it is a priori not clear if a first-order +phase transition occurs at some high temperature. +In particular, if the scalar’s field-dependent mass is +m2(φ), the 1-loop effective potential is +V1(φ) = 1 +24m2(φ)T 2 − +1 +12π +� +m2(φ) +�3/2 T + . . . +(2) +Using the square mass m2(φ) = V ′′ +cl (φ) gives the leading +contribution +V1(φ) = 1 +48λT 2φ2, +(3) +where I discarded field-independent terms. Now, for a +phase-transition to occur, this term must be of similar +size as the tree-level potential. This indicates that we +should add its contribution to Equation 1, +VLO(φ) = 1 +2 +� +−ν2 + 1 +24λT 2 +� +φ2 + 1 +4!λφ4, +(4) +and instead consider VLO(φ) as the leading-order poten- +tial. +This means that the high-temperature φ2-coefficient is +ν2 +eff = ν2 − 1 +24λT 2, +(5) +which vanishes at temperature T0. So this model seems +to describe a second order phase transition. +But we can find the effective—resummed—square mass +from the same potential, +m2 +eff(φ) = V ′′ +LO(φ) = ν2 +eff + 1 +2λφ2. +(6) +Should we not use this improved mass in the 1-loop po- +tential? Doing so gives +VLO(φ) = 1 +2ν2 +effφ2 − T +12π +� +ν2 +eff + 1 +2λφ2 +�3/2 ++ 1 +4!λφ4. (7) + +2 +For temperatures just above T0, where ν2 +eff ⪆ 0, there +indeed appears to be a barrier. +However, after looking a bit closer at the terms in- +volved; which scale as +φc ∼ +√ +λT, −ν2 +eff(Tc) ∼ λ2T 2 =⇒ meff ∼ λT, +(8) +we see that the effective mass is small compared to +the temperature T . Though this confirms that the high- +temperature expansion applies, this is actually bad news. +Loops involving the zero-Matsubara mode each come +with a power of λT , and the only other mass scale is +meff —the expansion parameter can only be λT/meff. So +once meff ∼ λT , as it is here, loops are not suppressed. +In the words of Arnold & Espinosa: +Now consider the loop expansion parame- +ter λT/meff at the minimum φc ∼ +√ +λT +corresponding to the apparent first-order +phase transition of the ring-improved one- +loop potential (2.7). Eqn. (2.8) implies that +meff(φc) ∼ λT , and so the loop expansion +parameter is O(1), verifying that the ring- +improved loop expansion cannot be trusted to +distinguish between a first and second-order +phase transition in this model. [1, p. 7]1 +I give two variations of this argument in the section on +scale hierarchies. +In fact, the real scalar theory studied above is known +to not have a first-order phase transition [1]. The barrier +turns out to have been fictitious all along. Let us call +such illusory barriers mirages, to emphasize the danger +they pose. We can contrast mirages with real radiative +barriers, such as that of Abelian Higgs at high tempera- +tures. This model is also studied in detail by Arnold & +Espinosa, and they state +The situation is quite different when the +gauge sector is included. As we now review, +the first-order phase transition seemingly de- +scribed by the ring-improved effective poten- +tial is associated with a small loop expansion +parameter if the Higgs mass is sufficiently +small. The phase transition is therefore in- +deed first order, and the ring-improved loop +expansion is a valid tool for studying it. [1, p. +7] +The comparison of these different models by Arnold & +Espinosa showcases how important it is to use a consis- +tent power-counting. Changing the shape of the classical +potential necessarily requires large perturbative correc- +tions, and care must be taken to ensure that the expan- +sion is performed in a small parameter. +1 The page numbers apply to the ArXiv version of [1]. +STOP COMPARING RESUMMATION METHODS +The resummation method laid out by Arnold & Espinosa +can be summarized as resumming the masses of the light +modes by integrating out the heavy modes: +[...] loops with quadratic UV divergences are +of order λT 2 and are dominated by large mo- +menta of order T . These O(λT 2) contribu- +tions are all absorbed by using ring-improved +propagators. [1, p. 6-7] +In essence we should only resum masses when the self- +energy is of the same order, m2 +eff ∼ Π(meff). +The resummation can at one loop be implemented by +adding a ring-improved term to the effective potential, +Vring(φ) = − T +12π +�� +m2 +eff(φ) +�3/2 − +� +m2(φ) +�3/2� +, +(9) +effectively replacing the mass of the particle in the zero- +mode contribution to the effective potential. +More formally, Arnold & Espinosa are constructing +a three-dimensional Euclidean effective field theory in +terms of Matsubara zero modes [9, 10], and their re- +summation method can be viewed as the leading order +prediction of this EFT. The process of constructing the +3D EFT is known as dimensional reduction, and it en- +ables systematic extension of the resummation method +to higher orders [2, 3]. I discuss dimensional reduction +further in the next section. +When +the +high-temperature +expansion +applies, +Arnold-Espinosa resummation (dimensional reduction) +is the best approach. But what if the high-temperature +expansion does not apply? Many then turn to Parwani +resummation, where thermal masses are inserted in +the full effective potential [11]. This can be imple- +mented with the help of a “thermal counterterm,” +effectively subtracting off terms that would otherwise be +doublecounted. +I will write the contribution to the MS -renormalized 1- +loop potential for a bosonic degree of freedom with square +mass m2 at temperature T as [12] +J(m2) = J0(m2) + T 4 +2π2 JB(m2 +T 2 ), +(10) +J0(m2) = m4 +64π2 +� +log +�m2 +µ2 +� +− 3 +2 +� +, +(11) +JB(y2) = +� +dx x2 log +� +1 − exp +� +− +� +x2 + y2 +�� +, +(12) +with µ the MS +scale.2 With this notation we can im- +plement Parwani resummation of the 1-loop potential by +2 The constant −3/2 in J0 is slightly different for a vector bo- +son [13]. Here I also ignore the analogous fermionic functions. + +3 +adding the term +VParwani = J(m2 +eff) − +∞ +� +i=0 +(δm2 +T )iJ(i)(m2), +(13) +where δm2 +T is the thermal counterterm that implements +the resummation. Each power of (δm2 +T ) raises the loop- +order: the subtracted terms should be sorted into their +appropriate loop orders to cancel terms and prevent +double-counting. Of course, if any other loop functions +are resummed then the corresponding terms should be +subtracted in a similar fashion. +Parwani resummation does not depend directly on the +high-temperature expansion, and as long as one is consis- +tent in subtracting the diagrams at each loop order, the +resummation mehtod should not introduce any problems. +Hence we might draw the conclusion that the Parwani +method is a safer bet when the high-temperature expan- +sion does not apply—as is often the case in phenomeno- +logical models with many particles of varying masses. +(See the subsection on linear terms for an example of +what can go wrong if one is not consistent.) +On the other hand, Arnold-Espinosa resummation has +conceptual clarity: only the modes which require resum- +mation are resummed. Double-counting is never even an +issue. +Because +the +two +methods +have +their +respective +strengths, they are compared against each other in nu- +merical studies (see e.g. [14, 15], and the studies cited +within). Yet such comparisons miss the point. +In fact, Arnold & Espinosa originally compared their +resummation method to that of Parwani, concluding that +The sum of all these contributions repro- +duces the previous resummed, two-loop result +(3.29), giving an explicit example that the ex- +act details of the resummation prescription +are unimportant. [1, p. 25] +But if this is case, how come the previously mentioned +studies keep finding that the methods give different re- +sults? The reason is that the conclusion of Arnold & Es- +pinosa rests on a few assumptions. First, that the high- +temperature expansion applies. Second, the existence of +a consistent power-counting—an expansion in a small pa- +rameter. +Any perturbative study of the electroweak phase tran- +sition in which the high-temperature expansion does not +apply, or if it is not an expansion in a small parameter, +will show a difference between these two methods. This +could give the illusion that the two methods give different +results, but a difference between the two methods sim- +ply reveals that the perturbative expansion, one way or +another, is not working. +This is why I stress that the true lesson of Arnold & Es- +pinosa’s paper is not their resummation method—which +now is supplanted by dimensional reduction anyway— +but the principle of using strict perturbative expansions. +Arnold & Espinosa also emphasize that consistency of +any one resummation method requires the protection of a +hierarchy of scales, as in this discussion of the momentum +dependent self-energy Π(K): +Resummation only affects perturbative ex- +pansions when Π cannot be treated as a per- +turbation to the inverse propagator G−1. This +happens only when K2 ≪ T 2, in which case +Π(K) ≈ Π(0). [1, p. 20] +Resumming masses is only reasonable if such a hierarchy +exists. To see this, consider a propagator D(p2) improved +by including the momentum-dependent self energy Π(p2), +with x ≪ 1 as a loop-counting parameter, +D(p2) = +1 +p2 + m2 + Π(p2), +(14) +Π(p2) = xΠ1(p2) + x2Π2(p2) + . . . . +(15) +By expanding the self energy in powers of momentum, +Π(p2) = Π(0) + p2Π′(0) + . . . , +(16) +we can see that we need +p2Π′(0) ≪ Π(0) +(17) +for the momentum-expansion to apply. Now we can con- +sider a generic example, in which the self-energy con- +tains contributions from a heavier particle with mass +M 2 ∼ m2/x. Then +xΠ1(p2) ∼ xM 2 + xp2. +(18) +In this case, with p2 ∼ m2, we have +m2 + Π(p2) = m2 + x +� +Π1(0) + p2Π′ +1(0) +� ++ O(x2) +∼ m2 + xM 2 + xp2 + O(x2). +(19) +So the hierarchy between m2 and M 2 tells us that we can +define a new effective mass +m2 +eff ∼ m2 + xM 2, +(20) +and that the momentum dependence of the self energy +can be neglected for this purpose. At higher orders the +momentum dependence can become important, and can +be included through higher order derivative operators in +the action. +If there is no hierarchy of scales, then we are not justi- +fied in simply resumming the mass: the whole self-energy +is needed.3 +3 As Arnold & Espinosa note in their appendix C [1, p. 58], this +poses a challenge for super-daisy resummations [16] and partial +dressing resummation [17–19], in which a gap equation is solved +to find the resummed mass. + +4 +ESTABLISH A HIERARCHY OF SCALES +In this section I apply the the previous considerations to +a few models, by studying different scale hierarchies. +Hierarchy I +To begin with, consider a theory with one mass scale m +at temperature T such that the hierarchy +m ≪ πT +(21) +holds. For concreteness we can consider the pure scalar +theory defined in equation (1), with −ν2 ∼ λT 2. The +hierarchy implies that we can integrate out the heavy +modes, to get a resummed theory of light modes with +φ2-coefficient ν2 +eff as given in equation (5). A cheap +way to implement this resummation is to use the ring- +improved potential of equation (9) defined by Arnold & +Espinosa [1]. +The more systematic version of this resummation +method is called dimensional reduction [2, 3, 9, 10]. +To give a brief motivation: the equilibrium quantities +of a finite temperature field theory can be studied in +an imaginary-time formalism. Bosonic fields are periodic +over this “time” direction, with period ∼ 1/T where T is +the temperature. This allows the fields to be decomposed +into Matsubara modes ϕn with masses m2+(2πnT )2. The +modes with n ̸= 0 are analogous to heavy particles, and +can be integrated out with standard field-theory tech- +niques. See [20–22] for modern and pedagogical reviews +of this concept; see [23] for software which automatizes +the matching procedure for generic models. +We are then left with a three-dimensional Euclidean +theory of zero-modes. The potential is now +V (φ3) = 1 +2m2 +3φ2 +3 + 1 +4!λ3φ4 +3, +(22) +with 3D quantities (note their mass dimensions) +φ2 +3 = φ2 +T + . . . , +(23) +m2 +3 = −ν2 + 1 +24λT 2 + . . . , +(24) +λ3 = λT + . . . . +(25) +Here the ellipses hide higher order corrections which can +be found by performing the matching to higher orders. +Now we can revisit why the pure scalar theory does +not support a radiative barrier. The contribution to the +1-loop potential in the 3D EFT for a field of square mass +M 2(φ3) is +f3(M 2) = − 1 +12π +� +M 2�3/2 . +(26) +Adding this to the tree-level potential gives +VLO(φ3) = 1 +2m2 +3φ2 +3 − +1 +12π +� +m2 +3 + 1 +2λ3φ2 +3 +�3/2 ++ 1 +4!λ3φ4 +3, +(27) +And the power-counting in equation (8) reads +φ3 ∼ +� +λ3, m2 +3 ∼ λ2 +3 =⇒ M(φ3) ∼ λ3. +(28) +Familiarly, even though a priori this counting seems +innocuous—the +high-temperature +expansion +clearly +applies—there is a problem here, as can be seen by di- +mensional analysis. Because this theory only has one cou- +pling constant λ3 (with mass-dimension 1) and one mass +M(φ3), each time we go up in loop order we must add a +factor of λ3. The dimensions then force us to remove one +factor of M(φ3), leading to the sequence +M 3, λ3M 2, λ2 +3M, λ3 +3, . . . +(29) +for loop orders one, two, three, four, . . . . And from the +power-counting in equation (28) we know M ∼ λ3. Hence +all loop orders contribute at the same order in perturba- +tion theory: λ3 +3. +More formally, rescaling the field as φ3 → √m3φ3, the +momenta as pi → m3pi and defining x = λ3/m3 yields +the dimensionless potential +V (φ3) +m3 +3 += 1 +2φ2 +3 + 1 +4!xφ4 +3. +(30) +The only coupling constant of this theory is x, and the +power-counting gives x ∼ 1. This is not an expansion in a +small parameter and our conclusions based on it cannot +be trusted. Its barriers are only mirages. +A more intuitive formulation: in order for the 1-loop +correction to change the shape of the potential, it needs +to be big enough to affect the classical potential. We need +something heavy to amplify the 1-loop potential, but we +only have one scalar—and it can not be heavier than +itself. +Hierarchy II +Next I will consider a scale hierarchy with an intermedi- +ate scale, +m ≪ M ≪ πT. +(31) +This scale hierarchy offers a rich set of possibilities. One +example is the second case considered by Arnold & Es- +pinosa, that of a gauge theory at high temperature. Again +we are dealing with a 3D EFT of Abelian Higgs, and the +authors state +The new procedure for resummation may be +viewed as follows: first integrate out all of +the heavy modes to obtain an effective the- +ory for momenta k0 = 0 and k ≪ T , next +integrate out the vectors (and scalars with +k ∼ M) to obtain an effective theory for mo- +menta k0 = 0 and k ≪ M, and only then fi- +nally integrate out scalar loops controlled by +k ∼ m ≪ M. [1, p. 30] + +5 +First a dimensional reduction is performed, and then the +gauge boson is integrated out. Note that the procedure +requires integrating out a field whose mass and couplings +depend on the background field of a lighter scalar. This +will yield a non-polynomial effective action [4, 24]. +In the end we get the leading-order potential and pa- +rameters +VLO(φ3) = 1 +2m2 +3φ2 +3 − +1 +12π +� +g2 +3φ2 +3 +�3/2 + 1 +4!λ3φ4 +3, +(32) +m2 +3 = m2 + g2 +12T 2 + 1 +18λT 2 + . . . , +(33) +λ3 = λT + . . . , +(34) +g2 +3 = g2T + . . . , +(35) +which indeed has a barrier. And this barrier is not a mi- +rage [1]. +By deriving the masses of the scalars from this po- +tential it is possible to extend previous studies [25, 26] +to study phase transitions in this theory accurately and +consistently [6–8, 27]. +Hierarchy III +We can also imagine another hierarchy in which the heavy +field is so heavy it is not excited by the temperature T , +m ≪ πT ≪ M. +(36) +In this case we should first integrate out all the modes +of the heavy field, and then integrate out the non-zero +Matsubara modes of the light field. See [4] for an example. +Hierarchy IV +Consider a heavy field at a scale close to the temperature +T , +m ≪ M ∼ πT. +(37) +In this case, the high-temperature expansion does not +apply to the field of mass M. But neither can the tem- +perature be neglected when integrating it out, as in equa- +tion (36). However, the high-temperature expansion still +applies for the light field of mass m. There should still +exist a 3D EFT for the zero-mode of the light field. This +method of “partial dimensional reduction” is not widely +studied, but see [28–31] for a few studies. +Here I want to highlight another example: a variant +of the Coleman-Weinberg (CW) model [13] as studied +in [32]. This model features radiative symmetry breaking: +there is no symmetry breaking at tree-level, but there is +at 1-loop level. This is the model of Abelian Higgs—a +complex scalar charged under a U(1) gauge field—with a +small positive mass term: +Vcl(φ) = 1 +2m2φ2 + 1 +4!λφ4, +(38) +This potential is of comparable size to the 1-loop contri- +bution of the gauge boson when λ ∼ g4 and m2 ∼ g4σ2 +with σ a characteristic size of the VEV. +There is a clear hierarchy of scales: the gauge boson is +heavier than the scalars and can be integrated out. This +gives a modified potential +VLO(φ) = 1 +2m2φ2 + 1 +4!λφ4 + 3JCW(g2φ2), +(39) +were JCW is the same as J0 of equation (11) but with +−3/2 replaced by −5/6. This potential has a non-zero +minimum, and it is from this potential which we should +find the scalar masses, +m2 +H(φ) = ∂2VLO(φ), +(40) +m2 +G(φ) = ∂VLO(φ) +φ +. +(41) +This is a consistent mass resummation derived from +power-counting rules, with a hierarchy of scales that pro- +tects it from double-counting diagrams and other issues. +This model has two different minima with a barrier +in between—could there be a first order phase transition +between them? To approach this question, we can assume +that the high-temperature expansion applies, and that a +φ3 barrier is induced: +VLO(φ) = 1 +2m2 +effφ2 + 1 +4!λφ4 − +3 +12πT g3φ3 ++ 3JCW(g2φ2), +(42) +but by balancing the powers of this expression we find +gφ ∼ T. +(43) +Which implies that the high-temperature expansion does +not apply. +But really we are asking for too much: we do not need +the high-temperature expansion to apply to the gauge +field. This model already has two minima, we do not need +the temperature effects to create a barrier. We only need +it to shift the energy of the different minima such that a +phase transition can occur [9, 33]; the resulting potential +reads +VLO(φ) = 1 +2m2 +effφ2 + 1 +4!λφ4 ++ 3JCW(g2φ2) + 3 T 4 +2π2 JB(g2φ2). +(44) +Balancing the powers here gives the same counting as in +equation (43). +But the question remains what to do with the scalar +field. After all, it is this field which will potentially un- +dergo a transition. +Because the high-temperature expansion still applies +to the scalar field, we should treat it using a 3D EFT +as before. To reach this EFT we integrate out the heavy +modes of the scalars: the non-zero Matsubara modes and + +6 +the high-momentum modes of the Matsubara zero-mode. +At the same time, we also integrate out all modes of the +vector. What we end up with is a Euclidean 3D EFT with +potential +VLO(φ3) = 1 +2m2 +3φ2 +3 + 1 +4!λ3φ4 +3 ++ 3JCW +T +(g2φ2) + 3 T 3 +2π2 JB(g2φ2). +(45) +The 3D parameters are determined by matching with the +4D theory. In this case we find +m2 +3 = m2 + 1 +18λT 2 + . . . , +(46) +λ3 = λT + . . . , +(47) +φ2 +3 = φ2 +T + . . . . +(48) +The difference between these expressions and those of +regular dimensional reduction of Abelian Higgs is that +here the non-zero Matsubara modes of the vector bo- +son do not contribute directly to the Wilson coefficients +(compare equations (46) and (33) and note the missing +g2T 2 term). Instead the vector modes contribute through +the non-polynomial term in the effective potential. This +contribution will in the end propagate to the mass of the +scalar. +To find the correct resummation to use in this theory +we use the same derivatives as in equations (40) and (41), +but with the potential given by equation (45). I empha- +size that this resummation contains parts that are not +utilizing the high-temperature expansion. And yet this +should be a wholly consistent resummation. Furthermore, +because the high temperature expansion applies to the +scalar field which undergoes the phase transition, the ma- +chinery of thermal escape (tunnelling at finite tempera- +ture) [5, 34, 35] should apply and all the usual formulas +ought to carry over.4 Though a detailed analysis of the +power-counting scheme and its convergence is warranted. +DISSOLVE ILLUSORY PROBLEMS +Studies of the electroweak phase transition have a long +laundry list of problems: gauge dependence, strong renor- +malization scale dependence, the Goldstone boson catas- +trophe, IR divergences, imaginary potentials, mirages, +perturbative breakdown, resummation method depen- +dence, and linear terms. Many of these problems were +4 A funny corollary of this is that a CW-like SU(2) gauge the- +ory would automatically suppress sphaleron transitions after the +phase transition. The power-counting in equation (43) implies +that the gauge field is not excited at the phase transition temper- +ature since it is too heavy. The suppression of thermal sphalerons +in models with radiative symmetry breaking hence make them +natural candidates for electroweak baryogenesis (see [36] for a +numerical study in agreement with this claim). +recently studied in [15], where it was shown that some +of them can yield big quantitative and qualitative uncer- +tainties. +I argue that these problems are dissolved if one uses a +consistent and strict perturbative expansion. +Gauge dependence +The gauge dependence of the effective action, and in par- +ticular the effective potential, is well-known and captured +in the famous Nielsen identities [37, 38]. Essentially the +effective potential is only gauge-invariant when evaluated +at an extremum—at a physical point. But in perturba- +tion theory there are implementation details: to get a +gauge-independent result we must use a strict perturba- +tive expansion. So if the effective potential is expanded +as +V = V0 + xV1 + x2V2 + . . . , +(49) +then we must find the extrema perturbatively, +φ = φ0 + xφ1 + x2φ2 + . . . , +(50) +by inserting this expansion of φ into the expansion of V , +and extremizing the potential order-by-order: +∂V |φ +!= 0 =⇒ ∂V0|φ0 +!= 0 +(51) +φ1 +!= − ∂V1 +∂2V0 +���� +φ0 +(52) +... +This expansion is sometimes called a tadpole expansion, +since it effectively reinserts scalar tadpoles into the 1PI +diagrams of the effective potential [38]. +Though it is well-established that the strict expan- +sion above gives gauge-independent results, there has re- +mained some confusion if it can also give accurate re- +sults. The strict expansion was popularized in [39] under +the name ℏ-expansion (now sometimes called the PRM +method). The authors then expressed concern that the +ℏ-expansion required a strict loop counting, while any +resummation necessarily mixes loop orders. +The way out of this dilemma is to realize that though +a strict expansion is necessary, it does not have to be a +loop expansion. All that is required is that the pertur- +bative expansion is performed using a consistent power- +counting [40]. In [27] it was shown that such a strict +expansion works if the expansion parameter x is small. +There is no conflict between gauge-independence and ac- +curate results. On the contrary they go hand-in-hand, as +predicted by Arnold & Espinosa:5 +5 Unfortunately this is not reflected in the wider literature. There +is much confusion as is evident by sampling the papers citing [39]. + +7 +In the best of worlds it would be nice to com- +pute results in a variety of gauges and check +that physical quantities are indeed gauge in- +variant, but we have not had the perseverance +to do so. [1, p. 26] +Strong renormalization scale dependence +There are many studies showing a strong renormalization +scale dependence in perturbative calculations of phase +transition quantities [15, 41, 42]. The problem is that re- +summations mix loop orders, which messes up the ordi- +nary cancellation between implicit running of parameters +and explicit running of loop-functions. As the thermal +masses arise at 1-loop, their running must be cancelled +by the next loop order: two loops. +As such, the solution is to use dimensional reduction +and calculate up to two-loop order. Constructing the +3D EFT consistently resums large contributions, and the +running within this EFT is tame. +The Goldstone boson catastrophe +To understand the possibility of IR-divergences we can +consider the form of the zero-temperature 1-loop function +J0, +J0 ∼ m4 log m2 =⇒ J′′ +0 ∼ log m2. +(53) +The second derivative of this function diverges in the +m2 → 0 limit. This divergence indicates two related prob- +lems. +The first problem can be seen if we think of these +derivatives as insertions of interactions. Then the diver- +gence of J′′ +0 implies that the 3-loop potential will diverge +in the same limit. This is known as the Goldstone bo- +son catastrophe, since the Goldstones are massless in the +broken minimum. The literature implies a resummation +is necessary to cure this divergence [43–45]. But this is +not correct. In fact, a resummation is not necessary to +remove the divergence in a regular loop expansion [46]. +It is enough to simply perform a strict expansion: any +IR divergences of the potential are cancelled by corre- +sponding divergences in the background field. However, +if a modified power-counting is used—i.e. when there is +a hierarchy of scales—resummation becomes necessary +again. +The second related problem arises if one attempts to +use a mass-dependent renormalization scheme and tries +to match the measured masses of the scalars through sec- +ond derivatives of the one-loop potential. The Goldstones +then again cause divergences, which are typically regu- +lated away with an IR regulator inserted by hand. +In my opinion it is much simpler to use MS and not +have to deal with these issues. But there is no real incon- +sistency with using the mass-dependent scheme together +with an IR cutoff. +IR divergences +Attempting to use the ℏ-expansion (strict loop expan- +sion) to find the critical temperature in a theory with a +radiative barrier leads to IR divergences [39, 47]. The di- +vergence comes from expanding the critical temperature +around T0, +Tc = T0 + ℏT1 + . . . , +(54) +∆V |Tc +!= 0 =⇒ m2 +3 +�� +T0 +!= 0. +(55) +In the strict loop expansion, T0 coincides with the tem- +perature where the classical potential of the 3D EFT +has a second order phase transition, as stated in equa- +tion (55). When evaluating the two-loop potential at this +temperature, there are divergences ∼ log m2 +3 +�� +T0. What is +going wrong? +The problem is that the wrong power-counting is used. +By using the modified power-counting λ ∼ g3 which a +radiative barrier necessitates [1], one instead finds the +critical temperature +Tc = TLO + xTNLO . . . , +(56) +∆V |Tc +!= 0 =⇒ ∆VLO|TLO +!= 0. +(57) +Expanding around TLO does not feature IR diver- +gences [27, 40]. Power-counting together with a strict +perturbative expansion is IR safe. +Imaginary potentials +A potential with a non-zero imaginary part signals an +instability: the 1-loop potential can develop an imaginary +part if evaluated at a field-value where a square mass is +negative. This can happen close to the broken minimum +where the Goldstone square mass changes sign. But it +can also happen at the origin for temperatures below T0, +where the scalar mass terms are negative. +These circumstances can arise if one is mixing loop +orders in V (φ)—if one is not using a strict perturba- +tive expansion. When searching field space and scanning +in temperatures this necessitates taking the real part of +the potential in order to get a sensible answer. But this +merely cures a symptom and does not fix the real prob- +lem: unstable modes are influencing the calculation. +But if one uses the resummed leading order effective +potential VLO in a strict perturbative expansion, then the +critical temperature and broken minimum can be found +order by order. The leading order quantities TLO, φLO +are found from VLO and subsequent orders are evalu- +ated there: VNLO|TLO,φLO. No imaginary parts develop; +all modes are correctly accounted for. + +8 +Mirages +Arnold & Espinosa demonstrated that the Z2 symmetric +pure scalar theory appears to have a barrier but in fact +does not, as we saw in an earlier section. In that case the +mirage was induced from a breakdown of perturbation +theory. More generally, mirages can arise when perturba- +tive orders are out of control. Either because the expan- +sion parameter is too large, as above, or when orders are +mixed haphazardly. +Then the only way to be safe against them is to use +a consistent power-counting, and perform a strict expan- +sion in a small parameter. +Perturbative breakdown +In the extreme case of a perturbative breakdown, mirages +are indicative of a larger problem: we cannot trust the +perturbative expansion. +Certain perturbative problems can be fixed by reorder- +ing perturbation theory—by resummations—while oth- +ers are incurable. Famously, non-Abelian gauge theories +suffer from the Linde problem: at high temperatures the +gauge boson must develop a “magnetic mass” M ∼ g2T +to cure IR divergences at four loops and higher [48]. This +results in a complete perturbative breakdown; perturba- +tive methods cannot reach O(M 3 ∼ g6T 4) (four loops). +To be fair, power-counting and a strict expansion can- +not solve this problem. But it can offer some perspec- +tive. In SU(2) gauge theory with a radiative barrier, the +Linde problem affects the sixth order and higher in strict +perturbation theory. The first five orders are calculable +and even the first three orders offer good accuracy for a +wide range of expansion parameter values [27]. Hence the +Linde problem should not occupy too much space in our +minds when studying phase transitions. +Resummation method dependence +Comparing resummation methods and finding a differ- +ence indicates that at least one of them is wrong. But it +cannot tell you which result is the correct one, or if any of +them are. Instead it is better to look to the assumptions +made on which the resummations are based. +To be confident in which resummation method to use +we should instead derive it from a consistent power- +counting, and make sure that perturbation theory is con- +verging. In the end there should only be one resummation +method available: that which is implied by the hierarchy +of scales. +Linear terms +Terms linear in φ will prevent the existence of the sym- +metric minimum at high temperatures. Such terms con- +tradict our usual understanding that the symmetry is +restored in this limit and in the early universe. In the +beginning of the 1990s there were several resummation +methods that produced such linear terms [16, 49], and +there was some doubt whether the linear terms should +exist or not. +But convincing argument against such terms can be es- +tablished from an effective field theory perspective, using +the methods of power-counting. In [50], the authors ar- +gue that the existence of an IR cutoff—the magnetic mass +∼ g2T discussed above—means that the effective poten- +tial must be analytic in |φ|2 as φ → 0. This prevents the +existence of linear terms. Any consistent resummation +method must respect this constraint. +In a more modern setting, it is sometimes argued +that Parwani resummation can give rise to such linear +terms [41]. But I think this is unfair to Parwani resumma- +tion. The linear terms arise if one uses a high-temperature +approximation for the thermal counter-term, but main- +tains the full unexpanded integrals in the potential [51]. If +the high-temperature expansion does not apply, then this +is inconsistent and there will remain uncancelled terms +that the counter-term insertion procedure cannot handle. +I think it is more fair to blame a faulty power-counting: +linear terms can arise if one is not using a consistent +expansion. +GO FORTH AND COUNT POWERS +The lesson of Arnold & Espinosa’s paper is a refrain that +is common in physics, and a lesson we have learnt again +and again: the proper way of doing physics is by way of +the effective field theory. +More formally, a strict perturbative expansion derived +from an effective field theory is an asymptotic expansion. +And asymptotic expansions are unique [52]: there should +not be any question as to which resummation method to +use once you are settled on an approximation scheme.6 +Any resummation method that will stand the test of time +must be implementing a power-counting protected by a +hierarchy of scales. +In the previous sections I showcased the strength of +power-counting by demonstrating how it dissolves the +laundry list of confusion surrounding phase transition +calculations (mirages, imaginary potentials, scheme de- +pendence, . . . ). But if it is so powerful, why is not every- +one using it? +There is one apparent drawback with applying the +method of power-counting: some amount of thought has +to be applied before the calculation starts. This means +that scanning large sections of parameter space with a +single method is out of the question. There is no silver +6 The uniqueness of the coefficient also implies the perturbative +expansion is gauge invariant order-by-order [27]. + +9 +bullet which can work for all of parameter space. Instead +one should divide parameter space into different regions +in which different perturbative expansions apply. +But how small should λ be before we start counting +it as g3 instead of g2? This is a question with no clear +answer within perturbation theory. The best we can do is +to look to lattice data to find approximate regions where +a first order phase transition occurs or not, and cross- +check our power-counting against it. +But I argue that this is a good problem to have, in +contrast to the dangerous confusions of the laundry list +which comes from using a faulty perturbative expansion. +The question of how large different contributions are is +an honest quantitative physical question. You have to be +realistic about these things. +On the other hand, much have been said about the +apparent small size of uncertainty due to gauge depen- +dence [53]. For a given gauge fixing method, the results +do not differ much between Landau gauge (ξ = 0) and +reasonable values of ξ. It is argued that even though a +gauge dependent result is uncomfortable, it is not a large +quantitative issue and can be ignored [41, 54]. +But I think this is a distraction from the true issue. +The real problem with gauge dependence was never that +one finds a span of values for different ξ and have to pick +one of them at the expense of accuracy. The real problem +is that gauge dependence signals that something is wrong +with the perturbative expansion. And when something is +wrong with perturbation theory, we are at the mercy of +the other problems in the laundry list.7 Gauge depen- +dence signals that we have lost control of perturbation +theory. By not taking the signal seriously we are bound +to be confused. +The reasons above are why I encourage anyone in- +terested in performing a perturbative study of a phase +transition to always begin with the power-counting, and +to monitor the convergence of perturbation theory—by +comparing the different orders to each other and check- +ing that the renormalization scale dependence is under +control. +Whether the high temperature expansion applies or +not, you should be able to establish a hierarchy of scales +and derive the correct mass resummation scheme. And if +no hierarchy exists you should not perform a resumma- +tion. There are many phenomenological models in which +first order phase transitions seem possible, but the exis- +tence of which have not yet been established using power- +counting.8 To do so would enable us to study phase tran- +sitions in such models consistently and accurately. +This is what Arnold & Espinosa tried to teach us 30 +years ago, but we failed to learn. But it is not too late. +Go forth and count powers. +ACKNOWLEDGMENTS +I thank Andreas Ekstedt and Tuomas Tenkanen for their +comments on the manuscript, and Oliver Gould for en- +lightening discussions. +[1] P. B. +Arnold and O. Espinosa, +Phys. Rev. D47, +3546 +(1993), +[Erratum: +Phys. +Rev.D50,6662(1994)], +arXiv:hep-ph/9212235 [hep-ph]. +[2] E. Braaten and A. Nieto, Phys. Rev. D51, 6990 (1995), +arXiv:hep-ph/9501375 [hep-ph]. +[3] K. +Kajantie, +M. +Laine, +K. +Rummukainen, +and +M. E. Shaposhnikov, Nucl. Phys. B 458, 90 (1996), +arXiv:hep-ph/9508379. +[4] J. Hirvonen, (2022), arXiv:2205.02687 [hep-ph]. +[5] O. Gould and J. Hirvonen, Phys. Rev. D 104, 096015 +(2021), arXiv:2108.04377 [hep-ph]. +[6] J. Hirvonen, J. L¨ofgren, M. J. Ramsey-Musolf, P. Schi- +cho, +and T. V. I. Tenkanen, JHEP 07, 135 (2022), +arXiv:2112.08912 [hep-ph]. +[7] J. L¨ofgren, M. J. Ramsey-Musolf, P. Schicho, and T. V. I. +Tenkanen, (2021), arXiv:2112.05472 [hep-ph]. +[8] A. +Ekstedt, +Phys. +Rev. +D +106, +095026 +(2022), +arXiv:2205.05145 [hep-ph]. +7 Though to be fair, varying ξ can lead not only to quantitative +uncertainty, but also to qualitative differences. Sometimes bar- +riers disappear, sometimes new minima are generated, as ξ is +varied [15, 27, 55]. +8 An exercise for the reader: develop a power-counting for a radia- +tive barrier to occur in the inert doublet model. +[9] P. H. Ginsparg, Nucl. Phys. B170, 388 (1980). +[10] T. Appelquist and R. D. Pisarski, Phys. Rev. D23, 2305 +(1981). +[11] R. R. Parwani, Phys. Rev. D45, 4695 (1992), [Erratum: +Phys. Rev.D48,5965(1993)], arXiv:hep-ph/9204216 [hep- +-ph]. +[12] L. Dolan and R. Jackiw, Phys. Rev. D 9, 3320 (1974). +[13] S. R. Coleman and E. J. Weinberg, Phys. Rev. D 7, 1888 +(1973). +[14] K. Kainulainen, V. Keus, L. Niemi, K. Rummukainen, +T. V. I. Tenkanen, +and V. Vaskonen, JHEP 06, 075 +(2019), arXiv:1904.01329 [hep-ph]. +[15] P. Athron, C. Balazs, A. Fowlie, L. Morris, G. White, +and Y. Zhang, (2022), arXiv:2208.01319 [hep-ph]. +[16] D. E. Brahm and S. D. H. Hsu, (1991). +[17] C. G. Boyd, D. E. Brahm, and S. D. H. Hsu, Phys. Rev. +D 48, 4963 (1993), arXiv:hep-ph/9304254. +[18] D. Curtin, P. Meade, and H. Ramani, Eur. Phys. J. C +78, 787 (2018), arXiv:1612.00466 [hep-ph]. +[19] D. +Curtin, +J. +Roy, +and +G. +White, +(2022), +arXiv:2211.08218 [hep-ph]. +[20] O. Gould, JHEP 04, 057 (2021), arXiv:2101.05528 [hep- +-ph]. +[21] L. Niemi, P. Schicho, and T. V. I. Tenkanen, Phys. Rev. +D 103, 115035 (2021), arXiv:2103.07467 [hep-ph]. +[22] P. M. Schicho, T. V. I. Tenkanen, +and J. ¨Osterman, +JHEP 06, 130 (2021), arXiv:2102.11145 [hep-ph]. + +10 +[23] A. Ekstedt, P. Schicho, and T. V. I. Tenkanen, (2022), +arXiv:2205.08815 [hep-ph]. +[24] E. +J. +Weinberg, +Phys. +Rev. +D +47, +4614 +(1993), +arXiv:hep-ph/9211314. +[25] M. Karjalainen and J. Peisa, Z. Phys. C76, 319 (1997), +arXiv:hep-lat/9607023 [hep-lat]. +[26] K. Kajantie, M. Karjalainen, M. Laine, +and J. Peisa, +Nucl. Phys. B 520, 345 (1998), arXiv:hep-lat/9711048. +[27] A. Ekstedt, O. Gould, +and J. L¨ofgren, +(2022), +arXiv:2205.07241 [hep-ph]. +[28] M. Laine and M. Losada, Nucl. Phys. B 582, 277 (2000), +arXiv:hep-ph/0003111. +[29] M. Laine and K. Rummukainen, Nucl. Phys. B 597, 23 +(2001), arXiv:hep-lat/0009025. +[30] T. Brauner, T. V. I. Tenkanen, A. Tranberg, A. Vuorinen, +and D. J. Weir, JHEP 03, 007 (2017), arXiv:1609.06230 +[hep-ph]. +[31] M. Laine, P. Schicho, +and Y. Schr¨oder, Phys. Rev. D +101, 023532 (2020), arXiv:1911.09123 [hep-ph]. +[32] D. Metaxas and E. J. Weinberg, Phys. Rev. D 53, 836 +(1996), arXiv:hep-ph/9507381. +[33] A. D. Linde, Rept. Prog. Phys. 42, 389 (1979). +[34] A. +Ekstedt, +Eur. +Phys. +J. +C +82, +173 +(2022), +arXiv:2104.11804 [hep-ph]. +[35] A. Ekstedt, JHEP 08, 115 (2022), arXiv:2201.07331 [hep- +-ph]. +[36] T. Prokopec, J. Rezacek, and B. ´Swie˙zewska, JCAP 02, +009 (2019), arXiv:1809.11129 [hep-ph]. +[37] N. K. Nielsen, Nucl. Phys. B 101, 173 (1975). +[38] R. Fukuda and T. Kugo, Phys. Rev. D 13, 3469 (1976). +[39] H. H. Patel and M. J. Ramsey-Musolf, JHEP 07, 029 +(2011), arXiv:1101.4665 [hep-ph]. +[40] A. Ekstedt and J. L¨ofgren, JHEP 12, 136 (2020), +arXiv:2006.12614 [hep-ph]. +[41] D. Croon, O. Gould, P. Schicho, T. V. I. Tenkanen, and +G. White, JHEP 04, 055 (2021), arXiv:2009.10080 [hep- +-ph]. +[42] O. Gould and T. V. I. Tenkanen, JHEP 06, 069 (2021), +arXiv:2104.04399 [hep-ph]. +[43] S. +P. +Martin, +Phys. +Rev. +D +90, +016013 +(2014), +arXiv:1406.2355 [hep-ph]. +[44] J. Elias-Miro, J. R. Espinosa, and T. Konstandin, JHEP +08, 034 (2014), arXiv:1406.2652 [hep-ph]. +[45] J. R. Espinosa, M. Garny, +and T. Konstandin, Phys. +Rev. D 94, 055026 (2016), arXiv:1607.08432 [hep-ph]. +[46] A. Ekstedt and J. L¨ofgren, JHEP 01, 226 (2019), +arXiv:1810.01416 [hep-ph]. +[47] M. +Laine, +Phys. +Lett. +B +335, +173 +(1994), +arXiv:hep-ph/9406268. +[48] A. D. Linde, Phys. Lett. 96B, 289 (1980). +[49] M. E. Shaposhnikov, Phys. Lett. B 277, 324 (1992), [Er- +ratum: Phys.Lett.B 282, 483 (1992)]. +[50] M. Dine, R. G. Leigh, P. Y. Huet, A. D. Linde, +and +D. A. Linde, +Phys. Rev. D +46, +550 +(1992), +arXiv:hep-ph/9203203. +[51] M. Laine, M. Meyer, and G. Nardini, Nucl. Phys. B920, +565 (2017), arXiv:1702.07479 [hep-ph]. +[52] C. M. Bender, S. Orszag, +and S. A. Orszag, Ad- +vanced mathematical methods for scientists and engineers +I: Asymptotic methods and perturbation theory, Vol. 1 +(Springer Science & Business Media, 1999). +[53] M. Garny and T. Konstandin, JHEP 07, 189 (2012), +arXiv:1205.3392 [hep-ph]. +[54] P. Schicho, T. V. I. Tenkanen, and G. White, JHEP 11, +047 (2022), arXiv:2203.04284 [hep-ph]. +[55] J. Zuk, C. Balazs, A. Papaefstathiou, +and G. White, +(2022), arXiv:2212.04046 [hep-ph]. + diff --git a/etE4T4oBgHgl3EQfqA2v/content/tmp_files/load_file.txt b/etE4T4oBgHgl3EQfqA2v/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..8364ec478a6d1a17fb6eb9fa2a5bee7fe4e5e389 --- /dev/null +++ b/etE4T4oBgHgl3EQfqA2v/content/tmp_files/load_file.txt @@ -0,0 +1,750 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf,len=749 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='05197v1 [hep-ph] 12 Jan 2023 Stop comparing resummation methods Johan L¨ofgren1, ∗ 1Department of Physics and Astronomy, Uppsala University, Box 516, SE-751 20 Uppsala, Sweden I argue that perturbative studies of phase transitions should always establish a consistent power- counting through a hierarchy of scales, as it allows one to derive a correct resummation method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hence any confusion over which resummation method to use is misplaced once an approximation scheme is settled on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Furthermore, adopting a consistent power-counting scheme and performing a strict perturbative expansion dissolves many other common problems of such studies: gauge depen- dence, strong renormalization scale dependence, the Goldstone boson catastrophe, IR divergences, imaginary potentials, mirages (illusory barriers), perturbative breakdown, and linear terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I il- lustrate the strength of power-counting through various examples, and by revisiting highlights of Arnold & Espinosa’s pioneering paper [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' IT HAS BEEN 30 YEARS 30 years ago, Peter Arnold and Olivier Espinosa demon- strated how to treat first order phase transitions pertur- batively in a groundbreaking paper [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Their paper is widely cited in the phenomenological thermal field the- ory community, where similar methods are applied to ex- tensions of the Standard Model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In particular to study plausible consequences of a first order electroweak phase transition—such as electroweak baryogenesis and gravi- tational wave signals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In this work I revisit the arguments of [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In particu- lar, I show that adhering to a strict power-counting by- passes many current issues such as gauge dependence and infrared divergences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Doing so provides a direct link be- tween [1] and modern effective field theories [2–4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The overarching lesson is a familiar story: consistent resummations are implemented by integrating out heavy modes with respect to the physics you are interested in.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Establishing a hierarchy of scales will enable a derivation of the correct resummation method to use, and hence I argue that it is time we stop comparing resummation methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Instead we should start all our perturbative studies with a power-counting analysis, and make sure we are expanding in a small parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' For simplicity I focus on equilibrium quantities such as the critical temperature, but proper perturbative expan- sions are equally important when calculating dynamical quantities such as the bubble nucleation rate [5–8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I also demonstrate the use of these techniques for a class of models with radiative symmetry breaking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' SEE THROUGH MIRAGES Arnold & Espinosa point out how easy it is to misin- terpret perturbative calculations of phase transitions;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I will repeat part of their demonstration here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Consider a ∗ johan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='lofgren@physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='uu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='se real-scalar theory: Vcl(φ) = −1 2ν2φ2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4, (1) with ν2 ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Though this potential exhibits spontaneous symmetry breaking, it is a priori not clear if a first-order phase transition occurs at some high temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In particular, if the scalar’s field-dependent mass is m2(φ), the 1-loop effective potential is V1(φ) = 1 24m2(φ)T 2 − 1 12π � m2(φ) �3/2 T + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (2) Using the square mass m2(φ) = V ′′ cl (φ) gives the leading contribution V1(φ) = 1 48λT 2φ2, (3) where I discarded field-independent terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Now, for a phase-transition to occur, this term must be of similar size as the tree-level potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This indicates that we should add its contribution to Equation 1, VLO(φ) = 1 2 � −ν2 + 1 24λT 2 � φ2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4, (4) and instead consider VLO(φ) as the leading-order poten- tial.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This means that the high-temperature φ2-coefficient is ν2 eff = ν2 − 1 24λT 2, (5) which vanishes at temperature T0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' So this model seems to describe a second order phase transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But we can find the effective—resummed—square mass from the same potential, m2 eff(φ) = V ′′ LO(φ) = ν2 eff + 1 2λφ2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (6) Should we not use this improved mass in the 1-loop po- tential?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Doing so gives VLO(φ) = 1 2ν2 effφ2 − T 12π � ν2 eff + 1 2λφ2 �3/2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (7) 2 For temperatures just above T0, where ν2 eff ⪆ 0, there indeed appears to be a barrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' However, after looking a bit closer at the terms in- volved;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' which scale as φc ∼ √ λT, −ν2 eff(Tc) ∼ λ2T 2 =⇒ meff ∼ λT, (8) we see that the effective mass is small compared to the temperature T .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Though this confirms that the high- temperature expansion applies, this is actually bad news.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Loops involving the zero-Matsubara mode each come with a power of λT , and the only other mass scale is meff —the expansion parameter can only be λT/meff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' So once meff ∼ λT , as it is here, loops are not suppressed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In the words of Arnold & Espinosa: Now consider the loop expansion parame- ter λT/meff at the minimum φc ∼ √ λT corresponding to the apparent first-order phase transition of the ring-improved one- loop potential (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Eqn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='8) implies that meff(φc) ∼ λT , and so the loop expansion parameter is O(1), verifying that the ring- improved loop expansion cannot be trusted to distinguish between a first and second-order phase transition in this model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 7]1 I give two variations of this argument in the section on scale hierarchies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In fact, the real scalar theory studied above is known to not have a first-order phase transition [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The barrier turns out to have been fictitious all along.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Let us call such illusory barriers mirages, to emphasize the danger they pose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' We can contrast mirages with real radiative barriers, such as that of Abelian Higgs at high tempera- tures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This model is also studied in detail by Arnold & Espinosa, and they state The situation is quite different when the gauge sector is included.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' As we now review, the first-order phase transition seemingly de- scribed by the ring-improved effective poten- tial is associated with a small loop expansion parameter if the Higgs mass is sufficiently small.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The phase transition is therefore in- deed first order, and the ring-improved loop expansion is a valid tool for studying it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 7] The comparison of these different models by Arnold & Espinosa showcases how important it is to use a consis- tent power-counting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Changing the shape of the classical potential necessarily requires large perturbative correc- tions, and care must be taken to ensure that the expan- sion is performed in a small parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 1 The page numbers apply to the ArXiv version of [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' STOP COMPARING RESUMMATION METHODS The resummation method laid out by Arnold & Espinosa can be summarized as resumming the masses of the light modes by integrating out the heavy modes: [.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='] loops with quadratic UV divergences are of order λT 2 and are dominated by large mo- menta of order T .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' These O(λT 2) contribu- tions are all absorbed by using ring-improved propagators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 6-7] In essence we should only resum masses when the self- energy is of the same order, m2 eff ∼ Π(meff).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The resummation can at one loop be implemented by adding a ring-improved term to the effective potential, Vring(φ) = − T 12π �� m2 eff(φ) �3/2 − � m2(φ) �3/2� , (9) effectively replacing the mass of the particle in the zero- mode contribution to the effective potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' More formally, Arnold & Espinosa are constructing a three-dimensional Euclidean effective field theory in terms of Matsubara zero modes [9, 10], and their re- summation method can be viewed as the leading order prediction of this EFT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The process of constructing the 3D EFT is known as dimensional reduction, and it en- ables systematic extension of the resummation method to higher orders [2, 3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I discuss dimensional reduction further in the next section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' When the high-temperature expansion applies, Arnold-Espinosa resummation (dimensional reduction) is the best approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But what if the high-temperature expansion does not apply?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Many then turn to Parwani resummation, where thermal masses are inserted in the full effective potential [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This can be imple- mented with the help of a “thermal counterterm,” effectively subtracting off terms that would otherwise be doublecounted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I will write the contribution to the MS -renormalized 1- loop potential for a bosonic degree of freedom with square mass m2 at temperature T as [12] J(m2) = J0(m2) + T 4 2π2 JB(m2 T 2 ), (10) J0(m2) = m4 64π2 � log �m2 µ2 � − 3 2 � , (11) JB(y2) = � dx x2 log � 1 − exp � − � x2 + y2 �� , (12) with µ the MS scale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='2 With this notation we can im- plement Parwani resummation of the 1-loop potential by 2 The constant −3/2 in J0 is slightly different for a vector bo- son [13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Here I also ignore the analogous fermionic functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 3 adding the term VParwani = J(m2 eff) − ∞ � i=0 (δm2 T )iJ(i)(m2), (13) where δm2 T is the thermal counterterm that implements the resummation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Each power of (δm2 T ) raises the loop- order: the subtracted terms should be sorted into their appropriate loop orders to cancel terms and prevent double-counting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Of course, if any other loop functions are resummed then the corresponding terms should be subtracted in a similar fashion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Parwani resummation does not depend directly on the high-temperature expansion, and as long as one is consis- tent in subtracting the diagrams at each loop order, the resummation mehtod should not introduce any problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hence we might draw the conclusion that the Parwani method is a safer bet when the high-temperature expan- sion does not apply—as is often the case in phenomeno- logical models with many particles of varying masses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (See the subsection on linear terms for an example of what can go wrong if one is not consistent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=') On the other hand, Arnold-Espinosa resummation has conceptual clarity: only the modes which require resum- mation are resummed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Double-counting is never even an issue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Because the two methods have their respective strengths, they are compared against each other in nu- merical studies (see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [14, 15], and the studies cited within).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Yet such comparisons miss the point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In fact, Arnold & Espinosa originally compared their resummation method to that of Parwani, concluding that The sum of all these contributions repro- duces the previous resummed, two-loop result (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='29), giving an explicit example that the ex- act details of the resummation prescription are unimportant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 25] But if this is case, how come the previously mentioned studies keep finding that the methods give different re- sults?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The reason is that the conclusion of Arnold & Es- pinosa rests on a few assumptions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' First, that the high- temperature expansion applies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Second, the existence of a consistent power-counting—an expansion in a small pa- rameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Any perturbative study of the electroweak phase tran- sition in which the high-temperature expansion does not apply, or if it is not an expansion in a small parameter, will show a difference between these two methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This could give the illusion that the two methods give different results, but a difference between the two methods sim- ply reveals that the perturbative expansion, one way or another, is not working.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This is why I stress that the true lesson of Arnold & Es- pinosa’s paper is not their resummation method—which now is supplanted by dimensional reduction anyway— but the principle of using strict perturbative expansions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Arnold & Espinosa also emphasize that consistency of any one resummation method requires the protection of a hierarchy of scales, as in this discussion of the momentum dependent self-energy Π(K): Resummation only affects perturbative ex- pansions when Π cannot be treated as a per- turbation to the inverse propagator G−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This happens only when K2 ≪ T 2, in which case Π(K) ≈ Π(0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 20] Resumming masses is only reasonable if such a hierarchy exists.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To see this, consider a propagator D(p2) improved by including the momentum-dependent self energy Π(p2), with x ≪ 1 as a loop-counting parameter, D(p2) = 1 p2 + m2 + Π(p2), (14) Π(p2) = xΠ1(p2) + x2Π2(p2) + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (15) By expanding the self energy in powers of momentum, Π(p2) = Π(0) + p2Π′(0) + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (16) we can see that we need p2Π′(0) ≪ Π(0) (17) for the momentum-expansion to apply.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Now we can con- sider a generic example, in which the self-energy con- tains contributions from a heavier particle with mass M 2 ∼ m2/x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Then xΠ1(p2) ∼ xM 2 + xp2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (18) In this case, with p2 ∼ m2, we have m2 + Π(p2) = m2 + x � Π1(0) + p2Π′ 1(0) � + O(x2) ∼ m2 + xM 2 + xp2 + O(x2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (19) So the hierarchy between m2 and M 2 tells us that we can define a new effective mass m2 eff ∼ m2 + xM 2, (20) and that the momentum dependence of the self energy can be neglected for this purpose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' At higher orders the momentum dependence can become important, and can be included through higher order derivative operators in the action.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' If there is no hierarchy of scales, then we are not justi- fied in simply resumming the mass: the whole self-energy is needed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='3 3 As Arnold & Espinosa note in their appendix C [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 58], this poses a challenge for super-daisy resummations [16] and partial dressing resummation [17–19], in which a gap equation is solved to find the resummed mass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 4 ESTABLISH A HIERARCHY OF SCALES In this section I apply the the previous considerations to a few models, by studying different scale hierarchies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hierarchy I To begin with, consider a theory with one mass scale m at temperature T such that the hierarchy m ≪ πT (21) holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' For concreteness we can consider the pure scalar theory defined in equation (1), with −ν2 ∼ λT 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The hierarchy implies that we can integrate out the heavy modes, to get a resummed theory of light modes with φ2-coefficient ν2 eff as given in equation (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' A cheap way to implement this resummation is to use the ring- improved potential of equation (9) defined by Arnold & Espinosa [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The more systematic version of this resummation method is called dimensional reduction [2, 3, 9, 10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To give a brief motivation: the equilibrium quantities of a finite temperature field theory can be studied in an imaginary-time formalism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Bosonic fields are periodic over this “time” direction, with period ∼ 1/T where T is the temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This allows the fields to be decomposed into Matsubara modes ϕn with masses m2+(2πnT )2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The modes with n ̸= 0 are analogous to heavy particles, and can be integrated out with standard field-theory tech- niques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' See [20–22] for modern and pedagogical reviews of this concept;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' see [23] for software which automatizes the matching procedure for generic models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' We are then left with a three-dimensional Euclidean theory of zero-modes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The potential is now V (φ3) = 1 2m2 3φ2 3 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λ3φ4 3, (22) with 3D quantities (note their mass dimensions) φ2 3 = φ2 T + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (23) m2 3 = −ν2 + 1 24λT 2 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (24) λ3 = λT + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (25) Here the ellipses hide higher order corrections which can be found by performing the matching to higher orders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Now we can revisit why the pure scalar theory does not support a radiative barrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The contribution to the 1-loop potential in the 3D EFT for a field of square mass M 2(φ3) is f3(M 2) = − 1 12π � M 2�3/2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (26) Adding this to the tree-level potential gives VLO(φ3) = 1 2m2 3φ2 3 − 1 12π � m2 3 + 1 2λ3φ2 3 �3/2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λ3φ4 3, (27) And the power-counting in equation (8) reads φ3 ∼ � λ3, m2 3 ∼ λ2 3 =⇒ M(φ3) ∼ λ3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (28) Familiarly, even though a priori this counting seems innocuous—the high-temperature expansion clearly applies—there is a problem here, as can be seen by di- mensional analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Because this theory only has one cou- pling constant λ3 (with mass-dimension 1) and one mass M(φ3), each time we go up in loop order we must add a factor of λ3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The dimensions then force us to remove one factor of M(φ3), leading to the sequence M 3, λ3M 2, λ2 3M, λ3 3, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (29) for loop orders one, two, three, four, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' And from the power-counting in equation (28) we know M ∼ λ3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hence all loop orders contribute at the same order in perturba- tion theory: λ3 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' More formally, rescaling the field as φ3 → √m3φ3, the momenta as pi → m3pi and defining x = λ3/m3 yields the dimensionless potential V (φ3) m3 3 = 1 2φ2 3 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='xφ4 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (30) The only coupling constant of this theory is x, and the power-counting gives x ∼ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This is not an expansion in a small parameter and our conclusions based on it cannot be trusted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Its barriers are only mirages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' A more intuitive formulation: in order for the 1-loop correction to change the shape of the potential, it needs to be big enough to affect the classical potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' We need something heavy to amplify the 1-loop potential, but we only have one scalar—and it can not be heavier than itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hierarchy II Next I will consider a scale hierarchy with an intermedi- ate scale, m ≪ M ≪ πT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (31) This scale hierarchy offers a rich set of possibilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' One example is the second case considered by Arnold & Es- pinosa, that of a gauge theory at high temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Again we are dealing with a 3D EFT of Abelian Higgs, and the authors state The new procedure for resummation may be viewed as follows: first integrate out all of the heavy modes to obtain an effective the- ory for momenta k0 = 0 and k ≪ T , next integrate out the vectors (and scalars with k ∼ M) to obtain an effective theory for mo- menta k0 = 0 and k ≪ M, and only then fi- nally integrate out scalar loops controlled by k ∼ m ≪ M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 30] 5 First a dimensional reduction is performed, and then the gauge boson is integrated out.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Note that the procedure requires integrating out a field whose mass and couplings depend on the background field of a lighter scalar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This will yield a non-polynomial effective action [4, 24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In the end we get the leading-order potential and pa- rameters VLO(φ3) = 1 2m2 3φ2 3 − 1 12π � g2 3φ2 3 �3/2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λ3φ4 3, (32) m2 3 = m2 + g2 12T 2 + 1 18λT 2 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (33) λ3 = λT + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (34) g2 3 = g2T + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (35) which indeed has a barrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' And this barrier is not a mi- rage [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' By deriving the masses of the scalars from this po- tential it is possible to extend previous studies [25, 26] to study phase transitions in this theory accurately and consistently [6–8, 27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hierarchy III We can also imagine another hierarchy in which the heavy field is so heavy it is not excited by the temperature T , m ≪ πT ≪ M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (36) In this case we should first integrate out all the modes of the heavy field, and then integrate out the non-zero Matsubara modes of the light field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' See [4] for an example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hierarchy IV Consider a heavy field at a scale close to the temperature T , m ≪ M ∼ πT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (37) In this case, the high-temperature expansion does not apply to the field of mass M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But neither can the tem- perature be neglected when integrating it out, as in equa- tion (36).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' However, the high-temperature expansion still applies for the light field of mass m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There should still exist a 3D EFT for the zero-mode of the light field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This method of “partial dimensional reduction” is not widely studied, but see [28–31] for a few studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Here I want to highlight another example: a variant of the Coleman-Weinberg (CW) model [13] as studied in [32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This model features radiative symmetry breaking: there is no symmetry breaking at tree-level, but there is at 1-loop level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This is the model of Abelian Higgs—a complex scalar charged under a U(1) gauge field—with a small positive mass term: Vcl(φ) = 1 2m2φ2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4, (38) This potential is of comparable size to the 1-loop contri- bution of the gauge boson when λ ∼ g4 and m2 ∼ g4σ2 with σ a characteristic size of the VEV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There is a clear hierarchy of scales: the gauge boson is heavier than the scalars and can be integrated out.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This gives a modified potential VLO(φ) = 1 2m2φ2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4 + 3JCW(g2φ2), (39) were JCW is the same as J0 of equation (11) but with −3/2 replaced by −5/6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This potential has a non-zero minimum, and it is from this potential which we should find the scalar masses, m2 H(φ) = ∂2VLO(φ), (40) m2 G(φ) = ∂VLO(φ) φ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (41) This is a consistent mass resummation derived from power-counting rules, with a hierarchy of scales that pro- tects it from double-counting diagrams and other issues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This model has two different minima with a barrier in between—could there be a first order phase transition between them?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To approach this question, we can assume that the high-temperature expansion applies, and that a φ3 barrier is induced: VLO(φ) = 1 2m2 effφ2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4 − 3 12πT g3φ3 + 3JCW(g2φ2), (42) but by balancing the powers of this expression we find gφ ∼ T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (43) Which implies that the high-temperature expansion does not apply.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But really we are asking for too much: we do not need the high-temperature expansion to apply to the gauge field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This model already has two minima, we do not need the temperature effects to create a barrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' We only need it to shift the energy of the different minima such that a phase transition can occur [9, 33];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' the resulting potential reads VLO(φ) = 1 2m2 effφ2 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λφ4 + 3JCW(g2φ2) + 3 T 4 2π2 JB(g2φ2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (44) Balancing the powers here gives the same counting as in equation (43).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But the question remains what to do with the scalar field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' After all, it is this field which will potentially un- dergo a transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Because the high-temperature expansion still applies to the scalar field, we should treat it using a 3D EFT as before.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To reach this EFT we integrate out the heavy modes of the scalars: the non-zero Matsubara modes and 6 the high-momentum modes of the Matsubara zero-mode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' At the same time, we also integrate out all modes of the vector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' What we end up with is a Euclidean 3D EFT with potential VLO(φ3) = 1 2m2 3φ2 3 + 1 4!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='λ3φ4 3 + 3JCW T (g2φ2) + 3 T 3 2π2 JB(g2φ2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (45) The 3D parameters are determined by matching with the 4D theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In this case we find m2 3 = m2 + 1 18λT 2 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (46) λ3 = λT + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (47) φ2 3 = φ2 T + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (48) The difference between these expressions and those of regular dimensional reduction of Abelian Higgs is that here the non-zero Matsubara modes of the vector bo- son do not contribute directly to the Wilson coefficients (compare equations (46) and (33) and note the missing g2T 2 term).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Instead the vector modes contribute through the non-polynomial term in the effective potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This contribution will in the end propagate to the mass of the scalar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To find the correct resummation to use in this theory we use the same derivatives as in equations (40) and (41), but with the potential given by equation (45).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I empha- size that this resummation contains parts that are not utilizing the high-temperature expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' And yet this should be a wholly consistent resummation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Furthermore, because the high temperature expansion applies to the scalar field which undergoes the phase transition, the ma- chinery of thermal escape (tunnelling at finite tempera- ture) [5, 34, 35] should apply and all the usual formulas ought to carry over.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='4 Though a detailed analysis of the power-counting scheme and its convergence is warranted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' DISSOLVE ILLUSORY PROBLEMS Studies of the electroweak phase transition have a long laundry list of problems: gauge dependence, strong renor- malization scale dependence, the Goldstone boson catas- trophe, IR divergences, imaginary potentials, mirages, perturbative breakdown, resummation method depen- dence, and linear terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Many of these problems were 4 A funny corollary of this is that a CW-like SU(2) gauge the- ory would automatically suppress sphaleron transitions after the phase transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The power-counting in equation (43) implies that the gauge field is not excited at the phase transition temper- ature since it is too heavy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The suppression of thermal sphalerons in models with radiative symmetry breaking hence make them natural candidates for electroweak baryogenesis (see [36] for a numerical study in agreement with this claim).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' recently studied in [15], where it was shown that some of them can yield big quantitative and qualitative uncer- tainties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I argue that these problems are dissolved if one uses a consistent and strict perturbative expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Gauge dependence The gauge dependence of the effective action, and in par- ticular the effective potential, is well-known and captured in the famous Nielsen identities [37, 38].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Essentially the effective potential is only gauge-invariant when evaluated at an extremum—at a physical point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But in perturba- tion theory there are implementation details: to get a gauge-independent result we must use a strict perturba- tive expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' So if the effective potential is expanded as V = V0 + xV1 + x2V2 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (49) then we must find the extrema perturbatively, φ = φ0 + xφ1 + x2φ2 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (50) by inserting this expansion of φ into the expansion of V , and extremizing the potential order-by-order: ∂V |φ !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= 0 =⇒ ∂V0|φ0 !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= 0 (51) φ1 !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= − ∂V1 ∂2V0 ���� φ0 (52) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This expansion is sometimes called a tadpole expansion, since it effectively reinserts scalar tadpoles into the 1PI diagrams of the effective potential [38].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Though it is well-established that the strict expan- sion above gives gauge-independent results, there has re- mained some confusion if it can also give accurate re- sults.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The strict expansion was popularized in [39] under the name ℏ-expansion (now sometimes called the PRM method).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The authors then expressed concern that the ℏ-expansion required a strict loop counting, while any resummation necessarily mixes loop orders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The way out of this dilemma is to realize that though a strict expansion is necessary, it does not have to be a loop expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' All that is required is that the pertur- bative expansion is performed using a consistent power- counting [40].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In [27] it was shown that such a strict expansion works if the expansion parameter x is small.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There is no conflict between gauge-independence and ac- curate results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' On the contrary they go hand-in-hand, as predicted by Arnold & Espinosa:5 5 Unfortunately this is not reflected in the wider literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There is much confusion as is evident by sampling the papers citing [39].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 7 In the best of worlds it would be nice to com- pute results in a variety of gauges and check that physical quantities are indeed gauge in- variant, but we have not had the perseverance to do so.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 26] Strong renormalization scale dependence There are many studies showing a strong renormalization scale dependence in perturbative calculations of phase transition quantities [15, 41, 42].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The problem is that re- summations mix loop orders, which messes up the ordi- nary cancellation between implicit running of parameters and explicit running of loop-functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' As the thermal masses arise at 1-loop, their running must be cancelled by the next loop order: two loops.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' As such, the solution is to use dimensional reduction and calculate up to two-loop order.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Constructing the 3D EFT consistently resums large contributions, and the running within this EFT is tame.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The Goldstone boson catastrophe To understand the possibility of IR-divergences we can consider the form of the zero-temperature 1-loop function J0, J0 ∼ m4 log m2 =⇒ J′′ 0 ∼ log m2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (53) The second derivative of this function diverges in the m2 → 0 limit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This divergence indicates two related prob- lems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The first problem can be seen if we think of these derivatives as insertions of interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Then the diver- gence of J′′ 0 implies that the 3-loop potential will diverge in the same limit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This is known as the Goldstone bo- son catastrophe, since the Goldstones are massless in the broken minimum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The literature implies a resummation is necessary to cure this divergence [43–45].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But this is not correct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In fact, a resummation is not necessary to remove the divergence in a regular loop expansion [46].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' It is enough to simply perform a strict expansion: any IR divergences of the potential are cancelled by corre- sponding divergences in the background field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' However, if a modified power-counting is used—i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' when there is a hierarchy of scales—resummation becomes necessary again.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The second related problem arises if one attempts to use a mass-dependent renormalization scheme and tries to match the measured masses of the scalars through sec- ond derivatives of the one-loop potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The Goldstones then again cause divergences, which are typically regu- lated away with an IR regulator inserted by hand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In my opinion it is much simpler to use MS and not have to deal with these issues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But there is no real incon- sistency with using the mass-dependent scheme together with an IR cutoff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' IR divergences Attempting to use the ℏ-expansion (strict loop expan- sion) to find the critical temperature in a theory with a radiative barrier leads to IR divergences [39, 47].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The di- vergence comes from expanding the critical temperature around T0, Tc = T0 + ℏT1 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (54) ∆V |Tc !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= 0 =⇒ m2 3 �� T0 !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (55) In the strict loop expansion, T0 coincides with the tem- perature where the classical potential of the 3D EFT has a second order phase transition, as stated in equa- tion (55).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' When evaluating the two-loop potential at this temperature, there are divergences ∼ log m2 3 �� T0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' What is going wrong?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The problem is that the wrong power-counting is used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' By using the modified power-counting λ ∼ g3 which a radiative barrier necessitates [1], one instead finds the critical temperature Tc = TLO + xTNLO .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' , (56) ∆V |Tc !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= 0 =⇒ ∆VLO|TLO !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='= 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' (57) Expanding around TLO does not feature IR diver- gences [27, 40].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Power-counting together with a strict perturbative expansion is IR safe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Imaginary potentials A potential with a non-zero imaginary part signals an instability: the 1-loop potential can develop an imaginary part if evaluated at a field-value where a square mass is negative.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This can happen close to the broken minimum where the Goldstone square mass changes sign.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But it can also happen at the origin for temperatures below T0, where the scalar mass terms are negative.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' These circumstances can arise if one is mixing loop orders in V (φ)—if one is not using a strict perturba- tive expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' When searching field space and scanning in temperatures this necessitates taking the real part of the potential in order to get a sensible answer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But this merely cures a symptom and does not fix the real prob- lem: unstable modes are influencing the calculation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But if one uses the resummed leading order effective potential VLO in a strict perturbative expansion, then the critical temperature and broken minimum can be found order by order.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The leading order quantities TLO, φLO are found from VLO and subsequent orders are evalu- ated there: VNLO|TLO,φLO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' No imaginary parts develop;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' all modes are correctly accounted for.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 8 Mirages Arnold & Espinosa demonstrated that the Z2 symmetric pure scalar theory appears to have a barrier but in fact does not, as we saw in an earlier section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In that case the mirage was induced from a breakdown of perturbation theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' More generally, mirages can arise when perturba- tive orders are out of control.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Either because the expan- sion parameter is too large, as above, or when orders are mixed haphazardly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Then the only way to be safe against them is to use a consistent power-counting, and perform a strict expan- sion in a small parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Perturbative breakdown In the extreme case of a perturbative breakdown, mirages are indicative of a larger problem: we cannot trust the perturbative expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Certain perturbative problems can be fixed by reorder- ing perturbation theory—by resummations—while oth- ers are incurable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Famously, non-Abelian gauge theories suffer from the Linde problem: at high temperatures the gauge boson must develop a “magnetic mass” M ∼ g2T to cure IR divergences at four loops and higher [48].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This results in a complete perturbative breakdown;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' perturba- tive methods cannot reach O(M 3 ∼ g6T 4) (four loops).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To be fair, power-counting and a strict expansion can- not solve this problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But it can offer some perspec- tive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In SU(2) gauge theory with a radiative barrier, the Linde problem affects the sixth order and higher in strict perturbation theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The first five orders are calculable and even the first three orders offer good accuracy for a wide range of expansion parameter values [27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hence the Linde problem should not occupy too much space in our minds when studying phase transitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Resummation method dependence Comparing resummation methods and finding a differ- ence indicates that at least one of them is wrong.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But it cannot tell you which result is the correct one, or if any of them are.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Instead it is better to look to the assumptions made on which the resummations are based.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' To be confident in which resummation method to use we should instead derive it from a consistent power- counting, and make sure that perturbation theory is con- verging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In the end there should only be one resummation method available: that which is implied by the hierarchy of scales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Linear terms Terms linear in φ will prevent the existence of the sym- metric minimum at high temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Such terms con- tradict our usual understanding that the symmetry is restored in this limit and in the early universe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In the beginning of the 1990s there were several resummation methods that produced such linear terms [16, 49], and there was some doubt whether the linear terms should exist or not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But convincing argument against such terms can be es- tablished from an effective field theory perspective, using the methods of power-counting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In [50], the authors ar- gue that the existence of an IR cutoff—the magnetic mass ∼ g2T discussed above—means that the effective poten- tial must be analytic in |φ|2 as φ → 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This prevents the existence of linear terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Any consistent resummation method must respect this constraint.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In a more modern setting, it is sometimes argued that Parwani resummation can give rise to such linear terms [41].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But I think this is unfair to Parwani resumma- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The linear terms arise if one uses a high-temperature approximation for the thermal counter-term, but main- tains the full unexpanded integrals in the potential [51].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' If the high-temperature expansion does not apply, then this is inconsistent and there will remain uncancelled terms that the counter-term insertion procedure cannot handle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I think it is more fair to blame a faulty power-counting: linear terms can arise if one is not using a consistent expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' GO FORTH AND COUNT POWERS The lesson of Arnold & Espinosa’s paper is a refrain that is common in physics, and a lesson we have learnt again and again: the proper way of doing physics is by way of the effective field theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' More formally, a strict perturbative expansion derived from an effective field theory is an asymptotic expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' And asymptotic expansions are unique [52]: there should not be any question as to which resummation method to use once you are settled on an approximation scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='6 Any resummation method that will stand the test of time must be implementing a power-counting protected by a hierarchy of scales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' In the previous sections I showcased the strength of power-counting by demonstrating how it dissolves the laundry list of confusion surrounding phase transition calculations (mirages, imaginary potentials, scheme de- pendence, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But if it is so powerful, why is not every- one using it?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There is one apparent drawback with applying the method of power-counting: some amount of thought has to be applied before the calculation starts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This means that scanning large sections of parameter space with a single method is out of the question.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There is no silver 6 The uniqueness of the coefficient also implies the perturbative expansion is gauge invariant order-by-order [27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 9 bullet which can work for all of parameter space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Instead one should divide parameter space into different regions in which different perturbative expansions apply.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But how small should λ be before we start counting it as g3 instead of g2?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This is a question with no clear answer within perturbation theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The best we can do is to look to lattice data to find approximate regions where a first order phase transition occurs or not, and cross- check our power-counting against it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But I argue that this is a good problem to have, in contrast to the dangerous confusions of the laundry list which comes from using a faulty perturbative expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The question of how large different contributions are is an honest quantitative physical question.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' You have to be realistic about these things.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' On the other hand, much have been said about the apparent small size of uncertainty due to gauge depen- dence [53].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' For a given gauge fixing method, the results do not differ much between Landau gauge (ξ = 0) and reasonable values of ξ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' It is argued that even though a gauge dependent result is uncomfortable, it is not a large quantitative issue and can be ignored [41, 54].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But I think this is a distraction from the true issue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The real problem with gauge dependence was never that one finds a span of values for different ξ and have to pick one of them at the expense of accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The real problem is that gauge dependence signals that something is wrong with the perturbative expansion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' And when something is wrong with perturbation theory, we are at the mercy of the other problems in the laundry list.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='7 Gauge depen- dence signals that we have lost control of perturbation theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' By not taking the signal seriously we are bound to be confused.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' The reasons above are why I encourage anyone in- terested in performing a perturbative study of a phase transition to always begin with the power-counting, and to monitor the convergence of perturbation theory—by comparing the different orders to each other and check- ing that the renormalization scale dependence is under control.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Whether the high temperature expansion applies or not, you should be able to establish a hierarchy of scales and derive the correct mass resummation scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' And if no hierarchy exists you should not perform a resumma- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' There are many phenomenological models in which first order phase transitions seem possible, but the exis- tence of which have not yet been established using power- counting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='8 To do so would enable us to study phase tran- sitions in such models consistently and accurately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' This is what Arnold & Espinosa tried to teach us 30 years ago, but we failed to learn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' But it is not too late.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Go forth and count powers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' ACKNOWLEDGMENTS I thank Andreas Ekstedt and Tuomas Tenkanen for their comments on the manuscript, and Oliver Gould for en- lightening discussions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [1] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Arnold and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Espinosa, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D47, 3546 (1993), [Erratum: Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='D50,6662(1994)], arXiv:hep-ph/9212235 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [2] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Braaten and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Nieto, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D51, 6990 (1995), arXiv:hep-ph/9501375 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [3] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Kajantie, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rummukainen, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Shaposhnikov, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 458, 90 (1996), arXiv:hep-ph/9508379.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [4] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hirvonen, (2022), arXiv:2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='02687 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [5] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Gould and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hirvonen, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 104, 096015 (2021), arXiv:2108.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='04377 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [6] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hirvonen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' L¨ofgren, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ramsey-Musolf, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schi- cho, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, JHEP 07, 135 (2022), arXiv:2112.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='08912 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [7] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' L¨ofgren, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ramsey-Musolf, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, (2021), arXiv:2112.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='05472 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [8] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 106, 095026 (2022), arXiv:2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='05145 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 7 Though to be fair, varying ξ can lead not only to quantitative uncertainty, but also to qualitative differences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Sometimes bar- riers disappear, sometimes new minima are generated, as ξ is varied [15, 27, 55].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 8 An exercise for the reader: develop a power-counting for a radia- tive barrier to occur in the inert doublet model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [9] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ginsparg, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B170, 388 (1980).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [10] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Appelquist and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Pisarski, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D23, 2305 (1981).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [11] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Parwani, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D45, 4695 (1992), [Erratum: Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='D48,5965(1993)], arXiv:hep-ph/9204216 [hep- ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [12] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Dolan and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Jackiw, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 9, 3320 (1974).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [13] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Coleman and E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Weinberg, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 7, 1888 (1973).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [14] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Kainulainen, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Keus, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Niemi, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rummukainen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Vaskonen, JHEP 06, 075 (2019), arXiv:1904.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='01329 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [15] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Athron, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Balazs, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Fowlie, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Morris, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' White, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Zhang, (2022), arXiv:2208.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='01319 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [16] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Brahm and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hsu, (1991).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [17] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Boyd, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Brahm, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Hsu, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 48, 4963 (1993), arXiv:hep-ph/9304254.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [18] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Curtin, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Meade, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ramani, Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' C 78, 787 (2018), arXiv:1612.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='00466 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [19] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Curtin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Roy, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' White, (2022), arXiv:2211.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='08218 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [20] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Gould, JHEP 04, 057 (2021), arXiv:2101.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='05528 [hep- ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [21] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Niemi, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 103, 115035 (2021), arXiv:2103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='07467 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [22] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' ¨Osterman, JHEP 06, 130 (2021), arXiv:2102.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='11145 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 10 [23] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, (2022), arXiv:2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='08815 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [24] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Weinberg, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 47, 4614 (1993), arXiv:hep-ph/9211314.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [25] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Karjalainen and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Peisa, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' C76, 319 (1997), arXiv:hep-lat/9607023 [hep-lat].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [26] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Kajantie, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Karjalainen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Peisa, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 520, 345 (1998), arXiv:hep-lat/9711048.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [27] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Gould, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' L¨ofgren, (2022), arXiv:2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='07241 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [28] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Losada, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 582, 277 (2000), arXiv:hep-ph/0003111.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [29] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rummukainen, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 597, 23 (2001), arXiv:hep-lat/0009025.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [30] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Brauner, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tranberg, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Vuorinen, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Weir, JHEP 03, 007 (2017), arXiv:1609.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='06230 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [31] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schr¨oder, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 101, 023532 (2020), arXiv:1911.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='09123 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [32] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Metaxas and E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Weinberg, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 53, 836 (1996), arXiv:hep-ph/9507381.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [33] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Linde, Rept.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Prog.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 42, 389 (1979).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [34] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt, Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' C 82, 173 (2022), arXiv:2104.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='11804 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [35] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt, JHEP 08, 115 (2022), arXiv:2201.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='07331 [hep- ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [36] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Prokopec, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rezacek, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' ´Swie˙zewska, JCAP 02, 009 (2019), arXiv:1809.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='11129 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [37] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Nielsen, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 101, 173 (1975).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [38] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Fukuda and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Kugo, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 13, 3469 (1976).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [39] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Patel and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ramsey-Musolf, JHEP 07, 029 (2011), arXiv:1101.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='4665 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [40] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' L¨ofgren, JHEP 12, 136 (2020), arXiv:2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='12614 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [41] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Croon, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Gould, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' White, JHEP 04, 055 (2021), arXiv:2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='10080 [hep- ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [42] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Gould and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, JHEP 06, 069 (2021), arXiv:2104.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='04399 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [43] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Martin, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 90, 016013 (2014), arXiv:1406.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='2355 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [44] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Elias-Miro, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Espinosa, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Konstandin, JHEP 08, 034 (2014), arXiv:1406.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='2652 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [45] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Espinosa, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Garny, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Konstandin, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 94, 055026 (2016), arXiv:1607.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='08432 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [46] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Ekstedt and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' L¨ofgren, JHEP 01, 226 (2019), arXiv:1810.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='01416 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [47] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 335, 173 (1994), arXiv:hep-ph/9406268.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [48] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Linde, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 96B, 289 (1980).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [49] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Shaposhnikov, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B 277, 324 (1992), [Er- ratum: Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='B 282, 483 (1992)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [50] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Dine, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Leigh, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Huet, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Linde, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Linde, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' D 46, 550 (1992), arXiv:hep-ph/9203203.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [51] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Laine, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Meyer, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Nardini, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' B920, 565 (2017), arXiv:1702.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='07479 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [52] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Bender, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Orszag, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Orszag, Ad- vanced mathematical methods for scientists and engineers I: Asymptotic methods and perturbation theory, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' 1 (Springer Science & Business Media, 1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [53] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Garny and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Konstandin, JHEP 07, 189 (2012), arXiv:1205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='3392 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [54] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Schicho, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Tenkanen, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' White, JHEP 11, 047 (2022), arXiv:2203.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='04284 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' [55] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Zuk, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Balazs, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' Papaefstathiou, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content=' White, (2022), arXiv:2212.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} +page_content='04046 [hep-ph].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/etE4T4oBgHgl3EQfqA2v/content/2301.05197v1.pdf'} diff --git a/fNA0T4oBgHgl3EQfHf-p/content/tmp_files/2301.02062v1.pdf.txt b/fNA0T4oBgHgl3EQfHf-p/content/tmp_files/2301.02062v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..91fd19d086d94c5c48a75ddbab8f7d317040cad1 --- /dev/null +++ b/fNA0T4oBgHgl3EQfHf-p/content/tmp_files/2301.02062v1.pdf.txt @@ -0,0 +1,2075 @@ +1 + +Stoic Conceptual Modeling Applied to +Business Process Modeling Notation (BPMN) + +Sabah Al-Fedaghi* +Computer Engineering Department +Kuwait University +Kuwait +salfedaghi@yahoo.com, sabah.alfedaghi@ku.edu.kw + +Abstract – Basic abstraction principles are reached through +ontology, which was traditionally conceived as a depiction of the +world itself. Ontology is also described using conceptual modeling +(CM) that defines fundamental concepts of reality. CM is one of +the central activities in computer science, especially as it is mainly +used in software engineering as an intermediate artifact for +system construction. To achieve such a goal, we propose Stoic CM +(SCM) as a description of what a system must do functionally +with minimal ambiguity. As a case study, we apply SCM to +investigate the ontology of BPMN (business process modeling +notation). Such an undertaking would demonstrate SCM notions +and simultaneously may offer a viable ontological foundation for +BPMN. SCM defines the being of things and actions in reality +based on Stoic notions of existence and subsistence. It has two +levels of specification: (1) a subsistence static model where things +and actions subsist and (2) an existence dynamic model where +things and actions exist in time. From the Stoic ontological point +of view, while a thing existing has a clear denotation, subsistence +indicates the thing is “being there,” but it is inactive (does not +participate in an event). We apply SCM to BPMN processes that +involve buying a new car with many notions, such as activity, +task, event, and message. The result indicates that SCM produces +a tighter representation of reality, thus providing the necessary +description of the part in the application world to be used as +requirements for developing the software system. + +Index Terms – conceptual modeling, Stoic, process, BPMN, +subsistence, existence, organization + +I. +INTRODUCTION +Anthropologists think that abstraction of reality is the +most important feature that gave homo sapiens a competitive +edge over less developed human races. Basic abstraction +principles are reached through ontology, which traditionally +conceived is a description of the world itself. Ever since the +Greek philosopher Aristotle, ontology has served as a basis for +human theories and the construction of models [1]. Ontology is +also used in conceptual modeling (CM) to define fundamental +concepts of reality. +CM is one of the central activities in computer science, +especially as it is mainly used in software engineering as an +intermediate artifact for system construction [2]. In this paper, +we propose Stoic CM (SCM) as a description of what a system +must do functionally with minimal ambiguity [3]. + +--------------------------------------------- +*Retired June 2021, seconded fall semester 2021/2022 + + +We claim that Stoic ontology can provide a foundation for CM +and as a way to evaluate the ontological soundness of a CM +language and its corresponding concepts. Ontology refers to +things whose existence is acknowledged by a system [4][5]. As +a case study, we investigate applying SCM to business process +modeling +notation +(BPMN). +Such +a +venture +would +demonstrate the SCM notions and may simultaneously provide +a possible ontological foundation for BPMN. + +A. Business Process Modeling Notation +The complexity of contemporary business information +systems has motivated interest in studies that focus on CM as +an aid to facilitate the comprehension of certain domain facts +relating to such systems, which would contribute to better +design decisions and eventually a better system [6]. +Understanding the CM of business domains is a key skill for +practitioners tasked with system analysis and design [6]. +BPMN has been utilized to create 1) descriptive business +processes models that can be communicated and analyzed and +2) technical view targets for technical developers who need +detailed specifications on the models to make them executable +[7]. BPMN enhances communication between business +analysts, technical developers, and business people [8]. +Theorists have proposed that BPMN can be used as a base +for CM for its simplicity and expressiveness. Moreover, it is +claimed that BPMN presents a well-defined semantic structure +and provides an easy working platform [9]. According to [10], +BPMN is characterized by its ability ―to describe and reflect +the real world of information systems better. This specification +is comprehensive and partially conflicting.‖ Therefore, +―several researchers present an ontology that provides a formal +definition of BPMN and can be used as a knowledge base‖ +[10]. +The focus of the BPMN standard is on providing an +intuitive graphical language, rather than formal semantics +specifications. This emphasis results in semantic ambiguities +regarding the interpretation of its modeling constructs [11]. +There is no guideline for using BPMN in ontology-based +systems, and different communities have been working on +BPMN-like ontologies for disparate application purposes [11]. +The following example motivates further study of BPMN +ontology. + +2 + + +B. Sample Problem +Amongst main BPMN modeling notions is the notion of +an event. In BPMN, events affect the flow of the model: +―throw events cause something to happen; catch events are +caused to happen. Moreover, depending on their position in +the process flow, they are: start events, end events, or +intermediate events‖ [11]. A BPMN event that references a +named message is known as a message event. A message +represents the content of communication between two +participants [12]. +Reference [11] scrutinized the BPMN diagram shown in +Fig. 1. Messages are exchanged between the process pools by +using tasks of the type, send message. BPMN includes +message as a throw event, which is used to model the sending +of a message as well. Since both message task (activity) and +message throw (event) model the same sending of a message, +there are differences in meaning. In general, an event conveys +the idea that when something happens..., whereas an activity +places more focus on the idea that something needs to be done. +Events represent changes in the domain being modeled, +whereas activities refer to a participant‘s commitment towards +the fulfillment of a specific goal. An event maps to a time +point on a timeline, whereas a task (or an activity) maps to a +time interval. This relationship between BPMN constructs and +the temporal line suggests that events are instantaneous while +activities last in time. ―BPMN does not commit to a theory of +time points or intervals; every reference to time beyond +atomicity remains vague within its modeling framework‖ [11]. + +C. Paper Structure +This paper has two aims: 1) exploring the ontological +features of SCM as a new modeling language and 2) applying +SCM in a new application area, the BPMN. Accordingly, the +paper is structured as follows. The next section gives a +description of SCM that includes materials that have appeared +in previous publications to provide a self-contained +manuscript. +Additionally, +the +paper +includes +a +new +contribution to SCM, which is an ongoing process. Section 3 +gives samples of SCM modeling. Section 4 includes a BPMN +case study modeled using SCM notions. + +II. SCM AND THINGING MACHINE MODEL +SCM utilizes thinging machine (TM) modeling [13] based +on Stoic ontology [14] and Lupascian logic [15]. In TM +modeling, a thing is a Heideggerian notion that indicates +something that becomes itself and announces its existence or +names the entering of the thing into the world (using Stoic +ontology terms, see e.g., [16]). Such things are conceptualized +as thimacs (things/machines). A thimac can be described +similar to a Romer‘s system: ―any part of reality, which can, at +least in principle, be separated from the rest of the world and +can be made an object of investigation‖ [17]. +Additionally, a thimac (Fig. 2) has a dual mode of being: +the machine side and the thing side. The machine has the five +potential actions: create, process, release, transfer, and receive. +The sense of ‗‖machinery‖ originated in the TM five actions +indicating that everything that creates, processes, and moves + + + + + + + + + + + + + + + + + + + + + + + + + + + (release-transfer-receive) +other +things +is +a +machine. +Simultaneously, what a machine creates, processes (changes), +releases, transfers, and/or receives is a thing. In this modeling +view, the ―world‖ is a totality of thimacs. +TM modeling has two levels of specification: +1) Static Model: represents static things and static +actions. From the SCM point of view, a thing‘s ―being‖ at this +level is a certain state of being, subsistence or a potential for +―becoming,‖ i.e., ―it is there,‖ inert, passive, waiting to exist +when it couples with time. Becoming refers to transferring to +the dynamic level to trigger creating an event. It is also the +―inactive‖ state (e.g., dormant volcano). The static level is the +retreating ―world‖ of events, e.g., doing something becomes a +negative event of ‗not doing‘ (a Lupascian logic term). +2) Dynamic Model: includes a static model subdiagram +(region) that unfolds with time, leading to events, i.e., the +realization of static things and actions. Thus, the event is the +existing being that was formerly a subsisting being as a region +in the static level. In this context, the Lupascian notion of a +negative event as reverting to the static level from the dynamic +level may explain Russell‘s [18] shadowy things, e.g., Today is +Wednesday when in fact it is Tuesday. Today is Wednesday is +a static (subsisting) thing that can exist (event) when It is +Tuesday. +Before presenting the TM modeling in details, we will +introduce its foundation in terms of Stoic ontology. + +A. Stoic Addition to TM Two-Level Modeling +Stoic ontology serves to define the being of TM things +and actions in reality. The Stoics concocted the idea of a +broader category of being: reality is made of things that exist +and things that subsist. This idea retains the commonsensical +notion that both static things and dynamic things are in some +Fig. 2. The thimac as a thing and machine +Thimac +Machine +Thing +Manifest + +Receive + + + + + + + +Process +Accept +Transfer +Release +Arrive + +Output + +Input +Create +Change + +Fig. 1 BPMN process diagram (From [12]) + +Poqunst for +Duotos +Send +Duote3 + +sense real. The notion of ―modes of being‖ appears in different +form in classical logic where the notions of existence and +subsistence +appear +[19]. +Reference +[20] +introduced +Meinongian metaphysics and has distinguished between being +and existence. Using Stoic ontology, we view the dynamic TM +description as in existence, whereas the static, mapped portion +of the dynamic description is in subsistence. +TM staticity refers to a static model that represents the +world of potentialities with atemporal and nonspatial +subsistence. The static model is self-contained (no new outside +thimacs added) and in a state in which time and its related +notions lose meaning. This static universe ―contains +everything there is or ever was or will be‖ (from [21] after +adding ―static‖ at the beginning). +In the dynamic TM level, events form among themselves +an interacting nexus (assemblages) that define, inform, and +constitute all thimac beings (existence). Things at the dynamic +level may present both object-like and PROCESS-like aspects +[22]. We use PROCESS with capital letters to distinguish this +term from process action used in a TM (Fig. 2). PROCESS in +a TM is another term for event and, more specifically, a net of +events, as will be discussed in detail when we model BPMN. +The event can be provisionally defined as a fundamental +happening that forms the basic building blocks of the existing +world. Everything in the world, including people and things, +can be constructed from events that form essential ontological +elements. Understanding their ―being‖ is the major field of +investigation conducted by Heidegger. + +B. Example: Regions and Events +Reference [22] considered a simple homogeneous, open- +ended PROCESS such as walking. According to [22], this +seems to have a ―universal‖ character as walking is present at +many +different +spatiotemporal +locations +with +many +participants. But if so, what is the spatiotemporal extent of that +walking? According to [22], in ordinary discourse, people +typically report the occurrence of such an event as ―I walked to +the station,‖ ―I walked for five minutes,‖ or ―I walked 500 +meters.‖ It is not clear that we need a notion of a ―process +token‖ distinct from the widely accepted notion of an event +token [22]. +Fig. 3 shows the static representation of Walking to the +station. This representation reflects a subsisting machine +(PROCESS) that has the potential to be actualized in reality. It +is subsisting because it has a being, but such a being is in the +inactive state. It is not a mental thing because even if human +beings vanished, it is possible that some non-human (e.g., a +dog) may walk to a station. It is an inactive ―natural‖ +PROCESS like a dormant volcano that had experienced +eruptions long before the existence of human beings. +Note that staticity here includes all variations of a certain +thimac to permit ontological changes at the dynamic level. For +example, ontological relationships among thimacs may change +to a different ontological arrangement, e.g., A is no more part +of B; in this case, the static level includes all these disparities. +A subsisting thimac moves to existence if it is embedded +with a time subthimac to become an event. We can establish +events to the generic actions (e.g., transfer, receive, etc.) in +Fig. 3, but we prefer to declare more meaningful events as +shown in Fig. 4. The region of an event is the subdiagram of +the static model where the event occurs. For simplification, +hereafter, we will use regions to indicate events. +Fig. 5 shows the event Walked 500 meters and Fig. 6 +shows the event Walked for five minutes. +To illustrate how the two levels of a TM world relate to +each other, Fig. 7 shows the TM PROCESS of ―becoming‖ +from a static region to an event (Event 1 is taken from Fig. 4). +Accordingly, following Whitehead‘s metaphysics, things in +reality are PROCESSes, i.e., are constituted by PROCESSes +(in TM, events or net of events), and their becoming is also a +PROCESS. + + + +Time +Create +Receive +Release +Process +Receive +Transfer +Transfer +Transfer +Transfer +Receive +Transfer +Release +Receive +Release + +Static level +Dynamic level +Transfer +Region +Release + + +Event +Fig. 7 Converting a region to an event + +Fig. 3 Static model of Walking to the station + +Station + +Walking + +Transfer +Release +Process +Create + + +Transfer +Receive + + + + + + + + + + + + +Station + +Walking + +Transfer +Fig. 4 Dynamic model +Release +Process +Event 1 + +Event 2 + +Create +Transfer +Release +Process +Receive +Transfer +Time + +Region + +Region + +Transfer +Release +Process +Receive +Transfer +Transfer +Release +Process +Receive +Transfer +Transfer +Receive +Region + +Transfer +Release +Process +Receive +Transfer +Event 3 + +Event 4 + +Region + + + + +Walking + +Transfer +Release +Process +Create +Region + +Fig. 5 The event of Walked 500 meters +Past +Distance is 500meters +Fig. 6 The event of Walked for five minutes + + + + +Walking + +Transfer +Release +Process +Create +Region + +Five minutes in past + +Eve nt 1 +Wa Iki +Create +RegianWaTking +Create4 + +In the context of SCM, such issues and other related ones (e.g., +nature of time, identity, and events calculus) need further +investigation. Note that the development of TM and SCM is an +ongoing venture further refined in each paper. + +C. The Thing Side of the Thimac +The thimac is a whole that is more than the sum of its +parts (i.e., has its own machine). Even if interiority has no +subthimacs (e.g., empty storage safe), the thimac has its +actions: create, process, release, transfer, and/or receive. Thing +subsistence means it is an identifiable thing along with its +related actions. For example it is like a city in a map. The city +can be described in terms of streets, population, connections +with other cities, interaction with the environment, windiness, +water resources, etc., but it is just a map with no activities. +Even though it is connected with another city, there are no +moving cars on the highways and no playing children in the +streets. ―Relations‖ between subsisting things are like dry river +beds. Even though a dry river (e.g., release, transfer, transfer, +receive) looks ―permanent‖ in the static model, it becomes a +flash event that may perish any time, i.e., alternate between +static and dynamic levels. +Only thimacs that can embed time are realizable (exist) at +the dynamic level. Thus, for example, ―square circle‖ is a +static thimac that cannot be injected with time to exist in the +dynamic model; neither does it subsist because it is not +mappable to the dynamic level. The TM universe is populated +by things that alternate between two different levels of being: +static and dynamic. This universe is a PROCESS (net of +events) where events occur and then perish or cease to be. +According to [23], the Stoic incorporeals (mapable +occupants of the static level) are conditions ―without which the +interaction of bodies in the world would neither be analyzable +nor intelligible.‖ The being of those incorporeals (entity-like +and PROCESS-like) does not depend on their material +occupants, for they can carry on in their own subsistent way +without being materialized (in TM existence). + +D. The Machine +TM actions seen in Fig. 2 can be described as follows. +1) Arrive: A thing moves to a machine. In Stoic ontology, +motion exists along the temporal dimension and ―the length of +a motion, or its duration, can be measured by how much of this +temporal dimension the motion covers. One motion has a +longer duration than another just in case it covers more of the +temporal dimension‖ [24]. + In TM modeling, not only time is the dimension in which +motion exists, but also, time is the dimension in which actions +(create, process, release, transfer, and receive) exist. +2) Accept: A thing enters the machine. For simplification, +we assume that arriving things are accepted; therefore, we can +combine arrive and accept stages into the receive stage. +3) Release: A thing is ready for transfer outside the machine. +4) Process: A thing is changed, handled, and examined, but +no new thing results. This also includes the action of +comparison, e.g., a number (thing) is greater than zero? Also +consider creating a wood fire then ―the wood is burning‖ is a +process indicating the physical situation has remained the +same, burning. +5) Transfer: A thing is input into or output from a machine. +The dynamic (not necessarily physical) ―movement‖ (event) is +from a previous region to a different region through yet a third +region. +6) Create: A new thing ―becoming‖ (found/manifested) is +realized from the moment it arises (emergence) in a thimac. +Simultaneously it also refers to the ―existence‖ of a thing, +especially where we want to emphasize persistence in time. +Ending the creation of an event returns the region of the +event to the static level. There are three types of creating: +existing, subsisting, and appearing (not subsisting in the static +level, e.g., square circle). Note that ―not subsisting‖ implies no +possibility of existence (see Fig. 8). These things ―neither +subsist nor exist but they are still objects [things] in their own +right—that is, they are part of the catalogue of the world in +their own right‖ [25]. Creation as existence is assertive (e.g., +true), creation as subsisting is expressive, creation as +appearance is inexpressive in the dynamic level. +Additionally, the TM model includes a triggering +mechanism (denoted by a dashed arrow in this article‘s +figures), which initiates a (non-sequential) flow from one +machine to another. Moreover, each action stage may have its +own memory storage (denoted by cylinder in the TM diagram) +of things. A memory has its own five actions forming a +memory thimac. +Note that for simplicity, we may omit create in some +diagrams because the box representing the thimac implies its +existence (in the TM model). +E. Two Levels of Representation and Lupascian Logic +The TM event is different from similarly named notions +currently used in the literature. Note that the TM approach +takes the side of philosophers (e.g., Whitehead) who conceive +of physical things as extended across time. Objects and events +are things of the same kind [26]. In this context, the TM differs +from such an ontological approach to objects by introducing +the notion of the thimac as a thing and machine extended in +time. +Note that objects are elements on the dynamic level and +typically viewed as durable solid things. Their static +counterparts are withdrawn to the static level and physically +imperceptible. Additionally, with respect to intentionality, the +same object may appear partially, hence differently, at the +dynamic level. However, we ignore the notion of intentionality +at this stage of research. + + +Physical +Things + + + + +What exists +What subsists +What does not subsist +TM +Model +Fig. 8 Things in SCM + +5 + +In SCM, instead of PROCESS vs. ―stop PROCESS,‖ the event +moves the region to the dynamic level in contrast to ―revert +(the event‘s region) to static level.‖ This method of eliminating +negativity stems from philosopher Stéphane Lupasco. +According to [27], every element e (in TM: an event, i.e., a +thimac that contains a region plus time) always associates with +a non-e (in TM: static thimac), such that the actualization of +one entails the potentialization of the other and vice versa, +alternatively, ―without either ever disappearing completely‖ +[27]. More illustration of this topic can be found in [15]. The +the next section and last section in this paper include examples +of negative events represented according to Lupascian logic. +Stoic ontology allows continuous passage from corporeals to +incorporeals and back again as a cycle of subsistence and +existence. +III. EXPLORATION IN SCM +The Stoics‘ ―body‖ (the subject matter of physics) is +anything that is capable of acting or being acted upon. It is the +mark of existence. A thimac is a machine that creates, +processes, releases, transfers, and/or receives; simultaneously, +it is a thing that can be created, processed, released, +transferred, and/or received. Thus, where as in Stoicism, two +principles together constitute physical reality: that which acts +and that which is acted on, in TM, acts are creates, processes, +releases, transfers, or receive; and is acted on are being +created, processed, released, transferred, and received. In TM, +thimacs can be actors and actees simultaneously. +A. Example 1 +The famous Heidegger‘s hammer is an existing tool as +long as it is ready-to-hand. However, this existence (ready-to- +hand) has two modes of being, one as a physical thing (create) +and as an actually used thing (PROCESS). When it breaks +down (present-at-hand), it subsists (See Fig. 9, left). What +makes a hammer a ―known‖ thing is its whole, e.g., function +and utility. Note that the head and hand of the hammer exist. +As Fig. 9 (left) shows, the hammer conceals itself (withdrawn) +into subsistence, thus, unavailable, but its pieces exist. In Fig. +9 (middle), it exists, but it may not be used. In Fig. 9 (right), +the hammer exists and is being used. Thus, it seems that +present-to-hand and ready-at-hand require a third mode of +being, used-at-hand. This illustrates one feature of SCM +representation. + +B. Example 2 +This example is more complex that previous examples in +this paper. It points to a need for a calculus of events in [28], +business process modeling based on TM. +According to [23], in My watch was made in Switzerland, +the expression ―my watch‖ refers not to a piece of metal but to +an immaterial office, a role for something to be, filled by some +piece of metal. If it becomes true that Tomorrow I lose my +watch and buy a new one, then a distinct piece of metal from +the one currently on my wrist will tomorrow become my +watch. Therefore, my watch hardly refers to any piece of +material. +Fig. 10 shows the TM static model for this scenario. First, +a person (pink 1) buys a watch (2) to become his/her watch +(3). ―Losing‖ is the negative event ―not owning‖ that has the +same region as the event owning but without time, as in the +dynamic model. Fig. 11 shows the dynamic model that +includes the following events. +E1: I am an existing object (i.e., entity-like event). +E2: A watch is an existing object. +E3: I bought the watch. +E4: The watch comes into my possession. +E5: I own the watch. +R5: I do not own the watch (lose it); R denotes region 5. + +Fig. 11 Dynamic model of I lose my watch and buy a new one + + + + + + +I + +Create +E1 +* +E5 +E4 +E3 +Ownership +watch + +Process: buy +E2 +* +Receive +Transfer +Release +Transfer +Transfer +Receive +Process: owning +Transfer +Watch + +Fig. 10 Static model of I lose my watch and buy a new one + + +Person + +Create +Person’s watch + +Process: buy +1 +Receive +Transfer +Release +Transfer +Transfer +Receive +Process: owning +Transfer +Watch + +2 +3 +Fig. 9 Subsisting hammer with existing head and hand (left), existing +hammer (middle), existing used hammer (right) + + + + + +Hammer +Create + + +Head +Hand +Create +Create + + +Hammer +Create + + +Head +Hand +Create +Create + Hammer Create +Transfer +Process +Release +Release +Transfer + +6 + + +In Fig. 11, E1 and E2 indicate my and the watch‘s +existence, and I bought the watch (E3). E4 denotes that the +watch becomes my watch; E5 registers the state of being of the +owner of this watch. R5 (the region of E5) indicates that I lost +my watch; hence, the watch is no longer in my ownership. +There is another watch in E2 that I bought in E3 (different time) +that becomes my watch. The interesting point in this example +is how the regions of E2, E3, and E4 are repeated in different +times. Hence, regions are elevated to events in different times. +Asterisks indicate extended events (persisted entity-like event). +It is possible to limit this extension (begin–end), but in this +example, we ignore such detail. +Fig. 12 shows the behavior model. + +IV. BPMN USING SCM + +A business PROCESS describes how an organization +performs the work necessary to produce outputs [29]. +Reference [30] defines a business PROCESS as ―a series of +steps designed to produce a product or service.‖ The BPMN +specification defines business PROCESS as ―a sequence of +activities leading from an initial state of the PROCESS +instance to some defined end state‖ [31]. An activity ―is an +action that is performed repeatedly in the course of business. +Each instance of the activity represents the same action (more +or less) on a different piece of work‖ [31]. Business +PROCESSes may have been carefully designed, or they may +have simply evolved over the years. PROCESSes are central to +understanding how anything in the organization creates value +[29]. Reference [29] used the famous input–process–output +model to define a business PROCESS as activities that +transform inputs into outputs valued by an organization. +In this paper, we develop the view that a business +PROCESS is a thimac realized as a net of TM PROCESSes. +The organization as a thimac can be decomposed into major +PROCESSes that in turn can be divided into subPROCESSes. + +A. New Car Sale Modeling +Reference [32] presented ―fact-based analysis‖ of BPMN +example models, as shown in Fig. 13 [31]. The following +concepts apply. +1) Activity: work performed within a business. +2) Process: any activity performed within or across +companies. +3) Event: something that ―happens‖ during the course of a +business process. +Then, [32] uses such a list and Fig. 13 to develop the +following fact verbalizations as ground facts (only a small +subset of the verbalizations is shown here): + +The Black-Box Pool: ―Customer‖ has an outgoing MessageFlow +―Order‖ to the Message Start Event ―Receive Order‖ within the Lane +―Sales‖ of The White-Box Pool: ―New Car Sales.‖ The Black-Box +Pool ―Customer‖ has an outgoing MessageFlow ―confirmation +response‖ to the SubProcess ―Order Car from Factory.‖ The +SubProcess ―order Car from Factory‖ has An outgoing MessageFlow +―confirmation request‖ to the Black-Box Pool ―Customer.‖ + + + + + + + + + + + + + + + + + + + +Such an approach is claimed to enhance the ―semantic +richness‖ of BPMN. Reference [32]‘s example provides a +typical example of a business PROCESS, meaning a network +of TM events. We claim that such an attempt to provide more +BPMN semantics is caused by the modeling in BPMN. To +demonstrate that, we develop a model based on SCM in order +to contrast the two models side by side. + +B. Static Model +Fig. 14 shows the TM static model (originally in the +subsisting state) of regions of the new car sale PROCESS that +can be described as follows. + +A customer creates an order for a new car (circle 1) that +flows to the new car sale (2) where it is processed (3). +According to the result of such a process, +a) If the car is not available (4), then this triggers creating a +message (5) that flows to the customer (6). +b) If the car is to be ordered from the factory (7), then a +request to confirm is sent to the customer (8), and with getting +this confirmation (9, 10 and 11), a request is sent to the +manufacturer (12) and a request is sent to the preparation +section (13) to prepare the car. +c) If the car is available (14), a request is sent to the +preparation section (15). + +In both cases of (b) and (c) above, a request for finance is +sent to the finance section (16 and 17). In the finance section, +the request for financing is processed (18) to trigger generation +of a loan request (19) that flows to a loaner (20). +The loaner processes the loan request (21) and sends the +response to the finance section (22). The finance section +processes +the +loaner +response +(23) +and +sends +an +approval/rejection decision to the preparation section (24). In +the preparation section, +a) If a preparation request (25) and a finance approval (26) +are received, the car is taken out of storage (27), prepared +(28), and sent to the customer (29). The horizontal thick +bar is a graphical simplification that indicates satisfying +both conditions in 25 and 26. +Fig. 12 Behaviour model + I +Watch +Bought +Receive +Lost Another +watch +Buy Receive +E1 +* E2 +* +E3 +E4 +R5 E2 +* +E3 +E4 +Tomorrow +------------------- +E5 +Own + +Fig. 13 BPMN example (From [32], originally in [31]) + +Confimatior +Confirmation + Deliser vehide +response +request +and tenporary +registration +Order +Order from +unavailable ? +Car +s3[es +factory ? +Enter Order +Order car from +Factory +ye: +yes +国 +Receive Order +New Car Sales +no +Sucpess +Prepare car for +Financing +daid +delivery +unavailable +国 +yes + ajqegentun +Financing +Finance +Arrange +Financing +Close and Delive +国 +- +no +国 +4 +Ioan.reque.d +4 +Factoryorder +Cancel factory1 +order +1Ship date +Loan response +文 +Loan decs +[approved] +Loan docs +executed +Factory +Lender7 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +In the case of a car ordered from the manufacturer, we assume +a third condition that the car has arrived to the storage from the +manufacturer (30). +b) In (a) above, if the finance is rejected (31), then the +actions in (a) are not executed (negative event as will be +indicated in the dynamic model). +C. Dynamic Modeling +Fig. 15 shows the dynamic model with the following +events. +E1: The customer sends an order for a new car that is received +by the new car sale. +E2: Processing the customer‘s order. +E3: The ordered car is unavailable; hence, a message is sent to +the customer informing him/her. +E4: Ordering a car involves a car from the factory. +E5: Making order to factory. +E6: Requesting and receiving confirmation of ordering from +factory. +E7: Sending a request for preparation of a car. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +E8: Sending a request for financing the car. +E9: The car is available. +E10: Creating a request for a loan that flows to a loaner. +E11: The loaner responds to the loan request. +E12: The finance section processes the loaner response and +sends approval/rejection to the preparation section. +E13: The preparation section receives an approval of financing. +E14: The preparation section receives a preparation request. +E15: The car has arrived from the factory. +E16: The car is delivered to the customer. +E17: Finance is rejected. +R16: (negative event) The car delivery is cancelled. +Fig. 16 shows the chronology of these events. +Thus, the new car sale PROCESS starts at the subsistence +(static) level where no events occur. Note the sense of +subsistence in this initial step. All regions have beings (they +are real), but they are inactive. For example, if a customer +asks, do you sell used cars? The answer is there is no such +PROCESS in our company. + + + + +Process: (a) +Confirm + +Customer + +Process: if confirm +Create +Transfer +Car +unavailable + +Create + + +(c) Order from factory? +Not available +Create +Receive +Preparation + +Create +Sale + +Finance + +Transfer +(b) Available +Create +Order to +Factory +Transfer +Receive +Process +Receive +Transfer +Receive +Process +Receive +Transfer +From factory + +Create +Transfer +Release +Transfer +Process +Create +Transfer +Process +Loan +request + +Create +Receive +Transfer +Receive +Receive +Order + +Release +Transfer +Transfer +Receive +Transfer +Transfer +Release +Create +Transfer +Release +To Factory + +Transfer +Release +Receive +Transfer +Process +Transfer +Create +Release +Request to +confirm +from +factory + +Receive +Release +Transfer +Loaner + +Receive +Loan +approval/reject +Transfer +Release +Finance +approval/ +rejected +Transfer +Release +Transfer +Release +Preparation +request +Transfer +Cars + +1 +New Car Sale + +Transfer +Release +2 +3 +4 +5 +6 +7 +8 +9 +10 +12 +11 +13 +14 +15 +17 +16 +18 +19 +20 +21 +22 +23 +24 +25 26 +Finance +approved +Finance +rejected +27 +28 +29 +9 +Transfer Release +30 +9 +31 +9 +Fig. 14 The static model of the new car sale + +8 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +Note how the events arise from the static level in +sequences for a single order, e.g., ordering available car (see +Fig. 17). Of course, there are several simultaneous orders. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +All types of studies can involve the number of events that are +currently occurring per region, the number of regions (negative +events), etc. that provide data for efficiency and activities +across all PROCESSes. + + + + + + + + + + + + + + + + + + + + + + +E2 +E4 +E3 +E5 +E6 +E7 +E1 + + + +Process: (a) +Confirm + +Customer + +Process: if confirm +Create +Transfer +Car +unavailable + +Create + +(c) Order from factory? +Not available +Create +Receive +Preparation + +Create +Sale + +Finance + +Transfer +(b) Available +Create +Order to +Factory +Transfer +Receive +Process +Receive +Transfer +Receive +Process +Receive +Transfer +From factory + +Create +Transfer +Release + +Transfer +Process +Create +Transfer +Process +Loan +request + +Create: +Receive +Transfer +Receive +Receive +Order + +Release +Transfer +Transfer +Receive +Transfer +Transfer +Release +Create +Transfer +Release +To Factory + +Transfer +Release +Receive +Transfer +Process +Transfer +Create +Release +Request to +confirm +from +factory + +Receive +Release +Transfer +Loaner + +Receive +Loan +approval/ +reject +Transfer +Release +Finance +approval/ +rejected +Transfer +Release +Transfer +Release +Preparation +request +Transfer +Cars + +New Car Sale + +Transfer +Release +Finance +approved +Finance +rejected +Transfer Release +Fig. 15 The dynamic model of the new car sale +E8 +E13 +E15 +E14 +E12 +E11 +E10 +E16 +E9 +E17 +E1 +E2 +E3 +E4 +E7 +E6 +E5 +E8 +E9 +E10 +E11 +E12 +E13 +E16 +E15 +E14 +E10 +E11 +E12 +E13 +E14 +E7 +E8 +Customer +order +Process +order +Not available +From factory +Order from factory +Customer +confirms +Request +preparation +Request +finance +Request +loan +Loaner +response +Finance +approval/ +reject +Car received +from factory +Finance +approved +Preparation request +received +Request preparation +Request finance +Available +Request loan +Loaner +response +Finance approval/reject +Finance approved +Preparation request +received +Fig. 16 Dynamic model +Finance rejected +E13 +R16 +Car delivery +Delivery cancelled +Finance rejected E13 +R16 +Delivery cancelled +E16 +Car delivery + +9 + + + + + + + + + + + + + + + + + + + + + + + + + + + + +V. CONCLUSION + +As a case study, we have applied SCM to investigate the +ontology of BPMN. Such an undertaking would demonstrate +the SCM notions and simultaneously may offer a viable +ontological foundation for BPMN. Specifically pinpointing the +disadvantages and disadvantages between the two modeling +techniques is difficult because they differ in every aspect. The +strategy is to compare them by modeling the same problem in +the two approaches in parallel and examine the overall results. +The judgement is left for the reader in such matters, for +example, in [10], the ability ―to describe and reflect the real +world of information systems better.‖ Further research is +needed to clarify the feasibility of SCM-based modeling. +However, one issue may be raised about the diagrammatic +―complexity‖ of the TM model. Such a complexity seems to be +caused by the repeated use of the TM actions. For example, by +assuming that the direction of arrows is sufficient to indicate +the flow, Fig. 14 can be simplified as shown in Fig. 18 that +eliminates the actions release, transfer, and receive. + +REFERENCES + +[1] D. Peuquet, B. Smith, and B. Brougaard, “The ontology of fields: +Report of a specialist meeting held under the auspices of the Varenius +Project,” Pan. Comp. Implem. Geogr. Concp. Spec. Mtg. Nat. Ctr. +Geogr. Inf. Analy. (NCGIA), Bar Harbor, Maine, June 11-13, 1998. pp. +1-42. +[2] B. Thalheim, ―The conceptual model ≡ An adequate and faithful artifact +enhanced by concepts,‖ Front. Artif. Intell., vol. 260, pp. 241-254, +January 2014. +[3] F. Bozyiğit, Ö. Aktaş, and D. Kılınç, ―Linking software requirements +and conceptual models: A systematic literature review,‖ Eng. Sci. +Technol. Int. J., vol. 24, no. 1, pp. 71-82, February 2021. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +[4] M. Verdonck, F. Gailly, S. d. Cesare, and G. Poels, ―Ontology-driven +conceptual modeling: A systematic literature mapping and review,‖ +Applied Ontology, vol. 10, nos. 3-4, pp. 197-227, December 2015. +[5] T. Honderich, The Oxford Companion to Philosophy, Oxford: Oxford +University Press, 2006. +[6] J. Mendling, J. Recker, H. A. Reijer, and H. Leopold, ―An empirical +review of the connection between model viewer characteristics and the +comprehension of conceptual process models,‖ Inf. Syst. Front., vol. 21, +no. 5, pp. 1111-1135, 2019. +[7] B. S. S. Onggo and O. Karpat, ―Agent-based conceptual model +representation using BPMN,‖ in Proc. Wtr. Simul. Conf., S. Jain, R. R. +Creasey, J. Himmelspach, K. P. White, and M. Fu, Eds., 2011, pp. 671- +682. +[8] B. Oliveira, Ó. Oliveira, and O. Belo, ―Using BPMN for ETL +conceptual modeling: A case study,‖ in Proc. 10th Int. Conf. Data Sci., +Paris, France, July 6-8, 2021, pp. 267-274. +[9] B. Oliveira, V. Santos, C. Gomes, R. Marques. and O. Belo, +―Conceptual–physical bridging—From BPMN models to physical +implementations on kettle,‖ CEUR Workshop, vol. 1418, pp. 55-59, +2015. +[10] M. Kchaou, W. Khlif, F. Gargouri, and M. Mahfoudh, ―Transformation +of BPMN model into an OWL2 ontology,‖ in Proc. 16th Int. Conf. +Eval. Novel Appro. Softw. Eng. (ENASE), Online, April 26-27, 2021, +pp. 380-388. +[11] E. M. Sanfilippo, S. Borgo, and C. Masolo, ―Events and activities: Is +there an ontology behind BPMN?,‖ in Proc. 8th Int. Conf. Formal +Ontol. Inf. Syst., Rio de Janeiro, Brazil, Sep 22-25, 2014. +[12] Object Management Group (OMG) Business Process Model and +Notation (BPMN). Version 2.0, 2011. Accessed: March 2014. +Available: http://www.bpmn.org +[13] S. Al-Fedaghi and G. Aldamkhi, ―Conceptual modeling of an IP phone +communication system: A case study,‖ Int. J. Interdisc. Telecom. Netw., +vol. 13, no. 3, pp. 83-94. +[14] S. Al-Fedaghi, ―Conceptual modeling founded on the Stoic ontology: +Reality with dynamic existence and static subsistence,‖ arXiv preprint, +arXiv:2212.05300. +[15] S. Al-Fedaghi, ―Modeling system events and negative events using +thinging machines based on Lupascian logic,‖ arXiv preprint +arXiv:2211.12586. + +Fig. 17 The trace of events for available car + + + + +Create +Transfer +Loan +request + +Release +Transfer +Loaner + +Receive +E10 + + + + + + + + + + +E2 +E7 +E1 +Process: (a) +Confirm + +Customer + +Process: if confirm +Create +Transfer +Car +unavailable + +Create + +(c) Order from factory? +Not available +Create +Receive +Preparation + +Create +Sale + +Finance + +Transfer +(b) Available +Order to +Factory +Transfer +Receive +Process +Receive +Transfer +Receive +Process +Receive +Transfer +From factory + +Create +Transfer +Release +Process +Create +Transfer +Process +Create: +Receive +Transfer +Receive +Receive +Order + +Release +Transfer +Transfer +Receive +Transfer +Transfer +Release +Create +Transfer +Release +To Factory + +Transfer +Release +Receive +Transfer +Process +Transfer +Create +Release +Request to +confirm +from +factory + +Receive +Loan +approval/reject +Transfer +Release +Finance +approval/r +ejected +Transfer +Release +Transfer +Release +Preparation +request +Transfer +Cars + +New Car Sale + +Transfer +Release +Finance +approved +Finance +rejected +Transfer +Release +E8 +E13 +E14 +E12 +E11 +E16 +E9 + +10 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +[16] A. K. Hai, ―Heidegger: The fourfold,‖ Blog, Accessed: Dec. 10, 2022. +https://that-which.com/heidegger-the-fourfold/ +[17] H. Romer, ―Complementarity of process and substance,‖ arXiv:quant- +ph/0602171v1. +[18] B. Russell, ―The philosophy of logical atomism,‖ in Logic and +Knowledge, London: George Allen & Unwin, 1968. +[19] R. F. A. Hoernlé, ―‗Existence‘ and ‗subsistence‘ in contemporary logic +and epistemology,‖ in Proc. Sixth Int. Cong. Phil., 1927, pp. 261-271. +https://doi.org/10.5840/wcp6192750 +[20] A. Meinong, ―On the Theory of Objects,‖ in Realism and the +Background of Phenomenology, R. Chisholm, Ed., Glencoe, Ill.: Free +Press, 1960, pp. 76-117. +[21] J. Post, Metaphysics: A Contemporary Introduction, New York: +Paragon House, 1991. +[22] A. Galton, 2012. ―The ontology of states, processes, and events,‖ in +Interdisciplinary Ontology, M. Okada and B. Smith, Eds., vol. 5. [Proc. +fifth Interdisc. Ontol. Mtg., Tokyo: Keio University, pp. 35-45.] +[23] T. J. Bailey, ―The Structure of Stoic metaphysics,‖ in Oxford Studies in +Ancient Philosophy, vol. XLVI, B. Inwood, Ed., Summer, 2014. +https://philpapers.org/archive/BAITSO-21.pdf +[24] B.A. Greene, ―The imperfect present: Stoic physics of time,‖ PhD +dissertation, +Univ. +California, +San +Diego, +2018. +https://escholarship.org/content/qt3jz5n2pk/qt3jz5n2pk_noSplash_0fb2 +f7fcc5c31c232cbaab793ef4f8ad.pdf + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +[25] A. Betti, Against Facts, Cambridge, MA: MIT Press, 2015. +[26] R. Casati and V. Achille, ―Events,‖ in The Stanford Encyclopedia of +Philosophy, E. N. Zalta, Ed., Summer 2020. +https://plato.stanford.edu/archives/sum2020/entries/events/ +[27] J. E. Brenner, ―Consciousness as process: A new logical perspective,‖ +Philosophy and Computers, vol. 18, no. 1, Fall 2018, pp. 10-24. +https://cdn.ymaws.com/www.apaonline.org/resource/collection +/EADE8D52-8D02-4136-9A2A-729368501E43/ComputersV18n1.pdf +[28] I. Chishti, ―Towards a general framework for business process +modelling,‖ Int. J. Intell. Comp. Res., vol. 5, no. 3, September 2014. +[29] P. Harmon, ―What is a business process,‖ Bus. Process Trends, vol. 8, +no. 21, December 14, 2010. +[30] G. A. Rummler and A. P. Brache, Improving Performance, 2nd ed., San +Francisco: Jossey-Bass, 1995. +[31] B. Silver, BPMN Method and Style, 2nd ed., Altadena, CA: Cody- +Cassidy Press, 2017. +[32] P. Bollen, ―Fact-based meta model for BPMN,‖ in ERP Future, F. +Piazolo et al., Eds. LNBIP 285, pp. 29–41, 2017. DOI: 10.1007/978-3- +319-58801-8_3 + + + + +Process: (a) +Confirm + +Customer + +Process: if confirm +Create +Car +unavailable + +Create + + +(c) Order from factory? +Not available +Create +Preparation + +Create +Sale + +Finance + +(b) Available +Create +Order to +Factory +Process +Process +From factory + +Create +Process +Create +Process +Loan +request + +Create: +OK +Receive +Order + +Create +To Factory + +Process +Create +Request to +confirm +from +factory + +Loaner + +Loan approval/rejection +Finance +approval/ +rejected +Preparation +request +Cars + +New Car Sale + +Finance +approved +Finance +rejected +Fig. 18 Simplification of the static model of the new car sale + diff --git a/fNA0T4oBgHgl3EQfHf-p/content/tmp_files/load_file.txt b/fNA0T4oBgHgl3EQfHf-p/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..bc7d374c240c102ea1965f5d245f46dcf874d0b9 --- /dev/null +++ b/fNA0T4oBgHgl3EQfHf-p/content/tmp_files/load_file.txt @@ -0,0 +1,1021 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf,len=1020 +page_content='1 Stoic Conceptual Modeling Applied to Business Process Modeling Notation (BPMN) Sabah Al Fedaghi Computer Engineering Department Kuwait University Kuwait salfedaghi@yahoo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='com, sabah.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='alfedaghi@ku.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='kw Abstract – Basic abstraction principles are reached through ontology, which was traditionally conceived as a depiction of the world itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ontology is also described using conceptual modeling (CM) that defines fundamental concepts of reality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' CM is one of the central activities in computer science, especially as it is mainly used in software engineering as an intermediate artifact for system construction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' To achieve such a goal, we propose Stoic CM (SCM) as a description of what a system must do functionally with minimal ambiguity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' As a case study, we apply SCM to investigate the ontology of BPMN (business process modeling notation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Such an undertaking would demonstrate SCM notions and simultaneously may offer a viable ontological foundation for BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' SCM defines the being of things and actions in reality based on Stoic notions of existence and subsistence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It has two levels of specification: (1) a subsistence static model where things and actions subsist and (2) an existence dynamic model where things and actions exist in time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' From the Stoic ontological point of view, while a thing existing has a clear denotation, subsistence indicates the thing is “being there,” but it is inactive (does not participate in an event).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' We apply SCM to BPMN processes that involve buying a new car with many notions, such as activity, task, event, and message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The result indicates that SCM produces a tighter representation of reality, thus providing the necessary description of the part in the application world to be used as requirements for developing the software system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Index Terms – conceptual modeling, Stoic, process, BPMN, subsistence, existence, organization I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' INTRODUCTION Anthropologists think that abstraction of reality is the most important feature that gave homo sapiens a competitive edge over less developed human races.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Basic abstraction principles are reached through ontology, which traditionally conceived is a description of the world itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ever since the Greek philosopher Aristotle, ontology has served as a basis for human theories and the construction of models [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ontology is also used in conceptual modeling (CM) to define fundamental concepts of reality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' CM is one of the central activities in computer science, especially as it is mainly used in software engineering as an intermediate artifact for system construction [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In this paper, we propose Stoic CM (SCM) as a description of what a system must do functionally with minimal ambiguity [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' --------------------------------------------- *Retired June 2021, seconded fall semester 2021/2022 We claim that Stoic ontology can provide a foundation for CM and as a way to evaluate the ontological soundness of a CM language and its corresponding concepts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ontology refers to things whose existence is acknowledged by a system [4][5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' As a case study, we investigate applying SCM to business process modeling notation (BPMN).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Such a venture would demonstrate the SCM notions and may simultaneously provide a possible ontological foundation for BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Business Process Modeling Notation The complexity of contemporary business information systems has motivated interest in studies that focus on CM as an aid to facilitate the comprehension of certain domain facts relating to such systems, which would contribute to better design decisions and eventually a better system [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Understanding the CM of business domains is a key skill for practitioners tasked with system analysis and design [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' BPMN has been utilized to create 1) descriptive business processes models that can be communicated and analyzed and 2) technical view targets for technical developers who need detailed specifications on the models to make them executable [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' BPMN enhances communication between business analysts, technical developers, and business people [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Theorists have proposed that BPMN can be used as a base for CM for its simplicity and expressiveness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Moreover, it is claimed that BPMN presents a well-defined semantic structure and provides an easy working platform [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to [10], BPMN is characterized by its ability ―to describe and reflect the real world of information systems better.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This specification is comprehensive and partially conflicting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ Therefore, ―several researchers present an ontology that provides a formal definition of BPMN and can be used as a knowledge base‖ [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The focus of the BPMN standard is on providing an intuitive graphical language, rather than formal semantics specifications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This emphasis results in semantic ambiguities regarding the interpretation of its modeling constructs [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' There is no guideline for using BPMN in ontology-based systems, and different communities have been working on BPMN-like ontologies for disparate application purposes [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The following example motivates further study of BPMN ontology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2 B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Sample Problem Amongst main BPMN modeling notions is the notion of an event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In BPMN, events affect the flow of the model: ―throw events cause something to happen;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' catch events are caused to happen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Moreover, depending on their position in the process flow, they are: start events, end events, or intermediate events‖ [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A BPMN event that references a named message is known as a message event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A message represents the content of communication between two participants [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Reference [11] scrutinized the BPMN diagram shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Messages are exchanged between the process pools by using tasks of the type, send message.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' BPMN includes message as a throw event, which is used to model the sending of a message as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Since both message task (activity) and message throw (event) model the same sending of a message, there are differences in meaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In general, an event conveys the idea that when something happens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', whereas an activity places more focus on the idea that something needs to be done.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Events represent changes in the domain being modeled, whereas activities refer to a participant‘s commitment towards the fulfillment of a specific goal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' An event maps to a time point on a timeline, whereas a task (or an activity) maps to a time interval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This relationship between BPMN constructs and the temporal line suggests that events are instantaneous while activities last in time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ―BPMN does not commit to a theory of time points or intervals;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' every reference to time beyond atomicity remains vague within its modeling framework‖ [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Paper Structure This paper has two aims: 1) exploring the ontological features of SCM as a new modeling language and 2) applying SCM in a new application area, the BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Accordingly, the paper is structured as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The next section gives a description of SCM that includes materials that have appeared in previous publications to provide a self-contained manuscript.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Additionally, the paper includes a new contribution to SCM, which is an ongoing process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Section 3 gives samples of SCM modeling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Section 4 includes a BPMN case study modeled using SCM notions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' SCM AND THINGING MACHINE MODEL SCM utilizes thinging machine (TM) modeling [13] based on Stoic ontology [14] and Lupascian logic [15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In TM modeling, a thing is a Heideggerian notion that indicates something that becomes itself and announces its existence or names the entering of the thing into the world (using Stoic ontology terms, see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', [16]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Such things are conceptualized as thimacs (things/machines).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A thimac can be described similar to a Romer‘s system: ―any part of reality, which can, at least in principle, be separated from the rest of the world and can be made an object of investigation‖ [17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Additionally, a thimac (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2) has a dual mode of being: the machine side and the thing side.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The machine has the five potential actions: create, process, release, transfer, and receive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The sense of ‗‖machinery‖ originated in the TM five actions indicating that everything that creates, processes, and moves (release-transfer-receive) other things is a machine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Simultaneously, what a machine creates, processes (changes), releases, transfers, and/or receives is a thing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In this modeling view, the ―world‖ is a totality of thimacs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' TM modeling has two levels of specification: 1) Static Model: represents static things and static actions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' From the SCM point of view, a thing‘s ―being‖ at this level is a certain state of being, subsistence or a potential for ―becoming,‖ i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', ―it is there,‖ inert, passive, waiting to exist when it couples with time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Becoming refers to transferring to the dynamic level to trigger creating an event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It is also the ―inactive‖ state (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', dormant volcano).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The static level is the retreating ―world‖ of events, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', doing something becomes a negative event of ‗not doing‘ (a Lupascian logic term).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2) Dynamic Model: includes a static model subdiagram (region) that unfolds with time, leading to events, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', the realization of static things and actions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thus, the event is the existing being that was formerly a subsisting being as a region in the static level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In this context, the Lupascian notion of a negative event as reverting to the static level from the dynamic level may explain Russell‘s [18] shadowy things, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Today is Wednesday when in fact it is Tuesday.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Today is Wednesday is a static (subsisting) thing that can exist (event) when It is Tuesday.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Before presenting the TM modeling in details, we will introduce its foundation in terms of Stoic ontology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Stoic Addition to TM Two-Level Modeling Stoic ontology serves to define the being of TM things and actions in reality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The Stoics concocted the idea of a broader category of being: reality is made of things that exist and things that subsist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This idea retains the commonsensical notion that both static things and dynamic things are in some Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The thimac as a thing and machine Thimac Machine Thing Manifest Receive Process Accept Transfer Release Arrive Output Input Create Change Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1 BPMN process diagram (From [12]) Poqunst for Duotos Send Duote3 sense real.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The notion of ―modes of being‖ appears in different form in classical logic where the notions of existence and subsistence appear [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Reference [20] introduced Meinongian metaphysics and has distinguished between being and existence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Using Stoic ontology, we view the dynamic TM description as in existence, whereas the static, mapped portion of the dynamic description is in subsistence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' TM staticity refers to a static model that represents the world of potentialities with atemporal and nonspatial subsistence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The static model is self-contained (no new outside thimacs added) and in a state in which time and its related notions lose meaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This static universe ―contains everything there is or ever was or will be‖ (from [21] after adding ―static‖ at the beginning).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In the dynamic TM level, events form among themselves an interacting nexus (assemblages) that define, inform, and constitute all thimac beings (existence).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Things at the dynamic level may present both object-like and PROCESS-like aspects [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' We use PROCESS with capital letters to distinguish this term from process action used in a TM (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' PROCESS in a TM is another term for event and, more specifically, a net of events, as will be discussed in detail when we model BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The event can be provisionally defined as a fundamental happening that forms the basic building blocks of the existing world.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Everything in the world, including people and things, can be constructed from events that form essential ontological elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Understanding their ―being‖ is the major field of investigation conducted by Heidegger.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Example: Regions and Events Reference [22] considered a simple homogeneous, open- ended PROCESS such as walking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to [22], this seems to have a ―universal‖ character as walking is present at many different spatiotemporal locations with many participants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' But if so, what is the spatiotemporal extent of that walking?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to [22], in ordinary discourse, people typically report the occurrence of such an event as ―I walked to the station,‖ ―I walked for five minutes,‖ or ―I walked 500 meters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ It is not clear that we need a notion of a ―process token‖ distinct from the widely accepted notion of an event token [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3 shows the static representation of Walking to the station.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This representation reflects a subsisting machine (PROCESS) that has the potential to be actualized in reality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It is subsisting because it has a being, but such a being is in the inactive state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It is not a mental thing because even if human beings vanished, it is possible that some non-human (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', a dog) may walk to a station.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It is an inactive ―natural‖ PROCESS like a dormant volcano that had experienced eruptions long before the existence of human beings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that staticity here includes all variations of a certain thimac to permit ontological changes at the dynamic level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' For example, ontological relationships among thimacs may change to a different ontological arrangement, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', A is no more part of B;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' in this case, the static level includes all these disparities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A subsisting thimac moves to existence if it is embedded with a time subthimac to become an event.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' We can establish events to the generic actions (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', transfer, receive, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=') in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3, but we prefer to declare more meaningful events as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The region of an event is the subdiagram of the static model where the event occurs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' For simplification, hereafter, we will use regions to indicate events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 5 shows the event Walked 500 meters and Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 6 shows the event Walked for five minutes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' To illustrate how the two levels of a TM world relate to each other, Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 7 shows the TM PROCESS of ―becoming‖ from a static region to an event (Event 1 is taken from Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Accordingly, following Whitehead‘s metaphysics, things in reality are PROCESSes, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', are constituted by PROCESSes (in TM, events or net of events), and their becoming is also a PROCESS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Time Create Receive Release Process Receive Transfer Transfer Transfer Transfer Receive Transfer Release Receive Release Static level Dynamic level Transfer Region Release Event Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 7 Converting a region to an event Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3 Static model of Walking to the station Station Walking Transfer Release Process Create Transfer Receive Station Walking Transfer Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 4 Dynamic model Release Process Event 1 Event 2 Create Transfer Release Process Receive Transfer Time Region Region Transfer Release Process Receive Transfer Transfer Release Process Receive Transfer Transfer Receive Region Transfer Release Process Receive Transfer Event 3 Event 4 Region Walking Transfer Release Process Create Region Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 5 The event of Walked 500 meters Past Distance is 500meters Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 6 The event of Walked for five minutes Walking Transfer Release Process Create Region Five minutes in past Eve nt 1 Wa Iki Create RegianWaTking Create4 In the context of SCM, such issues and other related ones (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', nature of time, identity, and events calculus) need further investigation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that the development of TM and SCM is an ongoing venture further refined in each paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The Thing Side of the Thimac The thimac is a whole that is more than the sum of its parts (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', has its own machine).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Even if interiority has no subthimacs (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', empty storage safe), the thimac has its actions: create, process, release, transfer, and/or receive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thing subsistence means it is an identifiable thing along with its related actions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' For example it is like a city in a map.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The city can be described in terms of streets, population, connections with other cities, interaction with the environment, windiness, water resources, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', but it is just a map with no activities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Even though it is connected with another city, there are no moving cars on the highways and no playing children in the streets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ―Relations‖ between subsisting things are like dry river beds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Even though a dry river (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', release, transfer, transfer, receive) looks ―permanent‖ in the static model, it becomes a flash event that may perish any time, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', alternate between static and dynamic levels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Only thimacs that can embed time are realizable (exist) at the dynamic level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thus, for example, ―square circle‖ is a static thimac that cannot be injected with time to exist in the dynamic model;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' neither does it subsist because it is not mappable to the dynamic level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The TM universe is populated by things that alternate between two different levels of being: static and dynamic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This universe is a PROCESS (net of events) where events occur and then perish or cease to be.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to [23], the Stoic incorporeals (mapable occupants of the static level) are conditions ―without which the interaction of bodies in the world would neither be analyzable nor intelligible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ The being of those incorporeals (entity-like and PROCESS-like) does not depend on their material occupants, for they can carry on in their own subsistent way without being materialized (in TM existence).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The Machine TM actions seen in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2 can be described as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1) Arrive: A thing moves to a machine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In Stoic ontology, motion exists along the temporal dimension and ―the length of a motion, or its duration, can be measured by how much of this temporal dimension the motion covers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' One motion has a longer duration than another just in case it covers more of the temporal dimension‖ [24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In TM modeling, not only time is the dimension in which motion exists, but also, time is the dimension in which actions (create, process, release, transfer, and receive) exist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2) Accept: A thing enters the machine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' For simplification, we assume that arriving things are accepted;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' therefore, we can combine arrive and accept stages into the receive stage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3) Release: A thing is ready for transfer outside the machine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 4) Process: A thing is changed, handled, and examined, but no new thing results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This also includes the action of comparison, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', a number (thing) is greater than zero?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Also consider creating a wood fire then ―the wood is burning‖ is a process indicating the physical situation has remained the same, burning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 5) Transfer: A thing is input into or output from a machine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The dynamic (not necessarily physical) ―movement‖ (event) is from a previous region to a different region through yet a third region.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 6) Create: A new thing ―becoming‖ (found/manifested) is realized from the moment it arises (emergence) in a thimac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Simultaneously it also refers to the ―existence‖ of a thing, especially where we want to emphasize persistence in time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ending the creation of an event returns the region of the event to the static level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' There are three types of creating: existing, subsisting, and appearing (not subsisting in the static level, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', square circle).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that ―not subsisting‖ implies no possibility of existence (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' These things ―neither subsist nor exist but they are still objects [things] in their own right—that is, they are part of the catalogue of the world in their own right‖ [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Creation as existence is assertive (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', true), creation as subsisting is expressive, creation as appearance is inexpressive in the dynamic level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Additionally, the TM model includes a triggering mechanism (denoted by a dashed arrow in this article‘s figures), which initiates a (non-sequential) flow from one machine to another.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Moreover, each action stage may have its own memory storage (denoted by cylinder in the TM diagram) of things.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A memory has its own five actions forming a memory thimac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that for simplicity, we may omit create in some diagrams because the box representing the thimac implies its existence (in the TM model).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Two Levels of Representation and Lupascian Logic The TM event is different from similarly named notions currently used in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that the TM approach takes the side of philosophers (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Whitehead) who conceive of physical things as extended across time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Objects and events are things of the same kind [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In this context, the TM differs from such an ontological approach to objects by introducing the notion of the thimac as a thing and machine extended in time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that objects are elements on the dynamic level and typically viewed as durable solid things.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Their static counterparts are withdrawn to the static level and physically imperceptible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Additionally, with respect to intentionality, the same object may appear partially, hence differently, at the dynamic level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' However, we ignore the notion of intentionality at this stage of research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Physical Things What exists What subsists What does not subsist TM Model Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 8 Things in SCM 5 In SCM, instead of PROCESS vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ―stop PROCESS,‖ the event moves the region to the dynamic level in contrast to ―revert (the event‘s region) to static level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ This method of eliminating negativity stems from philosopher Stéphane Lupasco.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to [27], every element e (in TM: an event, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', a thimac that contains a region plus time) always associates with a non-e (in TM: static thimac), such that the actualization of one entails the potentialization of the other and vice versa, alternatively, ―without either ever disappearing completely‖ [27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' More illustration of this topic can be found in [15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The the next section and last section in this paper include examples of negative events represented according to Lupascian logic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Stoic ontology allows continuous passage from corporeals to incorporeals and back again as a cycle of subsistence and existence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' EXPLORATION IN SCM The Stoics‘ ―body‖ (the subject matter of physics) is anything that is capable of acting or being acted upon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It is the mark of existence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A thimac is a machine that creates, processes, releases, transfers, and/or receives;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' simultaneously, it is a thing that can be created, processed, released, transferred, and/or received.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thus, where as in Stoicism, two principles together constitute physical reality: that which acts and that which is acted on, in TM, acts are creates, processes, releases, transfers, or receive;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' and is acted on are being created, processed, released, transferred, and received.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In TM, thimacs can be actors and actees simultaneously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Example 1 The famous Heidegger‘s hammer is an existing tool as long as it is ready-to-hand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' However, this existence (ready-to- hand) has two modes of being, one as a physical thing (create) and as an actually used thing (PROCESS).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' When it breaks down (present-at-hand), it subsists (See Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 9, left).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' What makes a hammer a ―known‖ thing is its whole, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', function and utility.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note that the head and hand of the hammer exist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' As Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 9 (left) shows, the hammer conceals itself (withdrawn) into subsistence, thus, unavailable, but its pieces exist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 9 (middle), it exists, but it may not be used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 9 (right), the hammer exists and is being used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thus, it seems that present-to-hand and ready-at-hand require a third mode of being, used-at-hand.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' This illustrates one feature of SCM representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Example 2 This example is more complex that previous examples in this paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It points to a need for a calculus of events in [28], business process modeling based on TM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to [23], in My watch was made in Switzerland, the expression ―my watch‖ refers not to a piece of metal but to an immaterial office, a role for something to be, filled by some piece of metal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' If it becomes true that Tomorrow I lose my watch and buy a new one, then a distinct piece of metal from the one currently on my wrist will tomorrow become my watch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Therefore, my watch hardly refers to any piece of material.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 10 shows the TM static model for this scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' First, a person (pink 1) buys a watch (2) to become his/her watch (3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ―Losing‖ is the negative event ―not owning‖ that has the same region as the event owning but without time, as in the dynamic model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 11 shows the dynamic model that includes the following events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E1: I am an existing object (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', entity-like event).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E2: A watch is an existing object.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E3: I bought the watch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E4: The watch comes into my possession.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E5: I own the watch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' R5: I do not own the watch (lose it);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' R denotes region 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 11 Dynamic model of I lose my watch and buy a new one I Create E1 E5 E4 E3 Ownership watch Process: buy E2 Receive Transfer Release Transfer Transfer Receive Process: owning Transfer Watch Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 10 Static model of I lose my watch and buy a new one Person Create Person’s watch Process: buy 1 Receive Transfer Release Transfer Transfer Receive Process: owning Transfer Watch 2 3 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 9 Subsisting hammer with existing head and hand (left), existing hammer (middle), existing used hammer (right) Hammer Create Head Hand Create Create Hammer Create Head Hand Create Create Hammer Create Transfer Process Release Release Transfer 6 In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 11, E1 and E2 indicate my and the watch‘s existence, and I bought the watch (E3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E4 denotes that the watch becomes my watch;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E5 registers the state of being of the owner of this watch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' R5 (the region of E5) indicates that I lost my watch;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' hence, the watch is no longer in my ownership.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' There is another watch in E2 that I bought in E3 (different time) that becomes my watch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The interesting point in this example is how the regions of E2, E3, and E4 are repeated in different times.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Hence, regions are elevated to events in different times.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Asterisks indicate extended events (persisted entity-like event).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' It is possible to limit this extension (begin–end), but in this example, we ignore such detail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 12 shows the behavior model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' BPMN USING SCM A business PROCESS describes how an organization performs the work necessary to produce outputs [29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Reference [30] defines a business PROCESS as ―a series of steps designed to produce a product or service.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ The BPMN specification defines business PROCESS as ―a sequence of activities leading from an initial state of the PROCESS instance to some defined end state‖ [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' An activity ―is an action that is performed repeatedly in the course of business.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Each instance of the activity represents the same action (more or less) on a different piece of work‖ [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Business PROCESSes may have been carefully designed, or they may have simply evolved over the years.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' PROCESSes are central to understanding how anything in the organization creates value [29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Reference [29] used the famous input–process–output model to define a business PROCESS as activities that transform inputs into outputs valued by an organization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In this paper, we develop the view that a business PROCESS is a thimac realized as a net of TM PROCESSes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The organization as a thimac can be decomposed into major PROCESSes that in turn can be divided into subPROCESSes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' New Car Sale Modeling Reference [32] presented ―fact-based analysis‖ of BPMN example models, as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 13 [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The following concepts apply.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1) Activity: work performed within a business.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 2) Process: any activity performed within or across companies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3) Event: something that ―happens‖ during the course of a business process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Then, [32] uses such a list and Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 13 to develop the following fact verbalizations as ground facts (only a small subset of the verbalizations is shown here): The Black-Box Pool: ―Customer‖ has an outgoing MessageFlow ―Order‖ to the Message Start Event ―Receive Order‖ within the Lane ―Sales‖ of The White-Box Pool: ―New Car Sales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ The Black-Box Pool ―Customer‖ has an outgoing MessageFlow ―confirmation response‖ to the SubProcess ―Order Car from Factory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ The SubProcess ―order Car from Factory‖ has An outgoing MessageFlow ―confirmation request‖ to the Black-Box Pool ―Customer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ Such an approach is claimed to enhance the ―semantic richness‖ of BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Reference [32]‘s example provides a typical example of a business PROCESS, meaning a network of TM events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' We claim that such an attempt to provide more BPMN semantics is caused by the modeling in BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' To demonstrate that, we develop a model based on SCM in order to contrast the two models side by side.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Static Model Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 14 shows the TM static model (originally in the subsisting state) of regions of the new car sale PROCESS that can be described as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A customer creates an order for a new car (circle 1) that flows to the new car sale (2) where it is processed (3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' According to the result of such a process, a) If the car is not available (4), then this triggers creating a message (5) that flows to the customer (6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' b) If the car is to be ordered from the factory (7), then a request to confirm is sent to the customer (8), and with getting this confirmation (9, 10 and 11), a request is sent to the manufacturer (12) and a request is sent to the preparation section (13) to prepare the car.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' c) If the car is available (14), a request is sent to the preparation section (15).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In both cases of (b) and (c) above, a request for finance is sent to the finance section (16 and 17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In the finance section, the request for financing is processed (18) to trigger generation of a loan request (19) that flows to a loaner (20).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The loaner processes the loan request (21) and sends the response to the finance section (22).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The finance section processes the loaner response (23) and sends an approval/rejection decision to the preparation section (24).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' In the preparation section, a) If a preparation request (25) and a finance approval (26) are received, the car is taken out of storage (27), prepared (28), and sent to the customer (29).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The horizontal thick bar is a graphical simplification that indicates satisfying both conditions in 25 and 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 12 Behaviour model I Watch Bought Receive Lost Another watch Buy Receive E1 * E2 * E3 E4 R5 E2 * E3 E4 Tomorrow ------------------- E5 Own Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 13 BPMN example (From [32], originally in [31]) Confimatior Confirmation Deliser vehide response request and tenporary registration Order Order from unavailable ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Car s3[es factory ?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Enter Order Order car from Factory ye: yes 国 Receive Order New Car Sales no Sucpess Prepare car for Financing daid delivery unavailable 国 yes ajqegentun Financing Finance Arrange Financing Close and Delive 国 no 国 4 Ioan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='reque.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='d 4 Factoryorder Cancel factory1 order 1Ship date Loan response 文 Loan decs [approved] Loan docs executed Factory Lender7 In the case of a car ordered from the manufacturer, we assume a third condition that the car has arrived to the storage from the manufacturer (30).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' b) In (a) above, if the finance is rejected (31), then the actions in (a) are not executed (negative event as will be indicated in the dynamic model).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Dynamic Modeling Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 15 shows the dynamic model with the following events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E1: The customer sends an order for a new car that is received by the new car sale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E2: Processing the customer‘s order.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E3: The ordered car is unavailable;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' hence, a message is sent to the customer informing him/her.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E4: Ordering a car involves a car from the factory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E5: Making order to factory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E6: Requesting and receiving confirmation of ordering from factory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E7: Sending a request for preparation of a car.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E8: Sending a request for financing the car.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E9: The car is available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E10: Creating a request for a loan that flows to a loaner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E11: The loaner responds to the loan request.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E12: The finance section processes the loaner response and sends approval/rejection to the preparation section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E13: The preparation section receives an approval of financing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E14: The preparation section receives a preparation request.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E15: The car has arrived from the factory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E16: The car is delivered to the customer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E17: Finance is rejected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' R16: (negative event) The car delivery is cancelled.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 16 shows the chronology of these events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thus, the new car sale PROCESS starts at the subsistence (static) level where no events occur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Note the sense of subsistence in this initial step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' All regions have beings (they are real), but they are inactive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' For example, if a customer asks, do you sell used cars?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The answer is there is no such PROCESS in our company.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Process: (a) Confirm Customer Process: if confirm Create Transfer Car unavailable Create (c) Order from factory?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Not available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='(b) Available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order to ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='From factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='To ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request to ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='confirm ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='from ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loaner ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/reject ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='rejected ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Cars ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='New Car Sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer Release 2 3 4 5 6 7 8 9 10 12 11 13 14 15 17 16 18 19 20 21 22 23 24 25 26 Finance approved Finance rejected 27 28 29 9 Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release 30 9 31 9 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 14 The static model of the new car sale 8 Note how the events arise from the static level in sequences for a single order, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', ordering available car (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Of course, there are several simultaneous orders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' All types of studies can involve the number of events that are currently occurring per region, the number of regions (negative events), etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' that provide data for efficiency and activities across all PROCESSes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E2 E4 E3 E5 E6 E7 E1 Process: (a) Confirm Customer Process: if confirm Create Transfer Car unavailable Create (c) Order from factory?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Not available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='(b) Available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order to ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='From factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create: ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='To ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request to ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='confirm ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='from ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loaner ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='reject ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='rejected ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Cars ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='New Car Sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer Release Finance approved Finance rejected Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='15 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='The ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='dynamic ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='model ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='of ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='the ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='new ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='car ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E15 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E14 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E12 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E11 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E16 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E9 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E7 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E6 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E5 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E9 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E11 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E12 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E16 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E15 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E14 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E11 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E12 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E14 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E7 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Customer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='order ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='order ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Not ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='From ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='from ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Customer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='confirms ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loaner ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='response ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='reject ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Car ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='received ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='from ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approved ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='received ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loaner ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='response ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/reject ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approved ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='received ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 16 Dynamic model Finance rejected E13 R16 Car delivery Delivery cancelled Finance rejected E13 R16 Delivery cancelled E16 Car delivery 9 V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' CONCLUSION As a case study, we have applied SCM to investigate the ontology of BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Such an undertaking would demonstrate the SCM notions and simultaneously may offer a viable ontological foundation for BPMN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Specifically pinpointing the disadvantages and disadvantages between the two modeling techniques is difficult because they differ in every aspect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The strategy is to compare them by modeling the same problem in the two approaches in parallel and examine the overall results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' The judgement is left for the reader in such matters, for example, in [10], the ability ―to describe and reflect the real world of information systems better.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='‖ Further research is needed to clarify the feasibility of SCM-based modeling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' However, one issue may be raised about the diagrammatic ―complexity‖ of the TM model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Such a complexity seems to be caused by the repeated use of the TM actions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' For example, by assuming that the direction of arrows is sufficient to indicate the flow, Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 14 can be simplified as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 18 that eliminates the actions release, transfer, and receive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' REFERENCES [1] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Peuquet, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Smith, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Brougaard, “The ontology of fields: Report of a specialist meeting held under the auspices of the Varenius Project,” Pan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Comp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Implem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Geogr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Concp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Spec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Mtg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ctr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Geogr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Inf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Analy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' (NCGIA), Bar Harbor, Maine, June 11-13, 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1-42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [2] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Thalheim, ―The conceptual model ≡ An adequate and faithful artifact enhanced by concepts,‖ Front.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Artif.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Intell.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 260, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 241-254, January 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [3] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Bozyiğit, Ö.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Aktaş, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Kılınç, ―Linking software requirements and conceptual models: A systematic literature review,‖ Eng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Technol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 24, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 71-82, February 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [4] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Verdonck, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Gailly, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Cesare, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Poels, ―Ontology-driven conceptual modeling: A systematic literature mapping and review,‖ Applied Ontology, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 10, nos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3-4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 197-227, December 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [5] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Honderich, The Oxford Companion to Philosophy, Oxford: Oxford University Press, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [6] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Mendling, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Recker, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Reijer, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Leopold, ―An empirical review of the connection between model viewer characteristics and the comprehension of conceptual process models,‖ Inf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Syst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Front.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 21, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1111-1135, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [7] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Onggo and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Karpat, ―Agent-based conceptual model representation using BPMN,‖ in Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Wtr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Simul.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Conf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Jain, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Creasey, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Himmelspach, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' White, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fu, Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', 2011, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 671- 682.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [8] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Oliveira, Ó.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Oliveira, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Belo, ―Using BPMN for ETL conceptual modeling: A case study,‖ in Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 10th Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Conf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Data Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Paris, France, July 6-8, 2021, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 267-274.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [9] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Oliveira, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Santos, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Gomes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Marques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Belo, ―Conceptual–physical bridging—From BPMN models to physical implementations on kettle,‖ CEUR Workshop, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1418, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 55-59, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [10] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Kchaou, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Khlif, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Gargouri, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Mahfoudh, ―Transformation of BPMN model into an OWL2 ontology,‖ in Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 16th Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Conf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Eval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Novel Appro.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Softw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Eng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' (ENASE), Online, April 26-27, 2021, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 380-388.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [11] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Sanfilippo, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Borgo, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Masolo, ―Events and activities: Is there an ontology behind BPMN?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=',‖ in Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 8th Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Conf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Formal Ontol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Inf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Syst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Rio de Janeiro, Brazil, Sep 22-25, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [12] Object Management Group (OMG) Business Process Model and Notation (BPMN).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Version 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='0, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Accessed: March 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Available: http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='bpmn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='org [13] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Al-Fedaghi and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Aldamkhi, ―Conceptual modeling of an IP phone communication system: A case study,‖ Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Interdisc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Telecom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Netw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 13, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 83-94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [14] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Al-Fedaghi, ―Conceptual modeling founded on the Stoic ontology: Reality with dynamic existence and static subsistence,‖ arXiv preprint, arXiv:2212.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='05300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [15] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Al-Fedaghi, ―Modeling system events and negative events using thinging machines based on Lupascian logic,‖ arXiv preprint arXiv:2211.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='12586.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 17 The trace of events for available car Create Transfer Loan request Release Transfer Loaner Receive E10 E2 E7 E1 Process: (a) Confirm Customer Process: if confirm Create Transfer Car unavailable Create (c) Order from factory?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Not available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='(b) Available ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order to ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='From factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create: ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Order ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='To ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Process ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Create ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Request to ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='confirm ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='from ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='factory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Receive ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Loan ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/reject ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approval/r ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='ejected ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Preparation ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='request ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Cars ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='New Car Sale ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='approved ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Finance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='rejected ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Transfer ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='Release ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E14 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E12 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E11 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E16 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='E9 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='[16] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Hai, ―Heidegger: The fourfold,‖ Blog, Accessed: Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 10, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' https://that-which.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='com/heidegger-the-fourfold/ [17] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Romer, ―Complementarity of process and substance,‖ arXiv:quant- ph/0602171v1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [18] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Russell, ―The philosophy of logical atomism,‖ in Logic and Knowledge, London: George Allen & Unwin, 1968.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [19] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Hoernlé, ―‗Existence‘ and ‗subsistence‘ in contemporary logic and epistemology,‖ in Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Sixth Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Cong.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Phil.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', 1927, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 261-271.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='5840/wcp6192750 [20] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Meinong, ―On the Theory of Objects,‖ in Realism and the Background of Phenomenology, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Chisholm, Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Glencoe, Ill.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=': Free Press, 1960, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 76-117.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [21] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Post, Metaphysics: A Contemporary Introduction, New York: Paragon House, 1991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [22] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Galton, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' ―The ontology of states, processes, and events,‖ in Interdisciplinary Ontology, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Okada and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Smith, Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' fifth Interdisc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Ontol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Mtg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Tokyo: Keio University, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 35-45.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='] [23] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Bailey, ―The Structure of Stoic metaphysics,‖ in Oxford Studies in Ancient Philosophy, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' XLVI, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Inwood, Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Summer, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' https://philpapers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='org/archive/BAITSO-21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='pdf [24] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Greene, ―The imperfect present: Stoic physics of time,‖ PhD dissertation, Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' California, San Diego, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' https://escholarship.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='org/content/qt3jz5n2pk/qt3jz5n2pk_noSplash_0fb2 f7fcc5c31c232cbaab793ef4f8ad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='pdf [25] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Betti, Against Facts, Cambridge, MA: MIT Press, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [26] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Casati and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Achille, ―Events,‖ in The Stanford Encyclopedia of Philosophy, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Zalta, Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Summer 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' https://plato.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='stanford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='edu/archives/sum2020/entries/events/ [27] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Brenner, ―Consciousness as process: A new logical perspective,‖ Philosophy and Computers, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 18, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 1, Fall 2018, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 10-24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' https://cdn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='ymaws.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='com/www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='apaonline.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='org/resource/collection /EADE8D52-8D02-4136-9A2A-729368501E43/ComputersV18n1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='pdf [28] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Chishti, ―Towards a general framework for business process modelling,‖ Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Intell.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Comp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 5, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 3, September 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [29] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Harmon, ―What is a business process,‖ Bus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Process Trends, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 8, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 21, December 14, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [30] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Rummler and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Brache, Improving Performance, 2nd ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', San Francisco: Jossey-Bass, 1995.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [31] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Silver, BPMN Method and Style, 2nd ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Altadena, CA: Cody- Cassidy Press, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' [32] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Bollen, ―Fact-based meta model for BPMN,‖ in ERP Future, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Piazolo et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=', Eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' LNBIP 285, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 29–41, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' DOI: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content='1007/978-3- 319-58801-8_3 Process: (a) Confirm Customer Process: if confirm Create Car unavailable Create (c) Order from factory?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' Not available Create Preparation Create Sale Finance (b) Available Create Order to Factory Process Process From factory Create Process Create Process Loan request Create: OK Receive Order Create To Factory Process Create Request to confirm from factory Loaner Loan approval/rejection Finance approval/ rejected Preparation request Cars New Car Sale Finance approved Finance rejected Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} +page_content=' 18 Simplification of the static model of the new car sale' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/fNA0T4oBgHgl3EQfHf-p/content/2301.02062v1.pdf'} diff --git a/g9E_T4oBgHgl3EQf3Rw9/content/tmp_files/2301.08345v1.pdf.txt b/g9E_T4oBgHgl3EQf3Rw9/content/tmp_files/2301.08345v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..70a2c2ac12aed000665c5b8aa396cb608721ff19 --- /dev/null +++ b/g9E_T4oBgHgl3EQf3Rw9/content/tmp_files/2301.08345v1.pdf.txt @@ -0,0 +1,2737 @@ +Noname manuscript No. +(will be inserted by the editor) +Complexity of linearized augmented Lagrangian for +optimization with nonlinear equality constraints +Lahcen El Bourkhissi · Ion Necoara +Received: date / Accepted: date +Abstract In this paper, we consider a nonconvex optimization problem with +nonlinear equality constraints. We assume that both, the objective function +and the functional constraints are locally smooth. For solving this problem, +we propose a linearized augmented Lagrangian method, i.e., we linearize the +functional constraints in the augmented Lagrangian at the current iterate and +add a quadratic regularization, yielding a subproblem that is easy to solve, and +whose solution is the next iterate. Under a dynamic regularization parameter +choice, we prove global asymptotic convergence of the iterates to a critical +point of the problem. We also derive convergence guarantees for the iterates +of our method to an ϵ first-order optimal solution in O(1/ϵ2) outer iterations. +Finally, we show that, when the problem data are e.g., semialgebraic, the +sequence generated by our algorithm converges and we derive convergence +rates. We validate the theory and the performance of the proposed algorithm +by numerically comparing it with the existing methods from the literature. +Keywords Nonconvex optimization · linearized augmented Lagrangian · +nonlinear functional constraints · convergence analysis. +The research leading to these results has received funding from: the European Union’s +Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant +agreement No. 953348; NO Grants 2014-2021, RO-NO- 2019-0184, under project ELO-Hyp, +contract no. 24/2020; UEFISCDI, PN-III-P4-PCE-2021-0720, under project L2O-MOC, nr. +70/2022. +Lahcen El Bourkhissi +Department of Automatic Control and Systems Engineering, University Politehnica +Bucharest, Bucharest, 060042, Romania +E-mail: lel@stud.acs.upb.ro +Ion Necoara +Department of Automatic Control and Systems Engineering, University Politehnica +Bucharest, Bucharest, 060042, Romania. +Gheorghe Mihoc-Caius Iacob Institute of Mathematical Statistics and Applied Mathematics +of the Romanian Academy, 050711 Bucharest, Romania. +E-mail: ion.necoara@upb.ro +arXiv:2301.08345v1 [math.OC] 19 Jan 2023 + +2 +L. El Bourkhissi, I. Necoara +Mathematics Subject Classification (2020) 68Q25 · 90C06 · 90C30. +1 Introduction +In many fields, such as machine learning, matrix optimization, statistics, con- +trol and signal processing, one finds applications that can be recast as noncon- +vex optimization problems with nonlinear functional equality constraints, see, +e.g., [10,17,19,24]. In this paper, we solve this optimization problem by means +of an augmented Lagrangian approach. The augmented Lagrangian method, +also known as the method of multipliers, was initially proposed in [14,16,22] +to minimize objective functions subject to (linear) equality constraints. It pro- +vides many theoretical advantages, even for non-convex problems, e.g., no du- +ality gap and exact penalty representation, see [23]. Moreover, the augmented +Lagrangian framework is at the heart of the Alternating Direction Method +of Multipliers (ADMM), a very efficient method for large-scale (distributed) +optimization [5–8,11,12,30]. +The augmented Lagrangian approach has been extensively studied in the lit- +erature for convex problems, see e.g., [2–4,7,12,26], and recently it has been +extended to non-convex (smooth/non-smooth) problems with linear equal- +ity constraints in [15, 17, 18, 30, 32]. However, there are very few studies on +the use of the augmented Lagrangian framework for nonconvex optimization +with nonlinear equality constraints, see [8,29]. In [29] a Proximal Augmented +Lagrangian (Proximal AL) is proposed to solve nonconvex, but smooth op- +timization problems. In Proximal AL a static regularization (proximal) term +is added to the original augmented Lagrangian function. The authors prove +that when an approximate first- (second)-order solution of the subproblem is +found, with an error approaching asymptotically zero, then an ϵ− first (ϵ− +second)-order solution of the original problem is obtained within O(1/ϵ2−η) +outer iterations, where η ∈ [0, 2] is an user-defined parameter. Note that when +η is close to 2, the complexity reduces to O(1) outer iterations, but the sub- +problem, which is already nonconvex due to the nonlinearity of the constraints, +gets ill-conditioned as the penalty parameter in the augmented Lagrangian is +inversely proportional to ϵη. In addition, the total iteration complexity is de- +rived when the Newton conjugate gradient method from [25] is used to solve +the subproblem at each outer iteration of Proximal AL algorithm. In [30], a +PL-ADMM was proposed to solve nonsmooth nonconvex problems with linear +equality constraints for two blocks. In this setting, the problem is smooth in +one block and nonsmooth in the other. PL-ADMM algorithm uses a dynamic +metric proximal term in the primal update and an overrelaxation step in the +dual update. It is proved in [30] that the sequence generated by this algo- +rithm is bounded and each limit point of this sequence is a stationary point +of the original problem. Additionally, it is shown that under the Kurdyka- +�Lojasiewicz (K�L) property, the error converges either in a finite number of +iterations, sublinearly or linearly, depending on the exponent associated with + +Linearized augmented Lagrangian for nonconvex optimization +3 +the K�L property. Similar convergence results are obtained in [8] for an al- +gorithm solving problems having nonlinear constraints in the first block (for +which the corresponding objective function is smooth) and linear constraints +in the second block. The algorithm proposed in [8], called DAM (Dynamic +linearized Alternating direction method of Multipliers), linearizes the smooth +part of the augmented Lagrangian and adds a proximal regularization, where +the proximal parameter is dynamically generated. +A different approach, that does not rely on the augmented Lagrangian frame- +work, is presented in [20, 27] and is called Sequential Convex Programming +(SCP). This method solves a sequence of convex approximations of the orig- +inal problem by linearizing the nonconvex parts of the objective and of the +functional constraints and preserving the structures that can be exploited by +convex optimization techniques. In this case the subproblem has a (strongly) +convex objective and linear constraints, for which efficient solution methods +exist, e.g., [13,21]. However, to the best of our knowledge, SCP methods con- +verge under mild assumptions only locally [20,27]. +Due to the high cost of solving the nonconvex subproblem in Proximal AL [29] +and the fact that SCP type schems [20,27] have only local convergence guar- +antees, in this paper we propose a novel Linearized Augmented Lagrangian +method (called Linearized AL) for solving smooth nonconvex problems with +nonlinear equality constraints, that overcomes these limitations. More pre- +cisely, in our algorithm we linearize the functional constraints in the aug- +mented Lagrangian at the current iterate and add a dynamic regularization +term, yielding a subproblem that is easy to solve, and whose solution is the next +iterate. Indeed, if the objective function is (weakly) convex, our subproblem is +strongly convex and thus can be solved efficiently. Hence, our method borrows +the advantages of both approaches, Proximal AL and SCP, since it converges +globally and the subproblem is strongly convex and thus easy to solve. We +prove that the sequence generated by our algorithm is bounded, each limit +point of this sequence is a stationary point and under the K�L property we +prove convergence for the whole sequence. More precisely, under a dynamic +regularization parameter choice, we prove global asymptotic convergence of +the iterates to a critical point of the problem. We also derive convergence +guarantees for the iterates of our method to an ϵ first-order optimal solution +in O(1/ϵ2) outer iterations. Finally, we show that under the K�L property the +whole sequence generated by the proposed algorithm converges and we derive +convergence rates depending on the KL parameter. Unlike [29], in our method +the penalty parameter is independent on the required accuracy ϵ. Moreover, +in comparison with [20], our convergence results here are global, while only +local convergence has been proved for SCP. The preliminary numerical results +confirm the efficiency of our Linearized AL algorithm. +The paper is structured as follows. In Section 2, we introduce our problem of +interest and some notions necessary for our analysis. In Section 3, we present +our algorithm, followed in Section 4 by its convergence analysis. Finally, in +Section 5, we compare numerically our method with existing algorithms. + +4 +L. El Bourkhissi, I. Necoara +2 Problem formulation and preliminaries +In this paper, we consider the following nonlinear optimization problem: +min +x∈Rn +f(x) +s.t. +F(x) = 0, +(1) +where f : Rn → R and F(x) ≜ (f1(x), ..., fm(x))T , with fi : Rn → R for +all i = 1 : m. We assume the functions f, fi ∈ C2 for all i = 1 : m, f is +(weakly) convex and F is nonlinear. For this problem we consider the following +assumptions: +Assumption 1 Assume that there exists ρ0 ≥ 0 such that f(x) + ρ0 +2 ∥F(x)∥2 +has compact level sets, i.e., for all α ∈ R, the following set is empty or compact: +S0 +α ≜ {x : f(x) + ρ0 +2 ∥F(x)∥2 ≤ α}. +Assumption 2 Given a compact set S ⊆ Rn, there exist positive constants +Mf, MF , σ, Lf, LF such that f and F satisfy the following conditions: +(i) ∥∇f(x)∥ ≤ Mf, +∥∇f(x) − ∇f(y)∥ ≤ Lf∥x − y∥ for all x, y ∈ S. +(ii) ∥∇F(x)∥ ≤ MF , +σmin(∇F(x)) ≥ σ > 0 for all x, y ∈ S. +(iii) ∥∇F(x) − ∇F(y)∥ ≤ LF ∥x − y∥ for all x, y ∈ S. +Assumption 3 There exists finite ¯U such that f(x) ≤ ¯U for all x ∈ {x : +∥F(x)∥ ≤ 1}. +Note that these assumptions are standard in the nonconvex optimization lit- +erature, see e.g., [8,29]. In fact, these assumptions are not restrictive because +they need to hold only locally. Indeed, large classes of problems satisfy these +assumptions as discussed below. +Remark 1 Assumption 1 holds e.g., when f(x) + ρ0∥F(x)∥2 is coercive for +some ρ ≥ 0, or when f(x) is strongly convex or f(x) is bounded bellow and +F(x) = xT x−1, as is the case in dictionary learning applications. It also holds +when f(x) = 1 +2xT Qx − pT x, F(x) = Ax − b and Q is a positive definite matrix +on null(A) := {x : Ax = 0}. +Remark 2 Assumption 2 allows general classes of problems. In particular, con- +ditions (i) hold if f(x) is differentiable and ∇f(x) is locally Lipschitz continu- +ous on a neighborhood of S. Conditions (ii) hold when F(x) is differentiable on +a neighborhood of S and satisfies an LICQ condition over S. Finally, condition +(iii) holds if ∇F(x) is locally Lipschitz continuous on S. +Remark 3 For Assumption 3 to hold, it is sufficient the set {x : ∥F(x)∥ ≤ 1} +to be compact. In fact, we do not need this assumption if we can choose the +starting point x0 such that F(x0) = 0, that is, the initial point is feasible. +The following lemma is an immediate consequence of Assumption 1. + +Linearized augmented Lagrangian for nonconvex optimization +5 +Lemma 1 If Assumption 1 holds, then f(x) + ρ0 +2 ∥F(x)∥2 is lower bounded: +¯L ≜ inf +x∈Rn{f(x) + ρ0 +2 ∥F(x)∥2} > −∞. +(2) +Proof See Appendix. +Further, let us introduce the following definitions (see also [29]): +Definition 1 [First-order solution and ϵ first-order solution of (1)] The vector +x∗ is said to be first-order solution of (1) if ∃λ∗ ∈ Rm such that: +∇f(x∗) + ∇F(x∗)T λ∗ = 0 +and +F(x∗) = 0. +Moreover, x∗ is an ϵ first-order solution of (1) if ∃λ∗ ∈ Rm such that: +∥∇f(x∗) + ∇F(∗x)T λ∗∥ ≤ ϵ +and +∥F(x∗)∥ ≤ ϵ. +Let Φ : Rd → ¯R be a proper lower semicontinuous function. For −∞ < τ1 < +τ2 ≤ +∞, we define [τ1 < Φ < τ2] = {x ∈ Rd : τ1 < Φ(x) < τ2}. Let +τ ∈ (0, +∞]. We denote by Ψτ the set of all continuous concave functions +ϕ : [0, τ] → [0, +∞) such that ϕ(0) = 0 and ϕ is continuously differentiable on +(0, τ), with ϕ′(s) > 0 over (0, τ). +Definition 2 Let Φ : Rd → ¯R be a proper lower semicontinuous function that +takes constant value on Ω. We say that Φ satisfies the K�L property on Ω if there +exists ϵ > 0, τ > 0, and ϕ ∈ Ψτ such that for every x∗ ∈ Ω and every element +x in the intersection {x ∈ Rd : dist(x, Ω) < ϵ} ∩ [Ψ(x∗) < Ψ(x) < Ψ(x∗) + τ], +we have: +ϕ′� +Φ(x) − Φ(x∗) +� +dist +� +0, ∂Φ(x) +� +≥ 1. +This definition covers many classes of functions arising in practical optimiza- +tion problems. For example, if f is a proper closed semialgebraic function, +then f is a K�L function with exponent ν ∈ [0, 1) [1]. The function g(Ax), +where g is strongly convex on a compact set and twice differentiable, and +A ∈ Rm×n, is a K�L function. Convex piecewise linear quadratic functions such +as ∥x∥1, ∥x∥0, γ �k +i=1 |x[i]|, where |x[i]| is the ith largest entry in x, k ≤ n +and γ ∈ (0, 1]; the indicator function δ∆(x), where ∆ = {x ∈ Rn : eT x = +1, x ≥ 0}; least-squares problems with the Smoothly Clipped Absolute Devia- +tion (SCAD) [9]; Minimax Concave Penalty (MCP) regularized functions [31] +are all K�L functions. The K�L property characterizes the local geometry of a +function around the set of critical points. + +6 +L. El Bourkhissi, I. Necoara +3 A linearized augmented Lagrangian method +In this section, we propose a new algorithm for solving problem (1) using the +augmented Lagrangian framework. Let us first introduce few notations. The +augmented Lagrangian associated with the problem (1) is: +Lρ(x, λ) = f(x) + ⟨λ , F(x)⟩ + ρ +2∥F(x)∥2 = f(x) + ψ(x, λ), +where ψ(x, λ) = ⟨λ , F(x)⟩ + ρ +2∥F(x)∥2. The gradient of ψ is: ∇xψ(x, λ) = +∇F(x)T (λ + ρF(x)) and ∇λψ(x, λ) = F(x). In the rest of this paper, for +the sake of clarity, we provide the proofs of all the lemmas in Appendix. +The following lemma shows that ψ is locally smooth, (i.e., it has Lipschitz +continuous gradient locally). +Lemma 2 [Smoothness of ψ] If Assumption 2 holds, then for any compact set +Λ ⊂ Rm there exists Lψ > 0 such that: +∥∇ψ(x, λ) − ∇ψ(x′, λ′)∥ ≤ Lψ +���� +�x − x′ +λ − λ′ +����� +∀(x, λ), (x′, λ′) ∈ S × Λ, +where Lψ = sup(x,λ)∈S×Λ{LF ∥λ + ρF(x)∥ + MF (2 + ρMF )}. +Proof See Appendix. +Further, let us denote the following function derived from linearization of the +functional constraints, at a given point ¯x, in the augmented Lagrangian: +¯Lρ(x, λ; ¯x) += f(x) + ⟨λ , F(¯x) + ∇F(¯x)(x − ¯x)⟩ + ρ +2∥F(¯x) + ∇F(¯x)(x − ¯x)∥2. +To solve the optimization problem (1) we propose the following Linearized +augmented Lagrangian (Linearized AL) algorithm, i.e., we linearize the func- +tional constraints in the augmented Lagrangian at the current iterate and add +a quadratic regularization. +Algorithm 1 Linearized augmented Lagrangian +1: Initialization: x0, λ0, and ρ > 0; +2: k ← 0 +3: while stopping criterion is not satisfied do +4: +generate a proximal parameter βk+1 > 0 +5: +xk+1 ← arg minx ¯Lρ(x, λk; xk) + βk+1 +2 +∥x − xk∥2 +▷ The subproblem +6: +λk+1 ← λk + ρ +� +F(xk) + ∇F(xk)(xk+1 − xk) +� +7: +k ← k + 1 +8: end while +To the best of our knowledge Linearized AL algorithm is new and its conver- +gence behaviour has not been analyzed before in the literature. Note that if + +Linearized augmented Lagrangian for nonconvex optimization +7 +f is (weakly) convex function, the objective function in the subproblem of +step 5 of Algorithm 1 is strongly convex, provided that βk+1 is chosen ad- +equately. In particular, if f is a quadratic function, then finding a solution +of the subproblem in step 5 is even equivalent with solving a linear system +of equalities. In these cases, efficient solution methods exist for solving the +subproblem, see e.g., [13, 21]. It is also important to note that our update +of the dual multipliers is different from the literature, i.e., instead of eval- +uating the functional constraints at the new test point xk+1 and updating +λk+1 = λk + ρF(xk+1) [8,29], we evaluate their linearization at xk in the new +point xk+1 and update λk+1 = λk + ρ(F(xk) + ∇F(xk)(xk+1 − xk)). In the +sequel, we denote: +∆xk = xk − xk−1 +and +∆λk = λk − λk−1 +∀k ≥ 1. +Let α ∈ (0, 1). Then, from Lemma 2 we can choose βk+1 such that: +ψ(xk+1, λk) − ψ(xk, λk) − ⟨∇xψ(xk, λk) , xk+1 − xk⟩ +≤ (1 − α)βk+1 +2 +∥xk+1 − xk∥2 +∀k ≥ 0. +(3) +Note that for any k ≥ 1, βk is well defined since ψ is smooth, see Lemma +2. This regularization parameter βk can be found using e.g., a backtracking +scheme as in [8]. In the sequel, we assume that βk satisfies the following: +Assumption 4 [Proximal parameter sequence is bounded] Assume that the +generated parameter sequence {βk}k≥1 is bounded, i.e., +β := sup +k≥1 +βk < ∞. +4 Convergence analysis +In this section, we derive the asymptotic convergence of Linearized AL algo- +rithm (Algorithm 1) and its efficiency to obtain an ϵ first-order solution for +problem (1). Moreover, we provide convergence rates under the K�L condition. +Note that for some intermediate steps in the convergence analysis we follow +a similar methodology as in [29]. Let us start by proving the decrease with +respect to the first argument for the augmented Lagrangian function. +Lemma 3 [Descent of Lρ w.r.t. primal variables] If Assumption 2 holds, then +we have the following descent: +Lρ(xk+1, λk) ≤ Lρ(xk, λk) − ρσ2 + αβk+1 +2 +∥xk+1 − xk∥2 +∀k ≥ 0. +Proof See Appendix. + +8 +L. El Bourkhissi, I. Necoara +Let us define the following Lyapunov function (inspired from [29]): +P(x, λ, y, γ) = Lρ(x, λ) + γ +2 ∥x − y∥2. +(4) +The evaluation of the Lyapunov function along the iterates of Linearized AL +algorithm is denoted by: +Pk = P(xk, λk, xk−1, γk) +∀k ≥ 1, +(5) +with γk > 0 to be defined later. We prove next that {Pk}k≥1 is decreasing and +bounded from bellow. Let us first prove that it is decreasing. Indeed: +P k+1 − Pk +=Lρ(xk+1, λk+1) − Lρ(xk+1, λk) + Lρ(xk+1, λk) − Lρ(xk, λk) ++ γk+1 +2 +∥xk+1 − xk∥2 − γk +2 ∥xk − xk−1∥2 +≤ +� +F(xk+1) − F(xk) − ∇F(xk)(xk+1 − xk) , ∆λk+1 +� ++ 1 +ρ∥∆λk+1∥2 +− ρσ2 + αβk+1 − γk+1 +2 +∥∆xk+1∥2 − γk +2 ∥∆xk∥2 +≤ +� +1 +2 η√ρη−1 + 1 +ρ +� +∥∆λk+1∥2 − γk +2 ∥∆xk∥2 +− +�ρσ2 + αβk+1 − γk+1 +2 +− 2M 2 +F +η√ρη−1 +� +∥∆xk+1∥2, +(6) +where the first inequality follows from Lemma 3 and from the update of the +dual multipliers in Step 6 of Algorithm 1. The second inequality holds for any +η > 1 and follows from Assumption 2 and the inequality: +⟨a , b⟩ ≤ 1 +2r∥a∥2 + r +2∥b∥2 +∀a, b ∈ Rm and r > 0. +(7) +Next, let us bound ∥∆λk+1∥2. +Lemma 4 [Bound for ∥∆λk+1∥] If Assumption 2 holds on some compact set +S and the sequence generated by Algorithm 1 is in S, then we have: +∥∆λk+1∥2 ≤ c1(βk+1)∥∆xk+1∥2 + c2(βk)∥∆xk∥2, +(8) +where c1(βk+1) = 2 (Lf +βk+1)2 +σ2 +and +c2(βk) = 2 (Mf LF +(2MF +σ)βk)2 +σ4 +. +Proof See Appendix. + +Linearized augmented Lagrangian for nonconvex optimization +9 +Subsequently, using (8) in (6), we obtain that: +Pk+1 − Pk ≤ +� �� +1 +2 η√ρη−1 + 1 +ρ +� +c1(βk+1) − γk+1 +2 +� +− +η√ρη−1( η√ρσ2 − 4M 2 +F ) + αβk+1 − 2γk+1 +2 +� +∥∆xk+1∥2 ++ +�� +1 +2 η√ρη−1 + 1 +ρ +� +c2(βk) − γk +2 +� +∥∆xk∥2. +Hence, if we choose γk ≤ αβk +2 +and ρ ≥ ( 4M 2 +F +σ2 )η, we get: +Pk+1 − Pk ≤ +�� +1 +2 η√ρη−1 + 1 +ρ +� +c1(βk+1) − γk+1 +2 +� +∥∆xk+1∥2 ++ +�� +1 +2 η√ρη−1 + 1 +ρ +� +c2(βk) − γk +2 +� +∥∆xk∥2. +Therefore, in order to obtain a decreasing Lyapunov function along iterates, +we must take into account when choosing γk the following: +γk > 2 +� +1 +2 η√ρη−1 + 1 +ρ +� +max{c1(βk), c2(βk)}. +Hence, we want the parameter γk to satisfy the following requirements: +2 +� +1 +2 η√ρη−1 + 1 +ρ +� +max{c1(βk), c2(βk)} < γk ≤ αβk +2 +∀k ≥ 1. +(9) +Choosing γk = 4 +� +1 +2 η√ρη−1 + 1 +ρ +� +max{c1(βk), c2(βk)}, one can see that it al- +ready satisfies the left inequality in (9). It remains to check the right inequality +in (9). To do so, we have to check if there exists a βk > 0 which satisfies the +following inequality: +8 +� +1 +2 η√ρη−1 + 1 +ρ +� +max{c1(βk), c2(βk)} ≤ αβk. +(10) +Let us determine max{c1(βk), c2(βk)}. Using the definitions of c1(βk) and +c2(βk), we have: +c1(βk) − c2(βk) += 2 +σ2 +�� +Lf − MfLF +σ +� +− 2MF +σ +βk +� �� +Lf + MfLF +σ +� ++ +�2MF +σ ++ 2 +� +βk +� +, +which leads to two different cases. + +10 +L. El Bourkhissi, I. Necoara +Case 1: If βk ≥ Lf σ−Mf LF +2MF +, then max{c1(βk), c2(βk)} = c2(βk). Let us now +check if βk, which satisfies (10), is well-defined in this case. To do this, we +replace the expression of c2(βk) in (10) and reformulate (10) as follows: +16 +σ4 +� +1 +2 η√ρη−1 + 1 +ρ +� � +(2MF + σ)2 β2 +k + 2 (2MF + σ) MfLF βk + (MfLF )2� +≤ αβk. +(11) +Denoting: +h1(βk) =16 +σ4 +� +1 +2 η√ρη−1 + 1 +ρ +� +(2MF + σ)2 β2 +k + 16 +σ4 +� +1 +2 η√ρη−1 + 1 +ρ +� +(MfLF )2 ++ +� +32 +σ4 +� +1 +2 η√ρη−1 + 1 +ρ +� +(2MF + σ) MfLF − α +� +βk, +the inequality (11) is then equivalent to h1(βk) ≤ 0. Since the roots of the +second order equation h1(βk) = 0 are: +¯ +β1 = δ +� +1 − +� +1 − 1/δ2 +� +MfLF +2MF + σ +and +¯β1 = δ +� +1 + +� +1 − 1/δ2 +� +MfLF +2MF + σ , +it follows that (9) and (11) are verified for any βk satisfying +¯ +β1 ≤ βk ≤ ¯β1. +Also, it is clear that when δ → ∞, the interval [ +¯ +β1, ¯β1] → (0, ∞). Therefore, we +should choose δ large enough so that ¯β1 ≥ β. If it happens that for some k ≥ 1, +we obtain from (3), Lf σ−Mf LF +2MF +≤ βk < +¯ +β1, then we take βk = +¯ +β1. Note that +when choosing α one has to note that for α → 0, ρ → ∞, whereas for α → 1, +the proximal parameter βk might be large and similar β from Assumption 4. +Case 2: Similarly, if βk < +Lf σ−Mf LF +2MF +, then max{c1(βk), c2(βk)} = c1(βk). +Let us check if βk, that satisfies (10), is well-defined. Replacing c1(βk) by its +expression in (10), it allows us to recast (10) as: +16 +σ2 +� +1 +2 η√ρη−1 + 1 +ρ +� +� +β2 +k + 2Lfβk + L2 +f +� +≤ αβk. +(12) +Let us denote: +h2(βk) =16 +σ2 +� +1 +2 η√ρη−1 + 1 +ρ +� +β2 +k + 16 +σ2 +� +1 +2 η√ρη−1 + 1 +ρ +� +L2 +f ++ +� +32 +σ2 +� +1 +2 η√ρη−1 + 1 +ρ +� +Lf − α +� +βk, +then the inequality (12) is equivalent to h2(βk) ≤ 0. Since the roots of the +equation h2(βk) = 0 are: +¯ +β2 = δ′ � +1 − +� +1 − 1/δ′2 +� +Lf +and +¯β2 = δ′ � +1 + +� +1 − 1/δ′2 +� +Lf, + +Linearized augmented Lagrangian for nonconvex optimization +11 +it follows that (9) and (12) are valid for all +¯ +β2 ≤ βk ≤ ¯β2. We choose δ′ +such that ¯β2 = Lf σ−Mf LF +2MF +(otherwise, if it happens that βk > Lf σ−Mf LF +2MF +, we +have c2(βk) = max{c1(βk), c2(βk)} and then we are in the first case discussed +previously). If for some k ≥ 1, we get from (3) βk < +¯ +β2, then we take βk = +¯ +β2. +In conclusion, from the two cases above, if we choose: +ρ ≥ max +��4M 2 +F +σ2 +�η +, +�48(δ′ + 1) +ασ2 +Lf +� +η +η−1 +, +�48(δ + 1) +ασ4 +(2MF + σ)MfLF +� +η +η−1� +, +γk = 4 +� +1 +2 η√ρη−1 + 1 +ρ +� +max{c1(βk), c2(βk)}, +then we have: +γk ≤ αβk +2 +∀k ≥ 1. +(13) +This, in turn, implies that the Lyapunov function decreases along the iterates: +Pk+1 − Pk ≤ −γk+1 +4 +∥∆xk+1∥2 − γk +4 ∥∆xk∥2 +∀k ≥ 1. +(14) +In the sequel, we assume that x0 is chosen such that: +∥F(x0)∥2 ≤ min +� +1, c0 +ρ +� +for some c0 > 0, +(15) +and that f(x0) ≤ ¯U. Let us define: +ˆα ≜ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 3, +(16) +and DS := max{∥x − y∥ : x, y ∈ S0 +ˆα}. Moreover, let us choose: +ρ ≥ max +��4M 2 +F +σ2 +�η +, +�48(δ′ + 1) +ασ2 +Lf +� +η +η−1 +, +�48(δ + 1) +ασ4 +(2MF + σ)MfLF +� +η +η−1 +, +3ρ0, ρ0 + +�Mf(2MF + σ) + 2δMfLF DS +√ +2σ(2MF + σ) +�2 � +. +(17) +Note that, if we choose δ large enough so that ¯β1 ≥ β, we get: +(Mf(2MF + σ) + 2δMfLF DS)2 +2σ2(2MF + σ)2 +≥ (Mf + βDS)2 +2σ2 +. +It follows that ρ ≥ ρ0 + (Mf +βDS)2 +2σ2 +and 4M 2 +F +σ2 +> 1, which imply that ρ ≥ 1. +Using the definition of Lρ, we have: +Lρ(x0, λ0) = f(x0) + ⟨λ0 , F(x0)⟩ + ρ +2∥F(x0)∥2 +(7) +≤f(x0) + ∥λ0∥2 +2ρ ++ ρ +2∥F(x0)∥2 + ρ +2∥F(x0)∥2 +(15) +≤ ¯U + 1 +2ρ∥λ0∥2 + c0. +(18) + +12 +L. El Bourkhissi, I. Necoara +It then follows, after some re-arrangements, that: +¯U + c0 − ¯L ≥ f(x0) + ⟨λ0 , F(x0)⟩ + ρ +2∥F(x0)∥2 − 1 +2ρ∥λ0∥2 − ¯L +(ρ≥3ρ0) +≥ +f(x0) + ρ0 +2 ∥F(x0)∥2 − ¯L + ⟨λ0 , F(x0)⟩ + ρ +3∥F(x0)∥2 − ∥λ0∥2 +2ρ +(2) +≥ ρ +3∥F(x0) + 3λ0 +2ρ ∥2 − 3 +4ρ∥λ0∥2− ∥λ0∥2 +2ρ +(ρ≥1) +≥ +− 5 +4∥λ0∥2. +(19) +The following lemma shows that the sequence {(xk, λk)}k≥1 generated by Al- +gorithm 1, is bounded. +Lemma 5 Consider Algorithm 1 and let {Pk}k≥1 as defined in (5). If As- +sumptions 1, 2, 3 and 4 hold with S = S0 +ˆα and ˆα defined in (16) for any fixed +constant c0 and DS the radius of Sˆα0. If ρ is chosen as in (17) and x0 is +chosen to satisfy (15), then we have the following: +xk ∈ S0 +ˆα, +(20a) +∥λk∥2 ≤ (Mf + βDS)2 +σ2 +≤ 2(ρ − ρ0), +(20b) +Pk ≤ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 2 +∀k ≥ 1. +(20c) +Proof See Appendix. +Next, we show that the Lyapunov sequence {Pk}k≥1 is bounded from below. +Lemma 6 Consider Algorithm 1 and let {Pk}k≥1 defined in (5). If Assump- +tions 1, 2, 3 and 4 hold with S = S0 +ˆα and ˆα defined in (16) for any fixed +constant c0 and DS the radius of Sˆα0. If ρ is chosen as in (17) and x0 is +chosen to satisfy (15), then we have the following: +Pk ≥ ¯L − 1 +∀k ≥ 1 +(21) +where ¯L is defined in (2). +Proof See Appendix. +Let us now bound the gradient of the augmented Lagrangian function. +Lemma 7 [Boundedness of ∇Lρ] Let {(xk, λk)}k≥1 be a sequence generated +by the Algorithm 1. If Assumption 2 holds, then: +∥∇Lρ(xk+1, λk+1)∥ ≤ Γ(βk+1)∥xk+1 − xk∥ + Γ(βk)∥xk − xk−1∥, +where the function +Γ(βk) = +� +MF + 1 +ρ +� MfLF + MF Lf + (3MF + σ)βk +σ2 ++ 2MF (ρMF + 1). + +Linearized augmented Lagrangian for nonconvex optimization +13 +Proof See Appendix. +Note that if Assumption 4 is satisfied, then {Γ(βk)}k≥1 is finite. Hence, in the +sequel we define this bound as follows: +Γmax := sup +k≥1 +{Γ(βk)}. +(22) +4.1 Global convergence +In this section we prove global convergence for the iterates generated by Algo- +rithm 1 and also convergence rates to an ϵ first-order solution. Based on the +previous lemmas, we are now ready to present the global asymptotic conver- +gence of the iterates of Algorithm 1. +Theorem 1 [Limit points are stationary points] If Assumptions 1, 2, 3 and 4 +hold with S = S0 +ˆα and ˆα defined in (16) for any fixed constant c0 and DS the +radius of Sˆα0. If ρ is chosen as in (17) and x0 is chosen to satisfy (15), then +any limit point (x∗, λ∗) of the sequence {(xk, λk)}k≥1, generated by Algorithm +1, is a stationary point, i.e., ∇Lρ(x∗, λ∗) = 0. Equivalently: +∇f(x∗) + ∇F(x∗)T λ∗ = 0, +F(x∗) = 0. +Proof Using (14) and the fact that γk > 0, we have: +γk+1 +4 +∥∆xk+1∥2 + γk +4 ∥∆xk∥2 ≤ Pk − Pk+1 +∀k ≥ 1. +Let k ≥ 1, by summing up the above inequality from i = 1 to i = k, we obtain: +k +� +i=1 +�γi+1 +4 +∥∆xi+1∥2 + γi +4 ∥∆xi∥2� +≤ P1 − Pk+1 +Lemma 6 +≤ +P1 − (¯L − 1) +(20c) +≤ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 3 − ¯L += ˆα − ¯L. +(23) +Since (23) holds for any k ≥ 1, we have: +∞ +� +i=1 +�γi+1 +4 +∥∆xi+1∥2 + γi +4 ∥∆xi∥2� +< ∞. +This, together with the fact that γk > 0, yields that: +lim +k→∞ ∥∆xk∥ = 0. +(24) +From (20b), (20a) and the fact that S0 +ˆα is compact, it follows that the sequence +{(xk, λk)}k≥1 is bounded and there exists a convergent subsequence, let us say +{(xk, λk)}k∈K, with the limit (x∗, λ∗). From Lemma 7 and (22), we have: +∥∇Lρ(x∗, λ∗)∥ = lim +k∈K ∥∇Lρ(xk, λk)∥ ≤ Γmax lim +k∈K(∥∆xk∥ + ∥∆xk−1∥) = 0. +Therefore, ∇Lρ(x∗, λ∗) = 0, which completes our proof. +⊓⊔ + +14 +L. El Bourkhissi, I. Necoara +For the remainder of this paper, we define: +¯γ := sup +k≥1 +{γk} +(13),Ass. 4 +< +∞ +and +¯ +γ := inf +k≥1{γk} > 0. +(25) +Let us now present one of the main results of this paper, which derives the +outer complexity of Algorithm 1 to find an ϵ first-order solution of problem (1). +Theorem 2 [First-order complexity] Consider Algorithm 1 and let {Pk}k≥1 +be defined as in (5). If Assumptions 1, 2, 3 and 4 hold with S = S0 +ˆα and ˆα +defined in (16) and ρ is chosen as in (17), then for any ϵ > 0, Algorithm +1 yields an ϵ first-order solution of (1) after K = 16Γmax +2 � +ˆα−¯L +¯ +γ +� +1 +ϵ2 outer +iterations. +Proof Let K ≥ 1, then from (23) and (25), we have: +¯ +γ +4 +K +� +i=1 +� +∥∆xi+1∥2 + ∥∆xi∥2� +≤ +K +� +i=1 +�γi+1 +4 +∥∆xi+1∥2 + γi +4 ∥∆xi∥2� +≤ ˆα − ¯L. +Therefore, there exists k∗ ∈ {1, ..., K} such that: +∥∆xk∗+1∥2 + ∥∆xk∗∥2 ≤ 4 ˆα − ¯L +K +¯ +γ . +It implies that: ∥∆xk∗+1∥ ≤ 2 +� +ˆα−¯L +K +¯ +γ +and +∥∆xk∗∥ ≤ 2 +� +ˆα−¯L +K +¯ +γ . Hence, from +Lemma 7 and (22), we get: +∥∇Lρ(xk∗+1, λk∗+1)∥ ≤ Γmax +� +∥∆xk∗+1∥ + ∥∆xk∗∥ +� +≤ 4Γmax +� +ˆα − ¯L +K +¯ +γ . +It follows that for any ϵ > 0, ∥∇Lρ(xk∗+1, λk∗+1)∥ ≤ ϵ when K ≥ 16Γmax +2 � +ˆα−¯L +¯ +γϵ2 +� +. +Consequently, after K = 16Γmax +2( ˆα−¯L +¯ +γ ) 1 +ϵ2 outer iterations Algorithm 1 yields +an ϵ first-order solution of (1). This concludes our proof. +⊓⊔ +4.2 Improved rates under KL +In this section, under the KL condition, we provide better convergence rates +for the iterates of Algorithm 1. In particular, we prove that the whole sequence +{(xk, λk)}k≥1 converges. In order to show these results, we first bound the full +gradient ∇P(·) (recall that P(·) is the function defined in (4)). +Lemma 8 [Boundedness of ∇P] Let {(xk, λk)}k≥1 be the sequence generated +by Algorithm 1. If Assumptions 1, 2, 3 and 4 hold with S = S0 +ˆα and ˆα defined +in (16) for any fixed constant c0, DS is the radius of S0 +ˆα and ρ is chosen as in +(17), then we have for any k ≥ 1: +∥∇P(xk+1, λk+1, xk, γk+1)∥ ≤ (Γmax+DS+2¯γ) (∥xk+1 − xk∥ + ∥xk − xk−1∥) , +where, Γmax and ¯γ are defined in (22) and (25), respectively. + +Linearized augmented Lagrangian for nonconvex optimization +15 +Proof See Appendix. +The above lemma directly implies the following: +∥∇P(xk+1, λk+1, xk, γk+1)∥2 ≤ 2(Γmax + DS + 2¯γ)2 � +∥∆xk+1∥2 + ∥∆xk∥2� +. +(26) +Then, it follows from (26) and (14), that: +Pk+1 − Pk ≤ − +¯ +γ +8(Γmax + DS + 2¯γ)2 ∥∇P(xk+1, λk+1, xk, γk+1)∥2 . +(27) +Let us denote zk = (xk, λk) and uk = (xk, λk, xk−1, γk). Moreover, crit P +denote the set of critical points of the function P(·) defined in (5). Furthermore, +we denote Ek = Pk − P ∗, where P ∗ = limk→∞ Pk (recall that the sequence +{Pk}k≥1 is decreasing and bounded from bellow according to (14) and Lemma +6, respectively, hence it is convergent). Let us denote the set of limit points of +{uk}k≥1 by: +Ω := {u∗ : ∃ a convergent subsequence {uk}k∈K such that lim +k∈K uk = u∗}. +Let us now prove the following lemma. +Lemma 9 Consider Algorithm 1 and let {Pk}k≥1 be defined as in (4). If +Assumptions 1, 2, 3 and 4 hold, with S = S0 +ˆα and ˆα defined in (16) for any +fixed constant c0, DS is the radius of S0 +ˆα and ρ is chosen as in (17), then the +following statements hold: +(i) Ω is a compact subset of crit P and limk→∞ dist(uk, Ω) = 0. +(ii) For any u ∈ Ω, we have P(u) = P ∗. +(iii) For any (x, λ, y, γ) ∈ crit P, we have (x, λ) a stationary point of (1). +Proof See Appendix. +Let us now prove that the sequence +� +∥∆xk∥ + ∥∆λk∥ +� +k≥1 has finite length, +provided that a KL condition holds. +Lemma 10 Let {(xk, λk)}k≥1 be the sequence generated by Algorithm 1. Let +Assumptions 1, 2, 3 and 4 hold, with S = S0 +ˆα and ˆα defined in (16) for any +fixed constant c0, and DS is the radius of S0 +ˆα. Moreover, assume that P(·) +defined in (4) satisfies the K�L property on Ω. Then, {zk}k≥1 = {(xk, λk)}k≥1 +satisfies the finite length property, i.e., +∞ +� +k=1 +∥∆xk∥ + ∥∆λk∥ < ∞, +and consequently converges to a stationary point of (1). +Proof See Appendix. + +16 +L. El Bourkhissi, I. Necoara +Lemma 10 shows that the set of limit points of the sequence {(xk, λk)}k≥1 is +a singleton. Let us denote its limit by (x∗, λ∗). We are now ready to present +the convergence rates of the whole sequence generated by Algorithm 1 (see +also [30] for a similar reasoning). +Theorem 3 [Convergence rates of {(xk, λk)}k≥1] If Assumptions 1, 2, 3 and +4 hold and z∗ := (x∗, λ∗) is a limit point of the sequence {zk := (xk, λk)}k≥1 +generated by Algorithm 1, then: +1. If P(·) satisfies K�L property at u∗ := (x∗, λ∗, x∗, γ∗), where γ∗ is a limit +point of the sequence {γk}k≥1, then there exists k1 ≥ 1 such that for all +k ≥ k1 we have: +∥zk − z∗∥ ≤ C max{ϕ(Ek), +� +Ek−1}, +where C > 0 and ϕ ∈ Ψτ, with τ > 0, denotes a desingularizing function. +2. Moreover, if P(·) satisfies K�L property with desingularizing function +ϕ : [0, τ) → [0, +∞), ϕ(s) = s1−ν, where ν ∈ [0, 1), +then the following rates hold +(a) If ν = 0, then zk converges to z∗ in a finite number of iterations. +(b) If ν ∈ (0, 1 +2), then for all k ≥ k1, we have: +∥zk − z∗∥ ≤ +� +Ek1 +� +(1 + ¯cE2ν−1 +k1 +)k−k1 +, +where ¯c = +¯ +γ +8(Γmax + DS + 2¯γ)2 . +(c) If ν ∈ ( 1 +2, 1), then for all k > k1, we have: +∥zk − z∗∥ ≤ +� +1 +µ(k − k1) + E1−2ν +k1 +� 1−ν +2ν−1 +. +Proof 1. Using Lemma 4, we get: +∥∆λk+1∥2 ≤ c1(β)∥∆xk+1∥2 + c2(β)∥∆xk∥2 +≤ max{c1(β), c2(β)} +� +∥∆xk+1∥2 + ∥∆xk∥2� +. +(28) +Adding the term ∥∆xk+1∥2 + ∥∆xk∥2 on both sides in (28), we obtain: +∥zk+1 − zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2 +≤ ∥∆xk+1∥2 + ∥∆λk+1∥2 + ∥∆xk∥2 +(28) +≤ +� +max{c1(β), c2(β)} + 1 +�� +∥∆xk+1∥2 + ∥∆xk∥2� +. +(29) + +Linearized augmented Lagrangian for nonconvex optimization +17 +Considering (25), we can then rewrite (14) as follows: +Pk+1 − Pk +(14) +≤ − ¯ +γ +4 +� +∥∆xk+1∥2 + ∥∆xk∥2� +(29) +≤ − +¯ +γ +4 +� +max{c1(β), c2(β)} + 1 +�∥zk+1 − zk∥2. +(30) +Based on our choice of ρ and γk, the sequence {Pk}k≥1 is monotonically +decreasing, see (14), and consequently {Ek}k≥1 is monotonically decreasing. +Using (30) and the fact that {Ek}k≥1 is non-negative, we have for all k ≥ 1: +∥∆xk+1∥ + ∥∆λk+1∥ ≤ 2 +� +2 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +� +Ek. +(31) +Without loss of generality, we assume that γ∗ is unique. Since Pk → P ∗, +uk → u∗ and P(·) satisfies the K�L property at u∗, then there exists k1 = +k1(ϵ, τ) ≥ 1 such that ∀k > k1, we have ∥uk − u∗∥ ≤ ϵ and P ∗ < Pk < +P ∗ + τ, and the following K�L property holds: +ϕ′(Ek)∥∇P(xk, λk, xk−1, γk)∥ ≥ 1. +(32) +Since ϕ is concave function, we have ϕ(Ek) − ϕ(Ek+1) ≥ ϕ′(Ek)(Ek − Ek+1). +Therefore, from (30) and (32), we get: +∥zk+1 − zk∥2 +≤ ϕ′(Ek)∥zk+1 − zk∥2∥∇P(xk, λk, xk−1, γk)∥ +≤ 4 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +ϕ′(Ek)(Ek − Ek+1)∥∇P(xk, λk, xk−1, γk)∥ +≤ 4 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +� +ϕ(Ek) − ϕ(Ek+1) +� +∥∇P(xk, λk, xk−1, γk)∥. +One can notice the following trivial inequality: for any a, b, c, d ≥ 0, if +a2 + b2 ≤ c × d, then (a + b)2 ≤ 2a2 + 2b2 ≤ 2c × d ≤ c2 + d2 ≤ (c + d)2. +Using this relation and the fact that ∥zk+1−zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2, +we have for any θ > 0: +∥∆xk+1∥ + ∥∆λk+1∥ ≤4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γ +� +ϕ(Ek) − ϕ(Ek+1) +� ++ 1 +θ∥∇P(xk, λk, xk−1, γk)∥. +(33) +Furthermore, we have: +∥∇P(xk, λk, xk−1, γk)∥ ≤ ∥∇Lρ(xk, λk)∥ + 2¯γ∥xk − xk−1∥ +≤(Γmax + DS + 2¯γ) (∥∆xk∥ + ∥∆λk∥) , + +18 +L. El Bourkhissi, I. Necoara +where the last inequality follows from the definition of Lρ and the properties +of the derivative together with Lemma 7 and (22). Then, (33) becomes: +∥∆xk+1∥ + ∥∆λk+1∥ ≤4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γ +� +ϕ(Ek) − ϕ(Ek+1) +� ++ Γmax + DS + 2¯γ +θ +� +∥∆xk∥ + ∥∆λk∥ +� +. +Let us now choose θ > 0 such that 0 < +Γmax+DS+2¯γ +θ +< 1 and define a +parameter δ0 as δ0 = 1 − Γmax+DS+2¯γ +θ +> 0. Then, summing up the above +inequality over k > k1, we get: +� +k≥k1 +∥∆xk+1∥ + ∥∆λk+1∥ ≤4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γδ0 +ϕ(Ek1) ++ Γmax + DS + 2¯γ +θδ0 +� +∥∆xk1∥ + ∥∆λk1∥ +� +. +Hence, using the triangle inequality, we get for any k ≥ k1: +∥zk − z∗∥ ≤ +� +l≥k +∥zl − zl+1∥ ≤ +� +l≥k +∥∆xl+1∥ + ∥∆λl+1∥ +≤ 4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γδ0 +ϕ(Ek) + Γmax + DS + 2¯γ +θδ0 +� +∥∆xk∥ + ∥∆λk∥ +� +. +Further, using (31), it follows that: +∥zk − z∗∥ ≤ 4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γδ0 +ϕ(Ek) ++ 2(Γmax + DS + 2¯γ) +θδ0 +� +2 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +� +Ek−1 +≤ C max{ϕ(Ek), +� +Ek−1}, +where +C = max +� +4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γδ0 +, +2(Γmax + DS + 2¯γ) +θδ0 +� +2 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +� +. +2. Let ν ∈ [0, 1) and for all s ∈ [0, τ), ϕ(s) = s1−ν and ϕ′(s) = (1 − ν)s−ν. It +follows that ∀k ≥ k1, we have: +∥zk − z∗∥ ≤ C max{E1−ν +k +, +� +Ek−1}. +(34) + +Linearized augmented Lagrangian for nonconvex optimization +19 +Furthermore, (32) yields: +Ek +ν ≤ ∥∇P(xk, λk, xk−1, γk)∥ +∀k ≥ k1. +Moreover, from (27), we have for any k ≥ 1: +∥∇P(xk, λk, xk−1, γk)∥2 ≤ 8(Γmax + DS + 2¯γ)2 +¯ +γ +(Ek−1 − Ek). +Hence, +Ek +2ν ≤ 8(Γmax + DS + 2¯γ)2 +¯ +γ +(Ek−1 − Ek) +∀k > k1. +Setting ¯c = +¯ +γ +8(Γmax+DS+2¯γ)2 > 0, we get the recurrence +¯cEk +2ν ≤ Ek−1 − Ek +∀k > k1. +(a) Let ν = 0. If Ek > 0 for any k > k1, we have ¯c ≤ Ek−1 − Ek. As k goes +to infinity, the right hand side approaches zero. Then, 0 < ¯c ≤ 0 which +is a contradiction. Hence, there exists k > k1 such that Ek = 0. Then, +Ek → 0 in a finite number of steps and from (34), zk → z∗ in a finite +number of steps. +(b) Let ν ∈ (0, 1 +2). Then, 2ν − 1 < 0. Let k > k1. Since {Ei}i≥k1 is mono- +tonically decreasing, then Ei ≤ Ek1 for any i ∈ {k1 + 1, k1 + 2, ..., k} +and +¯cEk1 +2ν−1Ek ≤ Ek−1 − Ek +∀k > k1. +Rearranging this, we get for all k > k1: +Ek ≤ +Ek−1 +1 + ¯cEk1 +2ν−1 ≤ +Ek−2 +(1 + ¯cEk1 +2ν−1)2 ≤ ... ≤ +Ek1 +(1 + ¯cEk1 +2ν−1)k−k1 . +Then, we have max{Ek +1−ν, +� +Ek−1} = +� +Ek−1. It then follows that: +∥zk − z∗∥ ≤ +� +Ek1 +� +(1 + ¯cE2ν−1 +k1 +)k−k1 +, +(c) Let ν ∈ (1/2, 1), we have: +¯c ≤ (Ek−1 − Ek)Ek +−2ν +∀k > k1. +(35) +Let h : R+ → R be defined as h(s) = s−2ν for any s ∈ R+. It is clear that +h is monotonically decreasing and ∀s ∈ R+, h′(s) = −2νs−(1+2ν) < 0. +Since Ek ≤ Ek−1 for all k > k1, then h(Ek−1) ≤ h(Ek) for all k > k1. +We consider two cases: + +20 +L. El Bourkhissi, I. Necoara +Case 1: Let r0 ∈ (1, +∞) such that: h(Ek) ≤ r0h(Ek−1) k > k1. Then, +from (35) we get: +¯c ≤ r0(Ek−1 − Ek)h(Ek−1) ≤ r0h(Ek−1) +� Ek−1 +Ek +1 ds +≤ r0 +� Ek−1 +Ek +h(s) ds = r0 +� Ek−1 +Ek +s−2ν ds = +r0 +1 − 2ν (Ek−1 +1−2ν − Ek +1−2ν). +Since ν > 1 +2, it follows that: +0 < ¯c(2ν − 1) +r0 +≤ Ek +1−2ν − Ek−1 +1−2ν. +Let us define ˆc = ¯c(2ν−1) +r0 +and ˆν = 1 − 2ν < 0. We then get: +0 < ˆc ≤ Ek +ˆν − Ek−1 +ˆν +∀k > k1. +(36) +Case 2: Let r0 ∈ (1, +∞) such that: h(Ek) > r0h(Ek−1), k > k1. We +then have Ek +−2ν ≥ r0Ek−1 +−2ν. This leads to +qEk−1 ≥ Ek, +where q = r0− 1 +2ν ∈ (0, 1). Since ˆν = 1 − 2ν < 0 we have qˆνEk−1 +ˆν ≤ Ek +ˆν +and then, it follows that: +(qˆν − 1)Ek−1 +ˆν ≤ Ek−1 +ˆν − Ek +ˆν. +Since qˆν − 1 > 0 and Ek → 0+ as k → ∞, there exists ˜c such that +(qˆν − 1)Ek−1 +ˆν ≥ ˜c for all k > k1. Therefore, we obtain: +0 < ˜c ≤ Ek +ˆν − Ek−1 +ˆν +∀k > k1. +(37) +By choosing µ = min{ˆc, ˜c} > 0, one can combine (36) and (37) to +obtain +0 < µ ≤ Ek +ˆν − Ek−1 +ˆν +∀k > k1. +Summing the above inequality from k1 + 1 to some k > k1 gives +µ(k − k1) + Ek1 +ˆν ≤ Ek +ˆν. +Hence, +Ek ≤ (µ(k − k1) + Ek1 +ˆν) +1 +ˆν = (µ(k − k1) + Ek1 +1−2ν) +1 +1−2ν . +Since ν ∈ ( 1 +2, 1), then max{Ek−1 +1−ν, +� +Ek−1} = Ek−1 +1−ν. Then, (34) +becomes: +∥zk − z∗∥ ≤ +� +1 +µ(k − k1) + E1−2ν +k1 +� 1−ν +2ν−1 +, +∀k > k1. +This concludes our proof. +⊓⊔ + +Linearized augmented Lagrangian for nonconvex optimization +21 +5 Numerical results +In this section we numerically compare Algorithm 1 (Linearized AL) with SCP +algorithm [20] and IPOPT [28] on quadratic problems with quadratic equal- +ity constraints (QPQCs). Note that Proximal Augmented Lagrangian method +presented in [29] is very slow compared to these three methods, hence we de- +cided not to include it in our comparisons. The simulations are implemented in +MATLAB and executed on a PC with (CPU 2.70GHz, 16GB RAM). For the +implementation of our method, we choose βk constant and equal to 1 and the +penalty parameter ρ = 103 if not specified differently. Since one cannot guar- +antee that the SCP iterates converge to a KKT point, we choose the following +stopping criteria: we stop the algorithms when the difference between two con- +secutive values of the objective function is less than a tolerance ϵ1 = 10−3 and +the norm of constraints is less than a tolerance ϵ2 = 10−5. The numerical +results are illustrated in Table 1 and Figure 1. +In Table 1, we compare the number of iterations, cpu time (sec), objective value +and feasibility violation for Linearized AL, SCP and IPOPT on QPQCs from +the CUTEst collection (the first part of the table) and randomly generated +QCQP (the last 10 test cases in Table 1). Note that for the test cases from +CUTEst our algorithm, Linearized AL, is the best in terms of cpu time. The +”-” indicates that an algorithm is unable to solve the corresponding problem +in less than two hours. As one can see from Table 1, IPOPT could not find +solutions of some random problems during the fixed time. Note that the last +five random cases (in the second part of the table) have sparse data, while the +first five random problems have dense data. This explains why the algorithms +are faster for the sparse problems compared to the first five ones in this second +part of the table, even though the latter have much smaller dimensions. Table +1 shows that our method is faster than SCP and IPOPT (in average it is +4 times faster than SCP and 10 times faster than IPOPT). For the optimal +value function, we can see that sometimes Linearized AL and SCP methods +obtain a worse optimal value than IPOPT. However, in our case by increasing +the penalty parameter ρ, we can also find the same optimal value as IPOPT, +whereas this is not the case for SCP. +Figure 1 shows the performance profiling for computation time (left) and num- +ber of iterations (right) for the 3 algorithms. In the performance profiling, the +vertical axis P(rp,s ≤ τ) (P(rp,s ≤ k)) represent the proportion of problems in +the numerical experiments for which rp,s does not exceed τ (k), respectively, +where rp,s is the ratio of the computational time (number of iterations) that the +s solver takes to solve problem p to the shortest computational time (number +of iterations) among the three algorithms to solve problem p, respectively. It +is clear from the computational time profile, Figure 1 (left), that the proposed +algorithm, Liniarized AL, approaches 1 faster than SCP and IPOPT. However, +for the number of iterations this is not always the case. From these prelim- +inary experiments we can conclude that our new algorithm, Linearized AL, +is an efficient method to solve optimization problems with nonlinear equality + +22 +L. El Bourkhissi, I. Necoara +XXXXXXXX +(n,m) +Alg +Linearized AL +SCP +IPOPT +# iter +cpu +# iter +cpu +# iter +cpu +f∗ +∥F∥ +f∗ +∥F∥ +f∗ +∥F∥ +OPTCTRL3 +5 +0.12 +5 +0.17 +7 +7.40 +(119,80) +2048.01 +5.42e-14 +2048.01 +4.18e-12 +2048.01 +1,84e-08 +OPTCTRL3; ρ = 103 +15 +1.38 +16 +3.34 +10 +11.99 +(1199,800) +19877.75 +1.10e-08 +19877.75 +1.56e-09 +18460.22 +6.33e-09 +OPTCTRL3; ρ = 7 × 105 +87 +9.44 +16 +3.34 +10 +11.99 +(1199,800) +18460.23 +4.09e-07 +19877.75 +1.56e-09 +18460.22 +6.33e-09 +OPTCTRL3 +44 +15.44 +44 +209.41 +11 +26.95 +(4499,3000) +74465.03 +8.43e-09 +74465.03 +2.82e-09 +74465.03 +1.09e-08 +DTOC4 +3 +0.05 +3 +0.24 +3 +7.61 +(299,198) +2.95 +3.65e-07 +2.95 +2.75e-07 +2.94 +1.02e-08 +DTOC4 +3 +0.35 +3 +0.91 +3 +12.45 +(1497,998) +2.88 +4.26e-07 +2.88 +7.85e-09 +2.88 +1.20e-08 +DTOC4 +3 +4.35 +3 +16.74 +3 +29.02 +(4497,2998) +2.87 +2.06e-07 +2.87 +4.87e-10 +2.87 +3.66e-08 +DTOC5; ρ = 103 +4 +0.33 +4 +0.51 +3 +12.06 +(998,499) +1.61 +1.67e-07 +1.61 +6.75e-08 +1.53 +7.76e-07 +DTOC5; ρ = 2 × 104 +120 +7.63 +4 +0.51 +3 +12.06 +(998,499) +1.54 +5.79e-07 +1.61 +6.75e-08 +1.53 +7.76e-07 +DTOC5 +5 +53.05 +4 +138.74 +3 +75.25 +(9998,4999) +1.62 +3.92e-08 +1.62 +1.68e-10 +1.53 +2.49e-07 +ORTHREGA +37 +0.91 +39 +1.73 +76 +10.14 +(517,256) +1414.05 +1.23e-06 +1664.80 +1.24e-06 +1414.05 +6.19e-10 +ORTHREGA +53 +13.27 +67 +31.78 +14 +23.99 +(2053,1024) +5661.43 +7.90e-07 +6654.78 +2.07e-06 +5661.43 +9,25e-07 +ORTHREGA +38 +65.72 +193 +3798.87 +20 +71.78 +(8197,4096) +22647.84 +1.83e-07 +22647.84 +3.12e-06 +22647.84 +1.86e-09 +(10,9) +27 +0.07 +34 +0.14 +7 +0.21 +0.21 +9.49e-06 +0.21 +1.05e-15 +0.21 +2.17e-10 +(20,13) +15 +0.12 +15 +0.39 +13 +0.65 +-2.61 +6.01e-06 +-2.61 +6.01e-06 +-2.61 +3.85e-15 +(50,43) +62 +3.90 +61 +4.68 +12 +87.75 +-12.95 +8.43e-06 +-12.95 +9.86e-06 +-1.98 +2.92e-14 +(100,91) +629 +22.53 +204 +6.42 +21 +143.74e +-742.21 +4.16e-06 +-300.72 +9.70e-06 +-113.52 +5.44e-12 +(150,140) +1165 +75.53 +505 +24.89 +37 +417.87 +-1027.24 +1.51e-06 +-358.67 +9.79e-06 +-231.26 +3.382-11 +(1000,500) +17 +1.37 +25 +1.76 +5 +140.34 +-0.27 +3.13e-06 +-0.27 +1.76e-08 +-0.27 +1.36e-10 +(10000,500) +7 +3.67 +7 +11.64 +- +- +1.46 +9.48e-07 +1.46 +3.40e-07 +- +- +(100000,500) +7 +34.12 +7 +73.43 +- +- +2.47 +6.59e-06 +2.47 +5.87e-06 +- +- +(100000,1000) +7 +76.21 +7 +134.07 +- +- +33.81 +1.59e-06 +33.81 +1.54e-09 +- +- +(300000,700) +7 +164.54 +7 +178.35 +- +- +33.59 +2.55e-06 +33.59 +2.44e-07 +- +- +Table 1 Comparison between Linearized AL, SCP and IPOPT on QCQPs from CUTEst +(top) and randomly generated QCQPs (bottom). + +Linearized augmented Lagrangian for nonconvex optimization +23 +Fig. 1 Performance profiles for the computation time (left) and the number of iterations +(right). +constraints, usually much faster than well established solvers such as IPOPT +and SCP. +6 Conclusion +In this paper, we have proposed a linearized augmented Lagrangian method for +solving (locally) smooth optimization problems with nonlinear equality con- +straints. In this method we have linearized the functional constraints within +the augmented Lagrangian function and added a regularization term. By dy- +namically generating the regularization (proximal) parameter, we have proved +global asymptotic convergence, convergence rate to an ϵ first-order optimal +solution and improved convergence rates under the K�L condition. Moreover, +we have numerically shown that the proposed algorithm is efficient, comparing +it with two known algorithms, SCP and IPOPT. +Conflict of interest +The authors declare that they have no conflict of interest. +Data availability +It is not applicable. + +0.8 +0.8 +0.6 +K. +0.6 +VI +s'd +s'd +P0.4 +0.4 +0.2 +0.2 +- Linearized AL +Linearized AL +- SCP +- SCP +---- IPOPT +----IPOPT +- +0 +5 +10 +15 +20 +2 +4 +6 +8 +10 +T: cpu time (s) +k: iterations24 +L. El Bourkhissi, I. Necoara +Appendix +Proof of Lemma 2 Using the definition of ψ and ∇xψ, we have: +∥∇xψ(x, λ) − ∇xψ(x′, λ′)∥ += ∥∇F(x)T (λ + ρF(x)) − ∇F(x′)T (λ′ + ρF(x′))∥ += +���� +� +∇F(x) − ∇F(x′) +�T � +λ + ρF(x) +� ++ ∇F(x′)T � +λ − λ′ + ρ +� +F(x) − F(x′) +������ +≤ ∥∇F(x) − ∇F(x′)∥∥λ + ρF(x)∥ + ∥∇F(x′)∥ +� +∥λ − λ′∥ + ρ∥F(x) − F(x′)∥ +� +Ass. 2 +≤ +� +LF ∥λ + ρF(x)∥ + ρM2 +F +� +∥x − x′∥ + MF ∥λ − λ′∥ +≤ +� +LF ∥λ + ρF(x)∥ + MF (1 + ρMF ) +� ���� +�x +λ +� +− +�x′ +λ′ +����� +≤ sup +x,λ +{Lψ(x, λ)} +���� +�x +λ +� +− +�x′ +λ′ +����� , +where Lψ(x, λ) = LF ∥λ + ρF(x)∥ + MF (1 + ρMF ). Similarly, using the expression of ∇λψ, +we get: +∥∇λψ(x, λ) − ∇λψ(x′, λ′)∥ = ∥F(x) − F(x′)∥ +Ass. 2 +≤ +MF ∥x − x′∥ ≤ MF +���� +�x +λ +� +− +�x′ +λ′ +����� . +Then, using basic properties of the Euclidean norm, we obtain: +∥∇ψ(x, λ) − ∇ψ(x′, λ′)∥ +≤ ∥∇xψ(x, λ) − ∇xψ(x′, λ′)∥ + ∥∇λψ(x, λ) − ∇λψ(x′, λ′)∥ +≤ +sup +(x,λ)∈S×Λ +� +LF ∥λ + ρF(x)∥ + MF (2 + ρMF ) +� ���� +�x − x′ +λ − λ′ +����� . +This concludes our proof. +⊓⊔ +Proof of Lemma 3 Using the optimality of xk+1, we have: +¯Lρ(xk+1, λk; xk) + βk+1 +2 +∥xk+1 − xk∥2 ≤ ¯Lρ(xk, λk; xk) = Lρ(xk, λk). +Further, from definitions of ¯Lρ and Lρ, we get: +f(xk+1) + ⟨λk , F(xk) + ∇F(xk)∆xk+1⟩ + ρ +2 ∥F(xk) + ∇F(xk)∆xk+1∥2 +≤ f(xk) + ⟨λk , F(xk)⟩ + ρ +2 ∥F(xk)∥2 − βk+1 +2 +∥∆xk+1∥2. +Rearranging this inequality, it follows: +f(xk+1) − f(xk) +≤ − ρ +2 ⟨∇F(xk)∆xk+1 , 2F(xk) + ∇F(xk)∆xk+1⟩ − ⟨∇F(xk)∆xk+1 , λk⟩ +− βk+1 +2 +∥∆xk+1∥2 +≤ − ρ +2 ∥∇F(xk)∆xk+1∥2 − ⟨∇F(xk)T (λk + ρF(xk)) , ∆xk+1⟩ − βk+1 +2 +∥∆xk+1∥2 +Ass. 2 +≤ +− ρ +2 σ2∥∆xk+1∥2 − βk+1 +2 +∥∆xk+1∥2 − ⟨∇F(xk)T (λk + ρF(xk)) , ∆xk+1⟩. + +Linearized augmented Lagrangian for nonconvex optimization +25 +Using the definitions of Lρ and ψ, we further obtain: +Lρ(xk+1, λk) − Lρ(xk, λk) = f(xk+1) − f(xk) + ψ(xk+1, λk) − ψ(xk, λk) +(3) +≤ − ρσ2 + αβk+1 +2 +∥xk+1 − xk∥2. +This proves our statement. +⊓⊔ +Proof of Lemma 4 Using the optimality condition for xk+1, we have: +∇f(xk+1) + ∇F(xk)T λk + ρ∇F(xk)T � +F(xk) + ∇F(xk)(xk+1 − xk) +� ++ βk+1(xk+1 − xk) = 0. +Combining this with the update in Step 6 of Algorithm 1, we get: +∇f(xk+1) + ∇F(xk)T λk+1 + βk+1(xk+1 − xk) = 0. +(38) +By replacing k with k − 1, we obtain: +∇f(xk) + ∇F(xk−1)T λk + βk(xk − xk−1) = 0. +(39) +Subtracting (39) from (38), we have: +∇f(xk+1) − ∇f(xk) + ∇F(xk)T ∆λk+1 + +� +∇F(xk) − ∇F(xk−1) +�T λk ++ βk+1∆xk+1 − βk∆xk = 0 +∀k ≥ 1. +Further, using Assumption 2, we have: +∥∆λk+1∥ ≤ 1 +σ +� +∥∇f(xk+1) − ∇f(xk)∥ + ∥∇F(xk) − ∇F(xk−1)∥∥λk∥ ++ βk+1∥∆xk+1∥ + βk∥∆xk∥ +� +∀k ≥ 1. +(40) +From (39), we further have: +∥λk∥ ≤ 1 +σ +� +∥∇f(xk)∥ + βk∥∆xk∥ +� +≤ 1 +σ +� +Mf + βk∥∆xk∥ +� +. +(41) +Moreover, from Assumption 2, we have: +∥∇F(xk) − ∇F(xk−1)∥ ≤ LF ∥∆xk∥ +and +∥∇F(xk) − ∇F(xk−1)∥ ≤ 2MF . +By replacing, the above inequalities and (41) in (40), we obtain: +∥∆λk+1∥ +≤ 1 +σ +� +Lf∥∆xk+1∥ + MfLF + 2MF βk +σ +∥∆xk∥ + βk+1∥∆xk+1∥ + βk∥∆xk∥ +� += Lf + βk+1 +σ +∥∆xk+1∥ + MfLF + (2MF + σ)βk +σ2 +∥∆xk∥. +(42) +Since (a + b)2 ≤ 2a2 + 2b2, we finally get (8). +⊓⊔ +Proof of Lemma 5 We prove this result using induction arguments. From Lemma 3 for +k = 0, we have the following: +f(x1) + ⟨λ0 , F(x1)⟩ + ρ +2 ∥F(x1)∥2 + ρσ2 + αβ1 +2 +∥x1 − x0∥2 +≤f(x0) + ⟨λ0 , F(x0)⟩ + ρ +2 ∥F(x0)∥2 +(18) +≤ ¯U + 1 +2ρ ∥λ0∥2 + c0. +(43) + +26 +L. El Bourkhissi, I. Necoara +For i = 0 and i = 1, we obtain: +f(xi) + ρ0 +2 ∥F(xi)∥2(ρ≥3ρ0) +≤ +f(xi) + ρ +6 ∥F(xi)∥2 +(18),(43) +≤ +¯U + 1 +2ρ ∥λ0∥2 + c0 − ⟨λ0 , F(xi)⟩ − ρ +3 ∥F(xi)∥2 +≤ ¯U + 1 +2ρ ∥λ0∥2 + c0 − ρ +3 ∥F(xi) + 3λ0 +2ρ ∥2 + 3∥λ0∥2 +4ρ +≤ ¯U + c0 + 5∥λ0∥2 +4ρ +(19) +≤ ¯U + c0 + 5∥λ0∥2 +4ρ ++ 3( ¯U + c0 − ¯L + 5∥λ0∥2 +4ρ +) +(ρ≥1) +≤ 4 ¯U + 4c0 − 3¯L + 5∥λ0∥2 ≤ ˆα. +Therefore, we find that x0, x1 ∈ S0 +ˆα. Moreover, using the optimality condition (39) for k = 1, +we have: +∇f(x1) + ∇F(x0)T λ1 + β1(x1 − x0) = 0. +Since x0, x1 ∈ S0 +ˆα and DS is the diameter of S0 +ˆα, then from Assumption 2: +∥λ1∥ ≤ 1 +σ (∥∇f(x1)∥ + β1∥∆x1∥) ≤ 1 +σ (Mf + βDS) ≤ 2(ρ − ρ0). +Furthermore, exploiting the definition of Pk for k = 1, we have: +P1 = Lρ(x1, λ1) + γ1 +2 ∥x1 − x0∥2 += Lρ(x1, λ1) − Lρ(x1, λ0) + Lρ(x1, λ0) − Lρ(x0, λ0) + Lρ(x0, λ0) ++ γ1 +2 ∥x1 − x0∥2 +Lemma 3 +≤ +⟨λ1 − λ0 , F(x1)⟩ − ρσ2 + αβ1 − γ1 +2 +∥x1 − x0∥2 + Lρ(x0, λ0) +(7) +≤ ρ +2 ∥F(x1)∥2 + 1 +2ρ ∥λ1 − λ0∥2 + Lρ(x0, λ0). +(44) +From (43), it follows that: +ρ +6 ∥F(x1)∥2 +≤ ¯U + 1 +2ρ ∥λ0∥2 + c0 − ρ +6 ∥F(x1)∥2 − ⟨λ0 , F(x1)⟩ − f(x1) − ρ +6 ∥F(x1)∥2 += ¯U + 1 +2ρ ∥λ0∥2 + c0 − ρ +6 ∥F(x1) + 3λ0 +ρ ∥2 + 3 +2ρ ∥λ0∥2 − f(x1) − ρ +6 ∥F(x1)∥2 +(ρ≥3ρ0) +≤ +¯U + 1 +2ρ ∥λ0∥2 + c0 + 3 +2ρ ∥λ0∥2 − f(x1) − ρ0 +2 ∥F(x1)∥2 +(2) +≤ ¯U + 2 +ρ ∥λ0∥2 + c0 − ¯L +(ρ≥1) +≤ +¯U + c0 − ¯L + 2∥λ0∥2. +(45) +Using (45) in (44), we obtain: +P1 ≤ 3( ¯U + c0 − ¯L + 2∥λ0∥2) + 1 +ρ ∥λ1∥2 + 1 +ρ ∥λ0∥2 + ¯U + c0 + 1 +2ρ ∥λ0∥2 +≤ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 2. + +Linearized augmented Lagrangian for nonconvex optimization +27 +It then follows that for k = 1, (20) is verified. Now, assume that (20) holds for some k ≥ 1 +(induction hypothesis) and we will prove that it continues to hold for k + 1. Using Lemma +3 together with the definition of Pk, we have: +Lρ(xk+1, λk) ≤ Lρ(xk, λk) ≤ Pk. +By using the expression of Lρ, we have that: +f(xk+1) + ⟨λk , F(xk+1)⟩ + ρ +2 ∥F(xk+1)∥2 ≤ Pk. +Thus, using (7), it follows that: +f(xk+1) − +∥λk∥2 +2(ρ − ρ0) − (ρ − ρ0)∥F(xk+1)∥2 +2 ++ ρ +2 ∥F(xk+1)∥2 ≤ Pk, +which yields the following: +f(xk+1) + ρ0 +2 ∥F(xk+1)∥2 ≤ Pk + +∥λk∥2 +2(ρ − ρ0) ≤Pk + 1 +≤4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 3 = ˆα, +where the last two inequalities are due to the induction hypothesis. Therefore, xk+1 ∈ S0 +ˆα. +Using the same arguments as for k = 1, the optimality condition (38) and the fact that +xk ∈ S0 +ˆα from the induction hypothesis, it follows that: +∥λk+1∥2 ≤ (Mf + βDS)2 +σ2 +≤ 2(ρ − ρ0). +Since xk, xk+1 ∈ S0 +ˆα and from (14), we have: +Pk+1 − Pk ≤ − γk+1 +4 +∥∆xk+1∥2 − γk +4 ∥∆xk∥2 ≤ 0. +Together with the induction hypothesis, we obtain: +Pk+1 ≤ Pk≤4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 2. +Finally, (20) is proved, which completes our proof. +⊓⊔ +Proof of Lemma 6 Using (5), we have: +Pk ≥ f(xk) + ρ +2 ∥F(xk)∥2 + ⟨λk , F(xk)⟩ +≥ f(xk) + ρ +2 ∥F(xk)∥2 − +∥λk∥2 +2(ρ − ρ0) − ρ − ρ0 +2 +∥F(xk)∥2 +(20b) +≥ f(xk) + ρ0 +2 ∥F(xk)∥2 − 1 +(Lemma 2) +≥ +¯L − 1. +It follows that the sequence {Pk}k≥1 is bounded from below. +⊓⊔ +Proof of Lemma 7 Using the optimality condition (38), we have: +∇f(xk+1) = −∇F(xk)T λk+1 − βk+1(xk+1 − xk). +It then follows, by exploiting the definition of Lρ and the properties of the derivative, that: +∇xLρ(xk+1, λk+1) = ∇f(xk+1) + ∇F(xk+1)T � +λk+1 + ρF(xk+1) +� += +� +∇F(xk+1) − ∇F(xk) +�T λk+1 + ∇F(xk+1)T ∆λk+1 − βk+1∆xk+1 ++ ρ∇F(xk+1)T � +F(xk+1) − F(xk) − ∇F(xk)∆xk+1 +� +. + +28 +L. El Bourkhissi, I. Necoara +Using basic properties of the Euclidean norm, we further get: +∥∇xLρ(xk+1, λk+1)∥ +≤∥∇F(xk+1) − ∇F(xk)∥∥λk+1∥ + ∥∇F(xk+1)∥∥∆λk+1∥ ++ βk+1∥∆xk+1∥ + ρ∥∇F(xk+1)∥∥F(xk+1) − F(xk) − ∇F(xk)∆xk+1∥ +Ass.2,(41) +≤ +MfLF + 2MF βk+1 +σ +∥∆xk+1∥ + MF ∥∆λk+1∥ ++ βk+1∥∆xk+1∥ + 2ρM2 +F ∥∆xk+1∥ += +� MfLF + (2MF + σ)βk+1 +σ ++ 2ρM2 +F +� +∥∆xk+1∥ + MF ∥∆λk+1∥ +(46) +(42) +≤ +� MfLF + MF Lf + (3MF + σ)βk+1 +σ ++ 2ρM2 +F +� +∥∆xk+1∥ ++ MF +σ +MfLF + (2MF + σ)βk +σ +∥∆xk∥ +≤ +� MF +σ +MfLF + MF Lf + (3MF + σ)βk+1 +σ ++ 2ρM2 +F +� +∥∆xk+1∥ ++ +� MF +σ +MfLF + MF Lf + (3MF + σ)βk +σ ++ 2ρM2 +F +� +∥∆xk∥ +Similarly, we have: +∥∇λLρ(xk+1, λk+1)∥ = ∥F(xk+1)∥ +≤∥F(xk+1) − F(xk) − ∇F(xk)∆xk+1∥ + 1 +ρ ∥∆λk+1∥ +Ass. 2 +≤ +2MF ∥∆xk+1∥ + 1 +ρ ∥∆λk+1∥ +(47) +(42) +≤ +� +2MF + 1 +ρ +Lf + βk+1 +σ +� +∥∆xk+1∥ + 1 +ρ +MfLF + (2MF + σ)βk +σ2 +∥∆xk∥ +≤ +� +2MF + 1 +ρ +Lfσ + MfLF + (2MF + σ)βk+1 +σ2 +� +∥∆xk+1∥ ++ +� +2MF + 1 +ρ +Lfσ + MfLF + (2MF + σ)βk +σ2 +� +∥∆xk∥ +where the first inequality is obtained from the multipliers update in Step 6 of Algorithm 1. +Hence, it follows that: +∥∇Lρ(xk+1, λk+1)∥ ≤ ∥∇xLρ(xk+1, λk+1)∥ + ∥∇λLρ(xk+1, λk+1)∥ +≤ Γ(βk+1)∥xk+1 − xk∥ + Γ(βk)∥xk − xk−1∥, +where +Γ(βk) = +� +MF + 1 +ρ +� MfLF + MF Lf + (3MF + σ)βk +σ2 ++ 2MF (ρMF + 1). +This proves our claim. +⊓⊔ +Proof of Lemma 8 By exploiting the definition of P(·) defined in (4), we have that for +any k ≥ 1: +∇xP(x, λ, y, γ) = ∇xLρ(x, λ) + γ(x − y), +∇λP(x, λ, y, γ) = ∇λLρ(x, λ) +∇yP(x, λ, y, γ) = γ(y − x) +and +∇γP(x, λ, y, γ) = 1 +2 ∥x − y∥2. + +Linearized augmented Lagrangian for nonconvex optimization +29 +Hence, +∥∇P(xk+1, λk+1, xk, γk+1)∥ ≤ ∥∇Lρ(xk+1, λk+1)∥ + +2γk+1∥xk+1 − xk∥ + 1 +2 ∥∆xk+1∥2 +≤(Γmax + DS + 2¯γ) (∥∆xk+1∥ + ∥∆xk∥) , +where the last inequality follows from Lemma 7, (22) and (25). This proves our claim. +⊓⊔ +Proof of Lemma 9 (i) From Lemma 5 and Assumption 4, it follows that {uk}k≥1 is +bounded and therefore, there exists a convergent subsequence {uk}k∈K such that limk∈K uk = +u∗. Hence Ω is nonempty. Moreover, Ω is compact since it is bounded and closed. On the +other hand, for any u∗ ∈ Ω, there exists a sequence of increasing integers K such that +limk∈K uk = u∗ and using Lemma 8 and (24), it follows that: +∥∇P(u∗)∥ = lim +k∈K ∥∇P(uk)∥ = 0. +Hence, u∗ ∈ crit P and 0 ≤ limk→∞ dist(uk, Ω) ≤ limk∈K dist(uk, Ω) = dist(u∗, Ω) = 0. +(ii) Since P(·) is continuous and {P(uk) = Pk}k≥1 converges to P ∗, then any subsequence +{P(uk) = Pk}k∈K that converges, it converges to the same limit P ∗. +(iii) Let (x, λ, y, γ) ∈ crit P that is ∇P(x, λ, y, γ) = 0. It then follows that: +∇xP(x, λ, y, γ) = ∇xLρ(x, λ) + γ(x − y) = 0, +∇λP(x, λ, y, γ) = ∇λLρ(x, λ) = 0 +∇yP(x, λ, y, γ) = γ(y − x) = 0 +and +∇γP(x, λ, y, γ) = 1 +2 ∥x − y∥2 = 0. +With some minor rearrangements, we obtain: +∇f(x) + ∇F(x)T λ = 0, +F(x) = 0. +Hence, (x, λ) is a stionary point of (1). This concludes our proof. +⊓⊔ +Proof of Lemma 10 From the boundedness of ∥∆λk+1∥2 derived in (8), we have: +∥∆λk+1∥2 ≤ c1(β)∥∆xk+1∥2 + c2(β)∥∆xk∥2 +≤ max{c1(β), c2(β)} +� +∥∆xk+1∥2 + ∥∆xk∥2� +. +(48) +Adding the term ∥∆xk+1∥2 + ∥∆xk∥2 on both sides in (48), we have: +∥zk+1 − zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2 +≤ ∥∆xk+1∥2 + ∥∆λk+1∥2 + ∥∆xk∥2 +(48) +≤ +� +max{c1(β), c2(β)} + 1 +�� +∥∆xk+1∥2 + ∥∆xk∥2� +. +(49) +Considering (25), we can then rewrite (14) as follows: +Pk+1 − Pk +(14) +≤ − ¯ +γ +4 +� +∥∆xk+1∥2 + ∥∆xk∥2� +(49) +≤ − +¯ +γ +4 +� +max{c1(β), c2(β)} + 1 +� ∥zk+1 − zk∥2. +(50) +Since Pk → P ∗ and {Pk}k≥1 is monotonically decreasing to P ∗, then it follows that the +error sequence {Ek}k≥1, is non-negative, monotonically decreasing and converges to 0. We +distinguish two cases. +Case 1: There exists k1 ≥ 1 such that Ek1 = 0. Then, Ek = 0 ∀k ≥ k1 and using (50), we +have: +∥zk+1 − zk∥2 ≤ 4 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +(Ek − Ek+1) = 0 +∀k ≥ k1. + +30 +L. El Bourkhissi, I. Necoara +From Lemma 5 the sequence {zk}k≥1 is bounded, and thus: +∞ +� +k=1 +∥∆xk∥ + ∥∆λk∥ = +k1 +� +k=1 +∥∆xk∥ + ∥∆λk∥<∞. +Case 2: The error Ek > 0 ∀k ≥ 1. Then, there exists k1 = k1(ϵ, τ) ≥ 1 such that ∀k ≥ k1 +we have dist(uk, Ω) ≤ ϵ, P ∗ < P(uk) < P ∗ + τ and +ϕ′(Ek)∥∇P(xk, λk, xk−1, γk)∥ ≥ 1, +(51) +where ϵ > 0, τ > 0 and ϕ ∈ Ψτ are well defined and correspond to those in Definition 2, recall +that P(·) satisfies the K�L property on Ω. Since ϕ is concave, we have ϕ(Ek) − ϕ(Ek+1) ≥ +ϕ′(Ek)(Ek − Ek+1). Then, from (50) and (51) we get: +∥zk+1 − zk∥2 +≤ ϕ′(Ek)∥zk+1 − zk∥2∥∇P(xk, λk, xk−1, γk)∥ +≤ 4 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +ϕ′(Ek)(Ek − Ek+1)∥∇P(xk, λk, xk−1, γk)∥ +≤ 4 +� +max{c1(β), c2(β)} + 1 +� +¯ +γ +� +ϕ(Ek) − ϕ(Ek+1) +� +∥∇P(xk, λk, xk−1, γk)∥. +Since ∥zk+1 − zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2. Using the fact that for any a, b, c, d ≥ 0, if +a2 + b2 ≤ c × d, then (a + b)2 ≤ 2a2 + 2b2 ≤ 2c × d ≤ c2 + d2 ≤ (c + d)2, it follows that for +any θ > 0, we have: +∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γ +� +ϕ(Ek) − ϕ(Ek+1) +� ++ 1 +θ ∥∇P(xk, λk, xk−1, γk)∥. +(52) +Furthermore, we have: +∥∇P(xk, λk, xk−1, γk)∥ ≤ ∥∇Lρ(xk, λk)∥ + +2¯γ∥xk − xk−1∥ +(46),(47) +≤ +(Γmax + DS + 2¯γ) (∥∆xk∥ + ∥∆λk∥) . +Then, (52) becomes: +∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γ +� +ϕ(Ek) − ϕ(Ek+1) +� ++ Γmax + DS + 2¯γ +θ +� +∥∆xk∥ + ∥∆λk∥ +� +. +Let us now choose θ > 0 so that 0 < Γmax+DS+2¯γ +θ +< 1 and define the parameter δ0 as: +δ0 = 1− Γmax+DS+2¯γ +θ +> 0. Then, by summing up the above inequality from k = ¯k to k = K +and using the property: �K +k=¯k ∥∆xk∥ = �K +k=¯k ∥∆xk+1∥ + ∥∆x +¯k∥ − ∥∆xK+1∥, we get: +K +� +k=¯k +∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γδ0 +� +ϕ(E +¯k) − ϕ(EK+1) +� ++ Γmax + DS + 2¯γ +θδ0 +� +∥∆x +¯k∥ + ∥∆λ +¯k∥ +� +− Γmax + DS + 2¯γ +θδ0 +� +∥∆xK+1∥ + ∥∆λK+1∥ +� +. + +Linearized augmented Lagrangian for nonconvex optimization +31 +Using the fact that {Ek}k≥k1 is monotonically decreasing and that the function ϕ is positive +and increasing, which yields ϕ(Ek) ≥ ϕ(Ek+1) > 0, then: +K +� +k=¯k +∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 +� +max{c1(β), c2(β)} + 1 +� +θ +¯ +γδ0 +ϕ(E +¯k) ++ Γmax + DS + 2¯γ +θδ0 +� +∥∆x +¯k∥ + ∥∆λ +¯k∥ +� +. +It is clear that the right-hand side of the above inequality is bounded for any K ≥ ¯k. Letting +K → ∞, we get that: +∞ +� +k=¯k +∥∆xk+1∥ + ∥∆λk+1∥ < ∞. +From Lemma 5, the sequence {(xk, λk)}k≥1 is bounded. Then, it follows that: +¯k +� +k=1 +∥∆xk∥ + ∥∆λk∥ < ∞. +Hence: �∞ +k=1 ∥∆xk∥ + ∥∆λk∥ < ∞. Let m, n ∈ Z+ such that n ≥ m, we have: +∥zn − zm∥ = ∥ +n−1 +� +k=m +∆zk+1∥ ≤ +n−1 +� +k=m +∥∆zk+1∥ ≤ +n−1 +� +k=m +∥∆xk+1∥ + ∥∆λk+1∥. +Since �∞ +k= ∥∆xk+1∥ + ∥∆λk+1∥ < ∞, it follows that ∀ε > 0, ∃N ∈ Z+ such that ∀m, n ∈ +Z+ where n ≥ m, we have: ∥zn − zm∥ ≤ ε. This implies that {zk}k≥1 is a Cauchy sequence +and converges. Moreover, by Theorem 1, {zk}k≥1 converges to a staionary point of (1). This +concludes our proof. +⊓⊔ +References +1. H. Attouch, J. Bolte and B. Svaiter, Convergence of descent methods for semi- +algebraic and tame problems: proximal algorithms, forward-backward splitting, and +regularized Gauss–Seidel methods, Math. Program. Ser. A 137, 91–129, 2013. +2. D. P. Bertsekas Constrained Optimization and Lagrange Multiplier Methods, Athena +Scientific, Belmont, MA 1996. +3. D. P. Bertsekas Convex Optimization Algorithms, Athena Scientific, Belmont, MA +2015. +4. D. P. Bertsekas, and J. N. Tsitsiklis, Parallel and Distributed Computation: Numer- +ical Methods, Athena Scientific, 2003. +5. J. Bolte, S. Sabach and M. Teboulle, Proximal alternating linearized minimization +for nonconvex and nonsmooth problems, Math. Program. 146(1–2), 459–494, 2014. +6. R. I. Bot and D. K. Nguyen, The proximal alternating direction method of multipliers +in the nonconvex setting: convergence analysis and rates, Math. Op. Res. 45(2), +682–712, 2020. +7. S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, Distributed optimization +and statistical learning via the alternating direction method of multipliers, Found. +Trends Mach. Learn. 3(1), 1–122, 2011. +8. E. Cohen, N. Hallak and M. Teboulle, A Dynamic Alternating Direction of Multi- +pliers for Nonconvex Minimization with Nonlinear Functional Equality Constraints, +Journal of Optimization Theory and Applications, 193, 324–353, 2022. +9. J. Fan, Comments on “wavelets in statistics: a review, by A. Antoniadis. J. Ital. +Stat. Soc. 6, 131–138, 1997. +10. J. A. Fessler, Optimization methods for magnetic resonance image reconstruction, +key models and optimization algorithms. IEEE Signal Process. Mag. 37(1), 33–40, +2020. + +32 +L. El Bourkhissi, I. Necoara +11. G. Gabay, and B. Mercier, A dual algorithm for the solution of nonlinear variational +problems via finite element approximation, Comput. Math. Appl. 2(1), 17–40, 1976. +12. R. Glowinski, and P. Le Tallec, Augmented Lagrangian and Operator-Splitting Meth- +ods in Nonlinear Mechanics, SIAM, vol. 9, 1989. +13. M. Grant and S. Boyd, CVX: Matlab Software for Disciplined Convex Programming, +version 2.1, 2014. [Online]. Available: http://cvxr.com/cvx,Mar.2014 +14. P. C. Haarhoff, J. D. Buys, A new method for the optimization of a nonlinear function +subject to nonlinear constraints, Computer Journal, 13, 178–184, 1970. +15. D. Hajinezhad and M. Hong, Perturbed proximal primal-dual algorithm for noncon- +vex nonsmooth optimization, Math. Program. 2019. +16. M. Hestenes, Multiplier and gradient methods, Journal of Optimization Theory and +Applications, 4, 303–320, 1969. +17. M. Hong, D. Hajinezhad, and M. M. Zhao, +Prox-PDA: The proximal primal-dual +algorithm for fast distributed nonconvex optimization and learning over networks. +In: D. Precup, Y.W. Teh (eds.) Proceedings of the 34th International Conference on +Machine Learning, Proceedings of Machine Learning Research, vol. 70, pp. 1529–1538, +PMLR 2017. +18. B. Jiang, T. Lin, S. Ma and S. Zhang, Structured nonconvex and nonsmooth opti- +mization: algorithms and iteration complexity analysis, Comput. Optim. Appl. 72(1), +115–157, 2019. +19. D. R. Luke, S. Sabach, and M. Teboulle, Optimization on spheres: models and prox- +imal algorithms with computational performance comparisons. SIAM J. Math. Data +Sci. 1(3), 408–445, 2019. +20. F. Messerer, K. Baumg¨artner and M. Diehl, Survey of sequential convex programming +and generalized Gauss-Newton methods, ESAIM: Proceedings and Surveys, 2021. +21. I. Necoara and S. Kvamme, DuQuad: A toolbox for solving convex quadratic pro- +grams using dual (augmented) first order algorithms, 2015 54th IEEE Conference on +Decision and Control (CDC), pp. 2043-2048, 2015. +22. M. J. D. Powell, A method for nonlinear optimization in minimization problems, in +Optimization (R. Fletcher, ed.), Academic Press, 283–298, 1969. +23. R. Rockafellar and R. Wets, Variational Analysis, Springer, Berlin, 1998. +24. J. O. Royset, Variational Analysis in Modern Statistics, Special Issue Mathematical +Programming, Series B, Volume 174, 2019. +25. C. W. Royer, M. O’Neill and S. J.Wright, A Newton-CG algorithm with complex- +ity guarantees for smooth unconstrained optimization, Mathematical Programming, +2019. +26. R. Shefi and M. Teboulle, Rate of convergence analysis of decompositionmethods +based on the proximal method of multipliers for convex minimization, SIAM J. Opt. +24(1), 269–297, 2014. +27. Q. Tran-Dinh and M. Diehl, Local convergence of sequential convex programming for +nonconvex optimization. In M. Diehl, F. Glineur, E. Jarlebring, and W. Michiels, +editors, Recent advances in optimization and its application in engineering, pages +93–103. Springer-Verl., 2010. +28. A. W¨achter and L. T. Biegler, On the Implementation of a Primal-Dual Interior +Point Filter Line Search Algorithm for Large-Scale Nonlinear Programming, Math- +ematical Programming 106(1), 25-57, 2006. +29. Y. Xie, S. J. Wright, Complexity of Proximal Augmented Lagrangian for Nonconvex +Optimization with Nonlinear Equality Constraints, Journal of Scientific Computing, +86, 38, 2021. +30. M. Yashtini, Convergence and rate analysis of a proximal linearized ADMM for +nonconvex nonsmooth optimization, Journal of Global Optimization, 2022. +31. C. Zhang, Nearly unbiased variable selection under minimax concave penalty, Ann. +Stat. 38, 894–942, 2010. +32. J. Zhang and Z. Q. Luo, A proximal alternating direction method of multiplier for lin- +early constrained nonconvex minimization, SIAM J. Optim. 30(3), 2272–2302, 2020. + diff --git a/g9E_T4oBgHgl3EQf3Rw9/content/tmp_files/load_file.txt b/g9E_T4oBgHgl3EQf3Rw9/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..027cd9cece4f8a149e775fa169a3b5d286a85e72 --- /dev/null +++ b/g9E_T4oBgHgl3EQf3Rw9/content/tmp_files/load_file.txt @@ -0,0 +1,929 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf,len=928 +page_content='Noname manuscript No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (will be inserted by the editor) Complexity of linearized augmented Lagrangian for optimization with nonlinear equality constraints Lahcen El Bourkhissi · Ion Necoara Received: date / Accepted: date Abstract In this paper, we consider a nonconvex optimization problem with nonlinear equality constraints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We assume that both, the objective function and the functional constraints are locally smooth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' For solving this problem, we propose a linearized augmented Lagrangian method, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', we linearize the functional constraints in the augmented Lagrangian at the current iterate and add a quadratic regularization, yielding a subproblem that is easy to solve, and whose solution is the next iterate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Under a dynamic regularization parameter choice, we prove global asymptotic convergence of the iterates to a critical point of the problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We also derive convergence guarantees for the iterates of our method to an ϵ first-order optimal solution in O(1/ϵ2) outer iterations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Finally, we show that, when the problem data are e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', semialgebraic, the sequence generated by our algorithm converges and we derive convergence rates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We validate the theory and the performance of the proposed algorithm by numerically comparing it with the existing methods from the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Keywords Nonconvex optimization · linearized augmented Lagrangian · nonlinear functional constraints · convergence analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The research leading to these results has received funding from: the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 953348;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' NO Grants 2014-2021, RO-NO- 2019-0184, under project ELO-Hyp, contract no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 24/2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' UEFISCDI, PN-III-P4-PCE-2021-0720, under project L2O-MOC, nr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 70/2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lahcen El Bourkhissi Department of Automatic Control and Systems Engineering, University Politehnica Bucharest, Bucharest, 060042, Romania E-mail: lel@stud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='upb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='ro Ion Necoara Department of Automatic Control and Systems Engineering, University Politehnica Bucharest, Bucharest, 060042, Romania.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Gheorghe Mihoc-Caius Iacob Institute of Mathematical Statistics and Applied Mathematics of the Romanian Academy, 050711 Bucharest, Romania.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' E-mail: ion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='necoara@upb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='ro arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='08345v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='OC] 19 Jan 2023 2 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Mathematics Subject Classification (2020) 68Q25 · 90C06 · 90C30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 1 Introduction In many fields, such as machine learning, matrix optimization, statistics, con- trol and signal processing, one finds applications that can be recast as noncon- vex optimization problems with nonlinear functional equality constraints, see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', [10,17,19,24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In this paper, we solve this optimization problem by means of an augmented Lagrangian approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The augmented Lagrangian method, also known as the method of multipliers, was initially proposed in [14,16,22] to minimize objective functions subject to (linear) equality constraints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It pro- vides many theoretical advantages, even for non-convex problems, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', no du- ality gap and exact penalty representation, see [23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, the augmented Lagrangian framework is at the heart of the Alternating Direction Method of Multipliers (ADMM), a very efficient method for large-scale (distributed) optimization [5–8,11,12,30].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The augmented Lagrangian approach has been extensively studied in the lit- erature for convex problems, see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', [2–4,7,12,26], and recently it has been extended to non-convex (smooth/non-smooth) problems with linear equal- ity constraints in [15, 17, 18, 30, 32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' However, there are very few studies on the use of the augmented Lagrangian framework for nonconvex optimization with nonlinear equality constraints, see [8,29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In [29] a Proximal Augmented Lagrangian (Proximal AL) is proposed to solve nonconvex, but smooth op- timization problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In Proximal AL a static regularization (proximal) term is added to the original augmented Lagrangian function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The authors prove that when an approximate first- (second)-order solution of the subproblem is found, with an error approaching asymptotically zero, then an ϵ− first (ϵ− second)-order solution of the original problem is obtained within O(1/ϵ2−η) outer iterations, where η ∈ [0, 2] is an user-defined parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that when η is close to 2, the complexity reduces to O(1) outer iterations, but the sub- problem, which is already nonconvex due to the nonlinearity of the constraints, gets ill-conditioned as the penalty parameter in the augmented Lagrangian is inversely proportional to ϵη.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In addition, the total iteration complexity is de- rived when the Newton conjugate gradient method from [25] is used to solve the subproblem at each outer iteration of Proximal AL algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In [30], a PL-ADMM was proposed to solve nonsmooth nonconvex problems with linear equality constraints for two blocks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In this setting, the problem is smooth in one block and nonsmooth in the other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' PL-ADMM algorithm uses a dynamic metric proximal term in the primal update and an overrelaxation step in the dual update.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It is proved in [30] that the sequence generated by this algo- rithm is bounded and each limit point of this sequence is a stationary point of the original problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Additionally, it is shown that under the Kurdyka- �Lojasiewicz (K�L) property, the error converges either in a finite number of iterations, sublinearly or linearly, depending on the exponent associated with Linearized augmented Lagrangian for nonconvex optimization 3 the K�L property.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Similar convergence results are obtained in [8] for an al- gorithm solving problems having nonlinear constraints in the first block (for which the corresponding objective function is smooth) and linear constraints in the second block.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The algorithm proposed in [8], called DAM (Dynamic linearized Alternating direction method of Multipliers), linearizes the smooth part of the augmented Lagrangian and adds a proximal regularization, where the proximal parameter is dynamically generated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' A different approach, that does not rely on the augmented Lagrangian frame- work, is presented in [20, 27] and is called Sequential Convex Programming (SCP).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This method solves a sequence of convex approximations of the orig- inal problem by linearizing the nonconvex parts of the objective and of the functional constraints and preserving the structures that can be exploited by convex optimization techniques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In this case the subproblem has a (strongly) convex objective and linear constraints, for which efficient solution methods exist, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', [13,21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' However, to the best of our knowledge, SCP methods con- verge under mild assumptions only locally [20,27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Due to the high cost of solving the nonconvex subproblem in Proximal AL [29] and the fact that SCP type schems [20,27] have only local convergence guar- antees, in this paper we propose a novel Linearized Augmented Lagrangian method (called Linearized AL) for solving smooth nonconvex problems with nonlinear equality constraints, that overcomes these limitations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' More pre- cisely, in our algorithm we linearize the functional constraints in the aug- mented Lagrangian at the current iterate and add a dynamic regularization term, yielding a subproblem that is easy to solve, and whose solution is the next iterate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Indeed, if the objective function is (weakly) convex, our subproblem is strongly convex and thus can be solved efficiently.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, our method borrows the advantages of both approaches, Proximal AL and SCP, since it converges globally and the subproblem is strongly convex and thus easy to solve.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We prove that the sequence generated by our algorithm is bounded, each limit point of this sequence is a stationary point and under the K�L property we prove convergence for the whole sequence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' More precisely, under a dynamic regularization parameter choice, we prove global asymptotic convergence of the iterates to a critical point of the problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We also derive convergence guarantees for the iterates of our method to an ϵ first-order optimal solution in O(1/ϵ2) outer iterations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Finally, we show that under the K�L property the whole sequence generated by the proposed algorithm converges and we derive convergence rates depending on the KL parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Unlike [29], in our method the penalty parameter is independent on the required accuracy ϵ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, in comparison with [20], our convergence results here are global, while only local convergence has been proved for SCP.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The preliminary numerical results confirm the efficiency of our Linearized AL algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The paper is structured as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In Section 2, we introduce our problem of interest and some notions necessary for our analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In Section 3, we present our algorithm, followed in Section 4 by its convergence analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Finally, in Section 5, we compare numerically our method with existing algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 4 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara 2 Problem formulation and preliminaries In this paper, we consider the following nonlinear optimization problem: min x∈Rn f(x) s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' F(x) = 0, (1) where f : Rn → R and F(x) ≜ (f1(x), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', fm(x))T , with fi : Rn → R for all i = 1 : m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We assume the functions f, fi ∈ C2 for all i = 1 : m, f is (weakly) convex and F is nonlinear.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' For this problem we consider the following assumptions: Assumption 1 Assume that there exists ρ0 ≥ 0 such that f(x) + ρ0 2 ∥F(x)∥2 has compact level sets, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', for all α ∈ R, the following set is empty or compact: S0 α ≜ {x : f(x) + ρ0 2 ∥F(x)∥2 ≤ α}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Assumption 2 Given a compact set S ⊆ Rn, there exist positive constants Mf, MF , σ, Lf, LF such that f and F satisfy the following conditions: (i) ∥∇f(x)∥ ≤ Mf, ∥∇f(x) − ∇f(y)∥ ≤ Lf∥x − y∥ for all x, y ∈ S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (ii) ∥∇F(x)∥ ≤ MF , σmin(∇F(x)) ≥ σ > 0 for all x, y ∈ S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (iii) ∥∇F(x) − ∇F(y)∥ ≤ LF ∥x − y∥ for all x, y ∈ S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Assumption 3 There exists finite ¯U such that f(x) ≤ ¯U for all x ∈ {x : ∥F(x)∥ ≤ 1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that these assumptions are standard in the nonconvex optimization lit- erature, see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', [8,29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In fact, these assumptions are not restrictive because they need to hold only locally.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Indeed, large classes of problems satisfy these assumptions as discussed below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Remark 1 Assumption 1 holds e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', when f(x) + ρ0∥F(x)∥2 is coercive for some ρ ≥ 0, or when f(x) is strongly convex or f(x) is bounded bellow and F(x) = xT x−1, as is the case in dictionary learning applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It also holds when f(x) = 1 2xT Qx − pT x, F(x) = Ax − b and Q is a positive definite matrix on null(A) := {x : Ax = 0}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Remark 2 Assumption 2 allows general classes of problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In particular, con- ditions (i) hold if f(x) is differentiable and ∇f(x) is locally Lipschitz continu- ous on a neighborhood of S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Conditions (ii) hold when F(x) is differentiable on a neighborhood of S and satisfies an LICQ condition over S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Finally, condition (iii) holds if ∇F(x) is locally Lipschitz continuous on S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Remark 3 For Assumption 3 to hold, it is sufficient the set {x : ∥F(x)∥ ≤ 1} to be compact.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In fact, we do not need this assumption if we can choose the starting point x0 such that F(x0) = 0, that is, the initial point is feasible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The following lemma is an immediate consequence of Assumption 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 5 Lemma 1 If Assumption 1 holds, then f(x) + ρ0 2 ∥F(x)∥2 is lower bounded: ¯L ≜ inf x∈Rn{f(x) + ρ0 2 ∥F(x)∥2} > −∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (2) Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Further, let us introduce the following definitions (see also [29]): Definition 1 [First-order solution and ϵ first-order solution of (1)] The vector x∗ is said to be first-order solution of (1) if ∃λ∗ ∈ Rm such that: ∇f(x∗) + ∇F(x∗)T λ∗ = 0 and F(x∗) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, x∗ is an ϵ first-order solution of (1) if ∃λ∗ ∈ Rm such that: ∥∇f(x∗) + ∇F(∗x)T λ∗∥ ≤ ϵ and ∥F(x∗)∥ ≤ ϵ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let Φ : Rd → ¯R be a proper lower semicontinuous function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' For −∞ < τ1 < τ2 ≤ +∞, we define [τ1 < Φ < τ2] = {x ∈ Rd : τ1 < Φ(x) < τ2}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let τ ∈ (0, +∞].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We denote by Ψτ the set of all continuous concave functions ϕ : [0, τ] → [0, +∞) such that ϕ(0) = 0 and ϕ is continuously differentiable on (0, τ), with ϕ′(s) > 0 over (0, τ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Definition 2 Let Φ : Rd → ¯R be a proper lower semicontinuous function that takes constant value on Ω.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We say that Φ satisfies the K�L property on Ω if there exists ϵ > 0, τ > 0, and ϕ ∈ Ψτ such that for every x∗ ∈ Ω and every element x in the intersection {x ∈ Rd : dist(x, Ω) < ϵ} ∩ [Ψ(x∗) < Ψ(x) < Ψ(x∗) + τ], we have: ϕ′� Φ(x) − Φ(x∗) � dist � 0, ∂Φ(x) � ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This definition covers many classes of functions arising in practical optimiza- tion problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' For example, if f is a proper closed semialgebraic function, then f is a K�L function with exponent ν ∈ [0, 1) [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The function g(Ax), where g is strongly convex on a compact set and twice differentiable, and A ∈ Rm×n, is a K�L function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Convex piecewise linear quadratic functions such as ∥x∥1, ∥x∥0, γ �k i=1 |x[i]|, where |x[i]| is the ith largest entry in x, k ≤ n and γ ∈ (0, 1];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' the indicator function δ∆(x), where ∆ = {x ∈ Rn : eT x = 1, x ≥ 0};' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' least-squares problems with the Smoothly Clipped Absolute Devia- tion (SCAD) [9];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Minimax Concave Penalty (MCP) regularized functions [31] are all K�L functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The K�L property characterizes the local geometry of a function around the set of critical points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 6 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara 3 A linearized augmented Lagrangian method In this section, we propose a new algorithm for solving problem (1) using the augmented Lagrangian framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us first introduce few notations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The augmented Lagrangian associated with the problem (1) is: Lρ(x, λ) = f(x) + ⟨λ , F(x)⟩ + ρ 2∥F(x)∥2 = f(x) + ψ(x, λ), where ψ(x, λ) = ⟨λ , F(x)⟩ + ρ 2∥F(x)∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The gradient of ψ is: ∇xψ(x, λ) = ∇F(x)T (λ + ρF(x)) and ∇λψ(x, λ) = F(x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In the rest of this paper, for the sake of clarity, we provide the proofs of all the lemmas in Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The following lemma shows that ψ is locally smooth, (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', it has Lipschitz continuous gradient locally).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 2 [Smoothness of ψ] If Assumption 2 holds, then for any compact set Λ ⊂ Rm there exists Lψ > 0 such that: ∥∇ψ(x, λ) − ∇ψ(x′, λ′)∥ ≤ Lψ ���� �x − x′ λ − λ′ ����� ∀(x, λ), (x′, λ′) ∈ S × Λ, where Lψ = sup(x,λ)∈S×Λ{LF ∥λ + ρF(x)∥ + MF (2 + ρMF )}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Further, let us denote the following function derived from linearization of the functional constraints, at a given point ¯x, in the augmented Lagrangian: ¯Lρ(x, λ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ¯x) = f(x) + ⟨λ , F(¯x) + ∇F(¯x)(x − ¯x)⟩ + ρ 2∥F(¯x) + ∇F(¯x)(x − ¯x)∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' To solve the optimization problem (1) we propose the following Linearized augmented Lagrangian (Linearized AL) algorithm, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', we linearize the func- tional constraints in the augmented Lagrangian at the current iterate and add a quadratic regularization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Algorithm 1 Linearized augmented Lagrangian 1: Initialization: x0, λ0, and ρ > 0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2: k ← 0 3: while stopping criterion is not satisfied do 4: generate a proximal parameter βk+1 > 0 5: xk+1 ← arg minx ¯Lρ(x, λk;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' xk) + βk+1 2 ∥x − xk∥2 ▷ The subproblem 6: λk+1 ← λk + ρ � F(xk) + ∇F(xk)(xk+1 − xk) � 7: k ← k + 1 8: end while To the best of our knowledge Linearized AL algorithm is new and its conver- gence behaviour has not been analyzed before in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that if Linearized augmented Lagrangian for nonconvex optimization 7 f is (weakly) convex function, the objective function in the subproblem of step 5 of Algorithm 1 is strongly convex, provided that βk+1 is chosen ad- equately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In particular, if f is a quadratic function, then finding a solution of the subproblem in step 5 is even equivalent with solving a linear system of equalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In these cases, efficient solution methods exist for solving the subproblem, see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', [13, 21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It is also important to note that our update of the dual multipliers is different from the literature, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', instead of eval- uating the functional constraints at the new test point xk+1 and updating λk+1 = λk + ρF(xk+1) [8,29], we evaluate their linearization at xk in the new point xk+1 and update λk+1 = λk + ρ(F(xk) + ∇F(xk)(xk+1 − xk)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In the sequel, we denote: ∆xk = xk − xk−1 and ∆λk = λk − λk−1 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let α ∈ (0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, from Lemma 2 we can choose βk+1 such that: ψ(xk+1, λk) − ψ(xk, λk) − ⟨∇xψ(xk, λk) , xk+1 − xk⟩ ≤ (1 − α)βk+1 2 ∥xk+1 − xk∥2 ∀k ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (3) Note that for any k ≥ 1, βk is well defined since ψ is smooth, see Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This regularization parameter βk can be found using e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', a backtracking scheme as in [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In the sequel, we assume that βk satisfies the following: Assumption 4 [Proximal parameter sequence is bounded] Assume that the generated parameter sequence {βk}k≥1 is bounded, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', β := sup k≥1 βk < ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 4 Convergence analysis In this section, we derive the asymptotic convergence of Linearized AL algo- rithm (Algorithm 1) and its efficiency to obtain an ϵ first-order solution for problem (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, we provide convergence rates under the K�L condition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that for some intermediate steps in the convergence analysis we follow a similar methodology as in [29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us start by proving the decrease with respect to the first argument for the augmented Lagrangian function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 3 [Descent of Lρ w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' primal variables] If Assumption 2 holds, then we have the following descent: Lρ(xk+1, λk) ≤ Lρ(xk, λk) − ρσ2 + αβk+1 2 ∥xk+1 − xk∥2 ∀k ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 8 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Let us define the following Lyapunov function (inspired from [29]): P(x, λ, y, γ) = Lρ(x, λ) + γ 2 ∥x − y∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (4) The evaluation of the Lyapunov function along the iterates of Linearized AL algorithm is denoted by: Pk = P(xk, λk, xk−1, γk) ∀k ≥ 1, (5) with γk > 0 to be defined later.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We prove next that {Pk}k≥1 is decreasing and bounded from bellow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us first prove that it is decreasing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Indeed: P k+1 − Pk =Lρ(xk+1, λk+1) − Lρ(xk+1, λk) + Lρ(xk+1, λk) − Lρ(xk, λk) + γk+1 2 ∥xk+1 − xk∥2 − γk 2 ∥xk − xk−1∥2 ≤ � F(xk+1) − F(xk) − ∇F(xk)(xk+1 − xk) , ∆λk+1 � + 1 ρ∥∆λk+1∥2 − ρσ2 + αβk+1 − γk+1 2 ∥∆xk+1∥2 − γk 2 ∥∆xk∥2 ≤ � 1 2 η√ρη−1 + 1 ρ � ∥∆λk+1∥2 − γk 2 ∥∆xk∥2 − �ρσ2 + αβk+1 − γk+1 2 − 2M 2 F η√ρη−1 � ∥∆xk+1∥2, (6) where the first inequality follows from Lemma 3 and from the update of the dual multipliers in Step 6 of Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The second inequality holds for any η > 1 and follows from Assumption 2 and the inequality: ⟨a , b⟩ ≤ 1 2r∥a∥2 + r 2∥b∥2 ∀a, b ∈ Rm and r > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (7) Next, let us bound ∥∆λk+1∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 4 [Bound for ∥∆λk+1∥] If Assumption 2 holds on some compact set S and the sequence generated by Algorithm 1 is in S, then we have: ∥∆λk+1∥2 ≤ c1(βk+1)∥∆xk+1∥2 + c2(βk)∥∆xk∥2, (8) where c1(βk+1) = 2 (Lf +βk+1)2 σ2 and c2(βk) = 2 (Mf LF +(2MF +σ)βk)2 σ4 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 9 Subsequently, using (8) in (6), we obtain that: Pk+1 − Pk ≤ � �� 1 2 η√ρη−1 + 1 ρ � c1(βk+1) − γk+1 2 � − η√ρη−1( η√ρσ2 − 4M 2 F ) + αβk+1 − 2γk+1 2 � ∥∆xk+1∥2 + �� 1 2 η√ρη−1 + 1 ρ � c2(βk) − γk 2 � ∥∆xk∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, if we choose γk ≤ αβk 2 and ρ ≥ ( 4M 2 F σ2 )η, we get: Pk+1 − Pk ≤ �� 1 2 η√ρη−1 + 1 ρ � c1(βk+1) − γk+1 2 � ∥∆xk+1∥2 + �� 1 2 η√ρη−1 + 1 ρ � c2(βk) − γk 2 � ∥∆xk∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, in order to obtain a decreasing Lyapunov function along iterates, we must take into account when choosing γk the following: γk > 2 � 1 2 η√ρη−1 + 1 ρ � max{c1(βk), c2(βk)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, we want the parameter γk to satisfy the following requirements: 2 � 1 2 η√ρη−1 + 1 ρ � max{c1(βk), c2(βk)} < γk ≤ αβk 2 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (9) Choosing γk = 4 � 1 2 η√ρη−1 + 1 ρ � max{c1(βk), c2(βk)}, one can see that it al- ready satisfies the left inequality in (9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It remains to check the right inequality in (9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' To do so, we have to check if there exists a βk > 0 which satisfies the following inequality: 8 � 1 2 η√ρη−1 + 1 ρ � max{c1(βk), c2(βk)} ≤ αβk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (10) Let us determine max{c1(βk), c2(βk)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using the definitions of c1(βk) and c2(βk), we have: c1(βk) − c2(βk) = 2 σ2 �� Lf − MfLF σ � − 2MF σ βk � �� Lf + MfLF σ � + �2MF σ + 2 � βk � , which leads to two different cases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 10 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Case 1: If βk ≥ Lf σ−Mf LF 2MF , then max{c1(βk), c2(βk)} = c2(βk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us now check if βk, which satisfies (10), is well-defined in this case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' To do this, we replace the expression of c2(βk) in (10) and reformulate (10) as follows: 16 σ4 � 1 2 η√ρη−1 + 1 ρ � � (2MF + σ)2 β2 k + 2 (2MF + σ) MfLF βk + (MfLF )2� ≤ αβk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (11) Denoting: h1(βk) =16 σ4 � 1 2 η√ρη−1 + 1 ρ � (2MF + σ)2 β2 k + 16 σ4 � 1 2 η√ρη−1 + 1 ρ � (MfLF )2 + � 32 σ4 � 1 2 η√ρη−1 + 1 ρ � (2MF + σ) MfLF − α � βk, the inequality (11) is then equivalent to h1(βk) ≤ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since the roots of the second order equation h1(βk) = 0 are: ¯ β1 = δ � 1 − � 1 − 1/δ2 � MfLF 2MF + σ and ¯β1 = δ � 1 + � 1 − 1/δ2 � MfLF 2MF + σ , it follows that (9) and (11) are verified for any βk satisfying ¯ β1 ≤ βk ≤ ¯β1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Also, it is clear that when δ → ∞, the interval [ ¯ β1, ¯β1] → (0, ∞).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, we should choose δ large enough so that ¯β1 ≥ β.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If it happens that for some k ≥ 1, we obtain from (3), Lf σ−Mf LF 2MF ≤ βk < ¯ β1, then we take βk = ¯ β1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that when choosing α one has to note that for α → 0, ρ → ∞, whereas for α → 1, the proximal parameter βk might be large and similar β from Assumption 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Case 2: Similarly, if βk < Lf σ−Mf LF 2MF , then max{c1(βk), c2(βk)} = c1(βk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us check if βk, that satisfies (10), is well-defined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Replacing c1(βk) by its expression in (10), it allows us to recast (10) as: 16 σ2 � 1 2 η√ρη−1 + 1 ρ � � β2 k + 2Lfβk + L2 f � ≤ αβk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (12) Let us denote: h2(βk) =16 σ2 � 1 2 η√ρη−1 + 1 ρ � β2 k + 16 σ2 � 1 2 η√ρη−1 + 1 ρ � L2 f + � 32 σ2 � 1 2 η√ρη−1 + 1 ρ � Lf − α � βk, then the inequality (12) is equivalent to h2(βk) ≤ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since the roots of the equation h2(βk) = 0 are: ¯ β2 = δ′ � 1 − � 1 − 1/δ′2 � Lf and ¯β2 = δ′ � 1 + � 1 − 1/δ′2 � Lf, Linearized augmented Lagrangian for nonconvex optimization 11 it follows that (9) and (12) are valid for all ¯ β2 ≤ βk ≤ ¯β2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We choose δ′ such that ¯β2 = Lf σ−Mf LF 2MF (otherwise, if it happens that βk > Lf σ−Mf LF 2MF , we have c2(βk) = max{c1(βk), c2(βk)} and then we are in the first case discussed previously).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If for some k ≥ 1, we get from (3) βk < ¯ β2, then we take βk = ¯ β2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In conclusion, from the two cases above, if we choose: ρ ≥ max ��4M 2 F σ2 �η , �48(δ′ + 1) ασ2 Lf � η η−1 , �48(δ + 1) ασ4 (2MF + σ)MfLF � η η−1� , γk = 4 � 1 2 η√ρη−1 + 1 ρ � max{c1(βk), c2(βk)}, then we have: γk ≤ αβk 2 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (13) This, in turn, implies that the Lyapunov function decreases along the iterates: Pk+1 − Pk ≤ −γk+1 4 ∥∆xk+1∥2 − γk 4 ∥∆xk∥2 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (14) In the sequel, we assume that x0 is chosen such that: ∥F(x0)∥2 ≤ min � 1, c0 ρ � for some c0 > 0, (15) and that f(x0) ≤ ¯U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us define: ˆα ≜ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 3, (16) and DS := max{∥x − y∥ : x, y ∈ S0 ˆα}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, let us choose: ρ ≥ max ��4M 2 F σ2 �η , �48(δ′ + 1) ασ2 Lf � η η−1 , �48(δ + 1) ασ4 (2MF + σ)MfLF � η η−1 , 3ρ0, ρ0 + �Mf(2MF + σ) + 2δMfLF DS √ 2σ(2MF + σ) �2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (17) Note that, if we choose δ large enough so that ¯β1 ≥ β, we get: (Mf(2MF + σ) + 2δMfLF DS)2 2σ2(2MF + σ)2 ≥ (Mf + βDS)2 2σ2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It follows that ρ ≥ ρ0 + (Mf +βDS)2 2σ2 and 4M 2 F σ2 > 1, which imply that ρ ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using the definition of Lρ, we have: Lρ(x0, λ0) = f(x0) + ⟨λ0 , F(x0)⟩ + ρ 2∥F(x0)∥2 (7) ≤f(x0) + ∥λ0∥2 2ρ + ρ 2∥F(x0)∥2 + ρ 2∥F(x0)∥2 (15) ≤ ¯U + 1 2ρ∥λ0∥2 + c0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (18) 12 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara It then follows, after some re-arrangements, that: ¯U + c0 − ¯L ≥ f(x0) + ⟨λ0 , F(x0)⟩ + ρ 2∥F(x0)∥2 − 1 2ρ∥λ0∥2 − ¯L (ρ≥3ρ0) ≥ f(x0) + ρ0 2 ∥F(x0)∥2 − ¯L + ⟨λ0 , F(x0)⟩ + ρ 3∥F(x0)∥2 − ∥λ0∥2 2ρ (2) ≥ ρ 3∥F(x0) + 3λ0 2ρ ∥2 − 3 4ρ∥λ0∥2− ∥λ0∥2 2ρ (ρ≥1) ≥ − 5 4∥λ0∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (19) The following lemma shows that the sequence {(xk, λk)}k≥1 generated by Al- gorithm 1, is bounded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 5 Consider Algorithm 1 and let {Pk}k≥1 as defined in (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If As- sumptions 1, 2, 3 and 4 hold with S = S0 ˆα and ˆα defined in (16) for any fixed constant c0 and DS the radius of Sˆα0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If ρ is chosen as in (17) and x0 is chosen to satisfy (15), then we have the following: xk ∈ S0 ˆα, (20a) ∥λk∥2 ≤ (Mf + βDS)2 σ2 ≤ 2(ρ − ρ0), (20b) Pk ≤ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 2 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (20c) Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Next, we show that the Lyapunov sequence {Pk}k≥1 is bounded from below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 6 Consider Algorithm 1 and let {Pk}k≥1 defined in (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If Assump- tions 1, 2, 3 and 4 hold with S = S0 ˆα and ˆα defined in (16) for any fixed constant c0 and DS the radius of Sˆα0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If ρ is chosen as in (17) and x0 is chosen to satisfy (15), then we have the following: Pk ≥ ¯L − 1 ∀k ≥ 1 (21) where ¯L is defined in (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us now bound the gradient of the augmented Lagrangian function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 7 [Boundedness of ∇Lρ] Let {(xk, λk)}k≥1 be a sequence generated by the Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If Assumption 2 holds, then: ∥∇Lρ(xk+1, λk+1)∥ ≤ Γ(βk+1)∥xk+1 − xk∥ + Γ(βk)∥xk − xk−1∥, where the function Γ(βk) = � MF + 1 ρ � MfLF + MF Lf + (3MF + σ)βk σ2 + 2MF (ρMF + 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 13 Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that if Assumption 4 is satisfied, then {Γ(βk)}k≥1 is finite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, in the sequel we define this bound as follows: Γmax := sup k≥1 {Γ(βk)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (22) 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='1 Global convergence In this section we prove global convergence for the iterates generated by Algo- rithm 1 and also convergence rates to an ϵ first-order solution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Based on the previous lemmas, we are now ready to present the global asymptotic conver- gence of the iterates of Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Theorem 1 [Limit points are stationary points] If Assumptions 1, 2, 3 and 4 hold with S = S0 ˆα and ˆα defined in (16) for any fixed constant c0 and DS the radius of Sˆα0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If ρ is chosen as in (17) and x0 is chosen to satisfy (15), then any limit point (x∗, λ∗) of the sequence {(xk, λk)}k≥1, generated by Algorithm 1, is a stationary point, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', ∇Lρ(x∗, λ∗) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Equivalently: ∇f(x∗) + ∇F(x∗)T λ∗ = 0, F(x∗) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof Using (14) and the fact that γk > 0, we have: γk+1 4 ∥∆xk+1∥2 + γk 4 ∥∆xk∥2 ≤ Pk − Pk+1 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let k ≥ 1, by summing up the above inequality from i = 1 to i = k, we obtain: k � i=1 �γi+1 4 ∥∆xi+1∥2 + γi 4 ∥∆xi∥2� ≤ P1 − Pk+1 Lemma 6 ≤ P1 − (¯L − 1) (20c) ≤ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 3 − ¯L = ˆα − ¯L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (23) Since (23) holds for any k ≥ 1, we have: ∞ � i=1 �γi+1 4 ∥∆xi+1∥2 + γi 4 ∥∆xi∥2� < ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This, together with the fact that γk > 0, yields that: lim k→∞ ∥∆xk∥ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (24) From (20b), (20a) and the fact that S0 ˆα is compact, it follows that the sequence {(xk, λk)}k≥1 is bounded and there exists a convergent subsequence, let us say {(xk, λk)}k∈K, with the limit (x∗, λ∗).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' From Lemma 7 and (22), we have: ∥∇Lρ(x∗, λ∗)∥ = lim k∈K ∥∇Lρ(xk, λk)∥ ≤ Γmax lim k∈K(∥∆xk∥ + ∥∆xk−1∥) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, ∇Lρ(x∗, λ∗) = 0, which completes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ 14 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara For the remainder of this paper, we define: ¯γ := sup k≥1 {γk} (13),Ass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 4 < ∞ and ¯ γ := inf k≥1{γk} > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (25) Let us now present one of the main results of this paper, which derives the outer complexity of Algorithm 1 to find an ϵ first-order solution of problem (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Theorem 2 [First-order complexity] Consider Algorithm 1 and let {Pk}k≥1 be defined as in (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If Assumptions 1, 2, 3 and 4 hold with S = S0 ˆα and ˆα defined in (16) and ρ is chosen as in (17), then for any ϵ > 0, Algorithm 1 yields an ϵ first-order solution of (1) after K = 16Γmax 2 � ˆα−¯L ¯ γ � 1 ϵ2 outer iterations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof Let K ≥ 1, then from (23) and (25), we have: ¯ γ 4 K � i=1 � ∥∆xi+1∥2 + ∥∆xi∥2� ≤ K � i=1 �γi+1 4 ∥∆xi+1∥2 + γi 4 ∥∆xi∥2� ≤ ˆα − ¯L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, there exists k∗ ∈ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', K} such that: ∥∆xk∗+1∥2 + ∥∆xk∗∥2 ≤ 4 ˆα − ¯L K ¯ γ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It implies that: ∥∆xk∗+1∥ ≤ 2 � ˆα−¯L K ¯ γ and ∥∆xk∗∥ ≤ 2 � ˆα−¯L K ¯ γ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, from Lemma 7 and (22), we get: ∥∇Lρ(xk∗+1, λk∗+1)∥ ≤ Γmax � ∥∆xk∗+1∥ + ∥∆xk∗∥ � ≤ 4Γmax � ˆα − ¯L K ¯ γ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It follows that for any ϵ > 0, ∥∇Lρ(xk∗+1, λk∗+1)∥ ≤ ϵ when K ≥ 16Γmax 2 � ˆα−¯L ¯ γϵ2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Consequently, after K = 16Γmax 2( ˆα−¯L ¯ γ ) 1 ϵ2 outer iterations Algorithm 1 yields an ϵ first-order solution of (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This concludes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='2 Improved rates under KL In this section, under the KL condition, we provide better convergence rates for the iterates of Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In particular, we prove that the whole sequence {(xk, λk)}k≥1 converges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In order to show these results, we first bound the full gradient ∇P(·) (recall that P(·) is the function defined in (4)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 8 [Boundedness of ∇P] Let {(xk, λk)}k≥1 be the sequence generated by Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If Assumptions 1, 2, 3 and 4 hold with S = S0 ˆα and ˆα defined in (16) for any fixed constant c0, DS is the radius of S0 ˆα and ρ is chosen as in (17), then we have for any k ≥ 1: ∥∇P(xk+1, λk+1, xk, γk+1)∥ ≤ (Γmax+DS+2¯γ) (∥xk+1 − xk∥ + ∥xk − xk−1∥) , where, Γmax and ¯γ are defined in (22) and (25), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 15 Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The above lemma directly implies the following: ∥∇P(xk+1, λk+1, xk, γk+1)∥2 ≤ 2(Γmax + DS + 2¯γ)2 � ∥∆xk+1∥2 + ∥∆xk∥2� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (26) Then, it follows from (26) and (14), that: Pk+1 − Pk ≤ − ¯ γ 8(Γmax + DS + 2¯γ)2 ∥∇P(xk+1, λk+1, xk, γk+1)∥2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (27) Let us denote zk = (xk, λk) and uk = (xk, λk, xk−1, γk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, crit P denote the set of critical points of the function P(·) defined in (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Furthermore, we denote Ek = Pk − P ∗, where P ∗ = limk→∞ Pk (recall that the sequence {Pk}k≥1 is decreasing and bounded from bellow according to (14) and Lemma 6, respectively, hence it is convergent).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us denote the set of limit points of {uk}k≥1 by: Ω := {u∗ : ∃ a convergent subsequence {uk}k∈K such that lim k∈K uk = u∗}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us now prove the following lemma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 9 Consider Algorithm 1 and let {Pk}k≥1 be defined as in (4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If Assumptions 1, 2, 3 and 4 hold, with S = S0 ˆα and ˆα defined in (16) for any fixed constant c0, DS is the radius of S0 ˆα and ρ is chosen as in (17), then the following statements hold: (i) Ω is a compact subset of crit P and limk→∞ dist(uk, Ω) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (ii) For any u ∈ Ω, we have P(u) = P ∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (iii) For any (x, λ, y, γ) ∈ crit P, we have (x, λ) a stationary point of (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us now prove that the sequence � ∥∆xk∥ + ∥∆λk∥ � k≥1 has finite length, provided that a KL condition holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lemma 10 Let {(xk, λk)}k≥1 be the sequence generated by Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let Assumptions 1, 2, 3 and 4 hold, with S = S0 ˆα and ˆα defined in (16) for any fixed constant c0, and DS is the radius of S0 ˆα.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, assume that P(·) defined in (4) satisfies the K�L property on Ω.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, {zk}k≥1 = {(xk, λk)}k≥1 satisfies the finite length property, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', ∞ � k=1 ∥∆xk∥ + ∥∆λk∥ < ∞, and consequently converges to a stationary point of (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof See Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 16 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Lemma 10 shows that the set of limit points of the sequence {(xk, λk)}k≥1 is a singleton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us denote its limit by (x∗, λ∗).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We are now ready to present the convergence rates of the whole sequence generated by Algorithm 1 (see also [30] for a similar reasoning).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Theorem 3 [Convergence rates of {(xk, λk)}k≥1] If Assumptions 1, 2, 3 and 4 hold and z∗ := (x∗, λ∗) is a limit point of the sequence {zk := (xk, λk)}k≥1 generated by Algorithm 1, then: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If P(·) satisfies K�L property at u∗ := (x∗, λ∗, x∗, γ∗), where γ∗ is a limit point of the sequence {γk}k≥1, then there exists k1 ≥ 1 such that for all k ≥ k1 we have: ∥zk − z∗∥ ≤ C max{ϕ(Ek), � Ek−1}, where C > 0 and ϕ ∈ Ψτ, with τ > 0, denotes a desingularizing function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, if P(·) satisfies K�L property with desingularizing function ϕ : [0, τ) → [0, +∞), ϕ(s) = s1−ν, where ν ∈ [0, 1), then the following rates hold (a) If ν = 0, then zk converges to z∗ in a finite number of iterations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (b) If ν ∈ (0, 1 2), then for all k ≥ k1, we have: ∥zk − z∗∥ ≤ � Ek1 � (1 + ¯cE2ν−1 k1 )k−k1 , where ¯c = ¯ γ 8(Γmax + DS + 2¯γ)2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (c) If ν ∈ ( 1 2, 1), then for all k > k1, we have: ∥zk − z∗∥ ≤ � 1 µ(k − k1) + E1−2ν k1 � 1−ν 2ν−1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Proof 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using Lemma 4, we get: ∥∆λk+1∥2 ≤ c1(β)∥∆xk+1∥2 + c2(β)∥∆xk∥2 ≤ max{c1(β), c2(β)} � ∥∆xk+1∥2 + ∥∆xk∥2� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (28) Adding the term ∥∆xk+1∥2 + ∥∆xk∥2 on both sides in (28), we obtain: ∥zk+1 − zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2 ≤ ∥∆xk+1∥2 + ∥∆λk+1∥2 + ∥∆xk∥2 (28) ≤ � max{c1(β), c2(β)} + 1 �� ∥∆xk+1∥2 + ∥∆xk∥2� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (29) Linearized augmented Lagrangian for nonconvex optimization 17 Considering (25), we can then rewrite (14) as follows: Pk+1 − Pk (14) ≤ − ¯ γ 4 � ∥∆xk+1∥2 + ∥∆xk∥2� (29) ≤ − ¯ γ 4 � max{c1(β), c2(β)} + 1 �∥zk+1 − zk∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (30) Based on our choice of ρ and γk, the sequence {Pk}k≥1 is monotonically decreasing, see (14), and consequently {Ek}k≥1 is monotonically decreasing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using (30) and the fact that {Ek}k≥1 is non-negative, we have for all k ≥ 1: ∥∆xk+1∥ + ∥∆λk+1∥ ≤ 2 � 2 � max{c1(β), c2(β)} + 1 � ¯ γ � Ek.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (31) Without loss of generality, we assume that γ∗ is unique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since Pk → P ∗, uk → u∗ and P(·) satisfies the K�L property at u∗, then there exists k1 = k1(ϵ, τ) ≥ 1 such that ∀k > k1, we have ∥uk − u∗∥ ≤ ϵ and P ∗ < Pk < P ∗ + τ, and the following K�L property holds: ϕ′(Ek)∥∇P(xk, λk, xk−1, γk)∥ ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (32) Since ϕ is concave function, we have ϕ(Ek) − ϕ(Ek+1) ≥ ϕ′(Ek)(Ek − Ek+1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, from (30) and (32), we get: ∥zk+1 − zk∥2 ≤ ϕ′(Ek)∥zk+1 − zk∥2∥∇P(xk, λk, xk−1, γk)∥ ≤ 4 � max{c1(β), c2(β)} + 1 � ¯ γ ϕ′(Ek)(Ek − Ek+1)∥∇P(xk, λk, xk−1, γk)∥ ≤ 4 � max{c1(β), c2(β)} + 1 � ¯ γ � ϕ(Ek) − ϕ(Ek+1) � ∥∇P(xk, λk, xk−1, γk)∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' One can notice the following trivial inequality: for any a, b, c, d ≥ 0, if a2 + b2 ≤ c × d, then (a + b)2 ≤ 2a2 + 2b2 ≤ 2c × d ≤ c2 + d2 ≤ (c + d)2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using this relation and the fact that ∥zk+1−zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2, we have for any θ > 0: ∥∆xk+1∥ + ∥∆λk+1∥ ≤4 � max{c1(β), c2(β)} + 1 � θ ¯ γ � ϕ(Ek) − ϕ(Ek+1) � + 1 θ∥∇P(xk, λk, xk−1, γk)∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (33) Furthermore, we have: ∥∇P(xk, λk, xk−1, γk)∥ ≤ ∥∇Lρ(xk, λk)∥ + 2¯γ∥xk − xk−1∥ ≤(Γmax + DS + 2¯γ) (∥∆xk∥ + ∥∆λk∥) , 18 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara where the last inequality follows from the definition of Lρ and the properties of the derivative together with Lemma 7 and (22).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, (33) becomes: ∥∆xk+1∥ + ∥∆λk+1∥ ≤4 � max{c1(β), c2(β)} + 1 � θ ¯ γ � ϕ(Ek) − ϕ(Ek+1) � + Γmax + DS + 2¯γ θ � ∥∆xk∥ + ∥∆λk∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us now choose θ > 0 such that 0 < Γmax+DS+2¯γ θ < 1 and define a parameter δ0 as δ0 = 1 − Γmax+DS+2¯γ θ > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, summing up the above inequality over k > k1, we get: � k≥k1 ∥∆xk+1∥ + ∥∆λk+1∥ ≤4 � max{c1(β), c2(β)} + 1 � θ ¯ γδ0 ϕ(Ek1) + Γmax + DS + 2¯γ θδ0 � ∥∆xk1∥ + ∥∆λk1∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, using the triangle inequality, we get for any k ≥ k1: ∥zk − z∗∥ ≤ � l≥k ∥zl − zl+1∥ ≤ � l≥k ∥∆xl+1∥ + ∥∆λl+1∥ ≤ 4 � max{c1(β), c2(β)} + 1 � θ ¯ γδ0 ϕ(Ek) + Γmax + DS + 2¯γ θδ0 � ∥∆xk∥ + ∥∆λk∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Further, using (31), it follows that: ∥zk − z∗∥ ≤ 4 � max{c1(β), c2(β)} + 1 � θ ¯ γδ0 ϕ(Ek) + 2(Γmax + DS + 2¯γ) θδ0 � 2 � max{c1(β), c2(β)} + 1 � ¯ γ � Ek−1 ≤ C max{ϕ(Ek), � Ek−1}, where C = max � 4 � max{c1(β), c2(β)} + 1 � θ ¯ γδ0 , 2(Γmax + DS + 2¯γ) θδ0 � 2 � max{c1(β), c2(β)} + 1 � ¯ γ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let ν ∈ [0, 1) and for all s ∈ [0, τ), ϕ(s) = s1−ν and ϕ′(s) = (1 − ν)s−ν.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It follows that ∀k ≥ k1, we have: ∥zk − z∗∥ ≤ C max{E1−ν k , � Ek−1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (34) Linearized augmented Lagrangian for nonconvex optimization 19 Furthermore, (32) yields: Ek ν ≤ ∥∇P(xk, λk, xk−1, γk)∥ ∀k ≥ k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, from (27), we have for any k ≥ 1: ∥∇P(xk, λk, xk−1, γk)∥2 ≤ 8(Γmax + DS + 2¯γ)2 ¯ γ (Ek−1 − Ek).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, Ek 2ν ≤ 8(Γmax + DS + 2¯γ)2 ¯ γ (Ek−1 − Ek) ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Setting ¯c = ¯ γ 8(Γmax+DS+2¯γ)2 > 0, we get the recurrence ¯cEk 2ν ≤ Ek−1 − Ek ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (a) Let ν = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' If Ek > 0 for any k > k1, we have ¯c ≤ Ek−1 − Ek.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' As k goes to infinity, the right hand side approaches zero.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, 0 < ¯c ≤ 0 which is a contradiction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, there exists k > k1 such that Ek = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, Ek → 0 in a finite number of steps and from (34), zk → z∗ in a finite number of steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (b) Let ν ∈ (0, 1 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, 2ν − 1 < 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since {Ei}i≥k1 is mono- tonically decreasing, then Ei ≤ Ek1 for any i ∈ {k1 + 1, k1 + 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', k} and ¯cEk1 2ν−1Ek ≤ Ek−1 − Ek ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Rearranging this, we get for all k > k1: Ek ≤ Ek−1 1 + ¯cEk1 2ν−1 ≤ Ek−2 (1 + ¯cEk1 2ν−1)2 ≤ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ≤ Ek1 (1 + ¯cEk1 2ν−1)k−k1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, we have max{Ek 1−ν, � Ek−1} = � Ek−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It then follows that: ∥zk − z∗∥ ≤ � Ek1 � (1 + ¯cE2ν−1 k1 )k−k1 , (c) Let ν ∈ (1/2, 1), we have: ¯c ≤ (Ek−1 − Ek)Ek −2ν ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (35) Let h : R+ → R be defined as h(s) = s−2ν for any s ∈ R+.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It is clear that h is monotonically decreasing and ∀s ∈ R+, h′(s) = −2νs−(1+2ν) < 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since Ek ≤ Ek−1 for all k > k1, then h(Ek−1) ≤ h(Ek) for all k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We consider two cases: 20 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Case 1: Let r0 ∈ (1, +∞) such that: h(Ek) ≤ r0h(Ek−1) k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, from (35) we get: ¯c ≤ r0(Ek−1 − Ek)h(Ek−1) ≤ r0h(Ek−1) � Ek−1 Ek 1 ds ≤ r0 � Ek−1 Ek h(s) ds = r0 � Ek−1 Ek s−2ν ds = r0 1 − 2ν (Ek−1 1−2ν − Ek 1−2ν).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since ν > 1 2, it follows that: 0 < ¯c(2ν − 1) r0 ≤ Ek 1−2ν − Ek−1 1−2ν.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us define ˆc = ¯c(2ν−1) r0 and ˆν = 1 − 2ν < 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We then get: 0 < ˆc ≤ Ek ˆν − Ek−1 ˆν ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (36) Case 2: Let r0 ∈ (1, +∞) such that: h(Ek) > r0h(Ek−1), k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We then have Ek −2ν ≥ r0Ek−1 −2ν.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This leads to qEk−1 ≥ Ek, where q = r0− 1 2ν ∈ (0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since ˆν = 1 − 2ν < 0 we have qˆνEk−1 ˆν ≤ Ek ˆν and then, it follows that: (qˆν − 1)Ek−1 ˆν ≤ Ek−1 ˆν − Ek ˆν.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since qˆν − 1 > 0 and Ek → 0+ as k → ∞, there exists ˜c such that (qˆν − 1)Ek−1 ˆν ≥ ˜c for all k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, we obtain: 0 < ˜c ≤ Ek ˆν − Ek−1 ˆν ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (37) By choosing µ = min{ˆc, ˜c} > 0, one can combine (36) and (37) to obtain 0 < µ ≤ Ek ˆν − Ek−1 ˆν ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Summing the above inequality from k1 + 1 to some k > k1 gives µ(k − k1) + Ek1 ˆν ≤ Ek ˆν.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, Ek ≤ (µ(k − k1) + Ek1 ˆν) 1 ˆν = (µ(k − k1) + Ek1 1−2ν) 1 1−2ν .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since ν ∈ ( 1 2, 1), then max{Ek−1 1−ν, � Ek−1} = Ek−1 1−ν.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, (34) becomes: ∥zk − z∗∥ ≤ � 1 µ(k − k1) + E1−2ν k1 � 1−ν 2ν−1 , ∀k > k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This concludes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Linearized augmented Lagrangian for nonconvex optimization 21 5 Numerical results In this section we numerically compare Algorithm 1 (Linearized AL) with SCP algorithm [20] and IPOPT [28] on quadratic problems with quadratic equal- ity constraints (QPQCs).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that Proximal Augmented Lagrangian method presented in [29] is very slow compared to these three methods, hence we de- cided not to include it in our comparisons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The simulations are implemented in MATLAB and executed on a PC with (CPU 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='70GHz, 16GB RAM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' For the implementation of our method, we choose βk constant and equal to 1 and the penalty parameter ρ = 103 if not specified differently.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since one cannot guar- antee that the SCP iterates converge to a KKT point, we choose the following stopping criteria: we stop the algorithms when the difference between two con- secutive values of the objective function is less than a tolerance ϵ1 = 10−3 and the norm of constraints is less than a tolerance ϵ2 = 10−5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The numerical results are illustrated in Table 1 and Figure 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In Table 1, we compare the number of iterations, cpu time (sec), objective value and feasibility violation for Linearized AL, SCP and IPOPT on QPQCs from the CUTEst collection (the first part of the table) and randomly generated QCQP (the last 10 test cases in Table 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that for the test cases from CUTEst our algorithm, Linearized AL, is the best in terms of cpu time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' The ”-” indicates that an algorithm is unable to solve the corresponding problem in less than two hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' As one can see from Table 1, IPOPT could not find solutions of some random problems during the fixed time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Note that the last five random cases (in the second part of the table) have sparse data, while the first five random problems have dense data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This explains why the algorithms are faster for the sparse problems compared to the first five ones in this second part of the table, even though the latter have much smaller dimensions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Table 1 shows that our method is faster than SCP and IPOPT (in average it is 4 times faster than SCP and 10 times faster than IPOPT).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' For the optimal value function, we can see that sometimes Linearized AL and SCP methods obtain a worse optimal value than IPOPT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' However, in our case by increasing the penalty parameter ρ, we can also find the same optimal value as IPOPT, whereas this is not the case for SCP.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Figure 1 shows the performance profiling for computation time (left) and num- ber of iterations (right) for the 3 algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In the performance profiling, the vertical axis P(rp,s ≤ τ) (P(rp,s ≤ k)) represent the proportion of problems in the numerical experiments for which rp,s does not exceed τ (k), respectively, where rp,s is the ratio of the computational time (number of iterations) that the s solver takes to solve problem p to the shortest computational time (number of iterations) among the three algorithms to solve problem p, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It is clear from the computational time profile, Figure 1 (left), that the proposed algorithm, Liniarized AL, approaches 1 faster than SCP and IPOPT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' However, for the number of iterations this is not always the case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' From these prelim- inary experiments we can conclude that our new algorithm, Linearized AL, is an efficient method to solve optimization problems with nonlinear equality 22 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara XXXXXXXX (n,m) Alg Linearized AL SCP IPOPT # iter cpu # iter cpu # iter cpu f∗ ∥F∥ f∗ ∥F∥ f∗ ∥F∥ OPTCTRL3 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='12 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='17 7 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='40 (119,80) 2048.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='01 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='42e-14 2048.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='01 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='18e-12 2048.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='01 1,84e-08 OPTCTRL3;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ρ = 103 15 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='38 16 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='34 10 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='99 (1199,800) 19877.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='10e-08 19877.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='56e-09 18460.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='22 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='33e-09 OPTCTRL3;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ρ = 7 × 105 87 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='44 16 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='34 10 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='99 (1199,800) 18460.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='23 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='09e-07 19877.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='56e-09 18460.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='22 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='33e-09 OPTCTRL3 44 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='44 44 209.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='41 11 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='95 (4499,3000) 74465.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='03 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='43e-09 74465.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='03 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='82e-09 74465.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='03 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='09e-08 DTOC4 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='05 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='24 3 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 (299,198) 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='95 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='65e-07 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='95 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75e-07 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='94 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='02e-08 DTOC4 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='35 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='91 3 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='45 (1497,998) 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='88 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='26e-07 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='88 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='85e-09 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='88 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='20e-08 DTOC4 3 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='35 3 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='74 3 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='02 (4497,2998) 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='06e-07 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87e-10 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='66e-08 DTOC5;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ρ = 103 4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='33 4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='51 3 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='06 (998,499) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='67e-07 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75e-08 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='53 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='76e-07 DTOC5;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ρ = 2 × 104 120 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='63 4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='51 3 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='06 (998,499) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='54 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='79e-07 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75e-08 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='53 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='76e-07 DTOC5 5 53.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='05 4 138.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='74 3 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='25 (9998,4999) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='62 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='92e-08 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='62 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='68e-10 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='53 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='49e-07 ORTHREGA 37 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='91 39 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='73 76 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='14 (517,256) 1414.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='05 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='23e-06 1664.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='80 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='24e-06 1414.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='05 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='19e-10 ORTHREGA 53 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='27 67 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='78 14 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='99 (2053,1024) 5661.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='43 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='90e-07 6654.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='78 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='07e-06 5661.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='43 9,25e-07 ORTHREGA 38 65.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='72 193 3798.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87 20 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='78 (8197,4096) 22647.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='84 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='83e-07 22647.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='84 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='12e-06 22647.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='84 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='86e-09 (10,9) 27 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='07 34 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='14 7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='21 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='21 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='49e-06 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='21 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='05e-15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='21 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='17e-10 (20,13) 15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='12 15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='39 13 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='65 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='01e-06 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='01e-06 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='61 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='85e-15 (50,43) 62 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='90 61 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='68 12 87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='75 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='95 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='43e-06 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='95 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='86e-06 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='98 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='92e-14 (100,91) 629 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='53 204 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='42 21 143.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='74e 742.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='21 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='16e-06 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='72 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='70e-06 113.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='52 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='44e-12 (150,140) 1165 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='53 505 24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='89 37 417.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87 1027.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='24 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='51e-06 358.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='67 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='79e-06 231.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='26 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='382-11 (1000,500) 17 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='37 25 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='76 5 140.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='34 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='27 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='13e-06 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='27 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='76e-08 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='27 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='36e-10 (10000,500) 7 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='67 7 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='64 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='46 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='48e-07 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='46 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='40e-07 (100000,500) 7 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='12 7 73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='43 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='47 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='59e-06 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='47 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='87e-06 (100000,1000) 7 76.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='21 7 134.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='07 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='81 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='59e-06 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='81 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='54e-09 (300000,700) 7 164.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='54 7 178.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='35 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='59 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='55e-06 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='59 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='44e-07 Table 1 Comparison between Linearized AL, SCP and IPOPT on QCQPs from CUTEst (top) and randomly generated QCQPs (bottom).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 23 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 1 Performance profiles for the computation time (left) and the number of iterations (right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' constraints, usually much faster than well established solvers such as IPOPT and SCP.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 6 Conclusion In this paper, we have proposed a linearized augmented Lagrangian method for solving (locally) smooth optimization problems with nonlinear equality con- straints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In this method we have linearized the functional constraints within the augmented Lagrangian function and added a regularization term.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' By dy- namically generating the regularization (proximal) parameter, we have proved global asymptotic convergence, convergence rate to an ϵ first-order optimal solution and improved convergence rates under the K�L condition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, we have numerically shown that the proposed algorithm is efficient, comparing it with two known algorithms, SCP and IPOPT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Conflict of interest The authors declare that they have no conflict of interest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Data availability It is not applicable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='6 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content="6 VI s'd s'd P0." metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='2 Linearized AL Linearized AL SCP SCP ---- IPOPT ----IPOPT 0 5 10 15 20 2 4 6 8 10 T: cpu time (s) k: iterations24 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Appendix Proof of Lemma 2 Using the definition of ψ and ∇xψ, we have: ∥∇xψ(x, λ) − ∇xψ(x′, λ′)∥ = ∥∇F(x)T (λ + ρF(x)) − ∇F(x′)T (λ′ + ρF(x′))∥ = ���� � ∇F(x) − ∇F(x′) �T � λ + ρF(x) � + ∇F(x′)T � λ − λ′ + ρ � F(x) − F(x′) ������ ≤ ∥∇F(x) − ∇F(x′)∥∥λ + ρF(x)∥ + ∥∇F(x′)∥ � ∥λ − λ′∥ + ρ∥F(x) − F(x′)∥ � Ass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2 ≤ � LF ∥λ + ρF(x)∥ + ρM2 F � ∥x − x′∥ + MF ∥λ − λ′∥ ≤ � LF ∥λ + ρF(x)∥ + MF (1 + ρMF ) � ���� �x λ � − �x′ λ′ ����� ≤ sup x,λ {Lψ(x, λ)} ���� �x λ � − �x′ λ′ ����� , where Lψ(x, λ) = LF ∥λ + ρF(x)∥ + MF (1 + ρMF ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Similarly, using the expression of ∇λψ, we get: ∥∇λψ(x, λ) − ∇λψ(x′, λ′)∥ = ∥F(x) − F(x′)∥ Ass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2 ≤ MF ∥x − x′∥ ≤ MF ���� �x λ � − �x′ λ′ ����� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, using basic properties of the Euclidean norm, we obtain: ∥∇ψ(x, λ) − ∇ψ(x′, λ′)∥ ≤ ∥∇xψ(x, λ) − ∇xψ(x′, λ′)∥ + ∥∇λψ(x, λ) − ∇λψ(x′, λ′)∥ ≤ sup (x,λ)∈S×Λ � LF ∥λ + ρF(x)∥ + MF (2 + ρMF ) � ���� �x − x′ λ − λ′ ����� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This concludes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 3 Using the optimality of xk+1, we have: ¯Lρ(xk+1, λk;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' xk) + βk+1 2 ∥xk+1 − xk∥2 ≤ ¯Lρ(xk, λk;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' xk) = Lρ(xk, λk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Further, from definitions of ¯Lρ and Lρ, we get: f(xk+1) + ⟨λk , F(xk) + ∇F(xk)∆xk+1⟩ + ρ 2 ∥F(xk) + ∇F(xk)∆xk+1∥2 ≤ f(xk) + ⟨λk , F(xk)⟩ + ρ 2 ∥F(xk)∥2 − βk+1 2 ∥∆xk+1∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Rearranging this inequality, it follows: f(xk+1) − f(xk) ≤ − ρ 2 ⟨∇F(xk)∆xk+1 , 2F(xk) + ∇F(xk)∆xk+1⟩ − ⟨∇F(xk)∆xk+1 , λk⟩ − βk+1 2 ∥∆xk+1∥2 ≤ − ρ 2 ∥∇F(xk)∆xk+1∥2 − ⟨∇F(xk)T (λk + ρF(xk)) , ∆xk+1⟩ − βk+1 2 ∥∆xk+1∥2 Ass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2 ≤ − ρ 2 σ2∥∆xk+1∥2 − βk+1 2 ∥∆xk+1∥2 − ⟨∇F(xk)T (λk + ρF(xk)) , ∆xk+1⟩.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 25 Using the definitions of Lρ and ψ, we further obtain: Lρ(xk+1, λk) − Lρ(xk, λk) = f(xk+1) − f(xk) + ψ(xk+1, λk) − ψ(xk, λk) (3) ≤ − ρσ2 + αβk+1 2 ∥xk+1 − xk∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This proves our statement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 4 Using the optimality condition for xk+1, we have: ∇f(xk+1) + ∇F(xk)T λk + ρ∇F(xk)T � F(xk) + ∇F(xk)(xk+1 − xk) � + βk+1(xk+1 − xk) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Combining this with the update in Step 6 of Algorithm 1, we get: ∇f(xk+1) + ∇F(xk)T λk+1 + βk+1(xk+1 − xk) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (38) By replacing k with k − 1, we obtain: ∇f(xk) + ∇F(xk−1)T λk + βk(xk − xk−1) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (39) Subtracting (39) from (38), we have: ∇f(xk+1) − ∇f(xk) + ∇F(xk)T ∆λk+1 + � ∇F(xk) − ∇F(xk−1) �T λk + βk+1∆xk+1 − βk∆xk = 0 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Further, using Assumption 2, we have: ∥∆λk+1∥ ≤ 1 σ � ∥∇f(xk+1) − ∇f(xk)∥ + ∥∇F(xk) − ∇F(xk−1)∥∥λk∥ + βk+1∥∆xk+1∥ + βk∥∆xk∥ � ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (40) From (39), we further have: ∥λk∥ ≤ 1 σ � ∥∇f(xk)∥ + βk∥∆xk∥ � ≤ 1 σ � Mf + βk∥∆xk∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (41) Moreover, from Assumption 2, we have: ∥∇F(xk) − ∇F(xk−1)∥ ≤ LF ∥∆xk∥ and ∥∇F(xk) − ∇F(xk−1)∥ ≤ 2MF .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' By replacing, the above inequalities and (41) in (40), we obtain: ∥∆λk+1∥ ≤ 1 σ � Lf∥∆xk+1∥ + MfLF + 2MF βk σ ∥∆xk∥ + βk+1∥∆xk+1∥ + βk∥∆xk∥ � = Lf + βk+1 σ ∥∆xk+1∥ + MfLF + (2MF + σ)βk σ2 ∥∆xk∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (42) Since (a + b)2 ≤ 2a2 + 2b2, we finally get (8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 5 We prove this result using induction arguments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' From Lemma 3 for k = 0, we have the following: f(x1) + ⟨λ0 , F(x1)⟩ + ρ 2 ∥F(x1)∥2 + ρσ2 + αβ1 2 ∥x1 − x0∥2 ≤f(x0) + ⟨λ0 , F(x0)⟩ + ρ 2 ∥F(x0)∥2 (18) ≤ ¯U + 1 2ρ ∥λ0∥2 + c0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (43) 26 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara For i = 0 and i = 1, we obtain: f(xi) + ρ0 2 ∥F(xi)∥2(ρ≥3ρ0) ≤ f(xi) + ρ 6 ∥F(xi)∥2 (18),(43) ≤ ¯U + 1 2ρ ∥λ0∥2 + c0 − ⟨λ0 , F(xi)⟩ − ρ 3 ∥F(xi)∥2 ≤ ¯U + 1 2ρ ∥λ0∥2 + c0 − ρ 3 ∥F(xi) + 3λ0 2ρ ∥2 + 3∥λ0∥2 4ρ ≤ ¯U + c0 + 5∥λ0∥2 4ρ (19) ≤ ¯U + c0 + 5∥λ0∥2 4ρ + 3( ¯U + c0 − ¯L + 5∥λ0∥2 4ρ ) (ρ≥1) ≤ 4 ¯U + 4c0 − 3¯L + 5∥λ0∥2 ≤ ˆα.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, we find that x0, x1 ∈ S0 ˆα.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, using the optimality condition (39) for k = 1, we have: ∇f(x1) + ∇F(x0)T λ1 + β1(x1 − x0) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since x0, x1 ∈ S0 ˆα and DS is the diameter of S0 ˆα, then from Assumption 2: ∥λ1∥ ≤ 1 σ (∥∇f(x1)∥ + β1∥∆x1∥) ≤ 1 σ (Mf + βDS) ≤ 2(ρ − ρ0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Furthermore, exploiting the definition of Pk for k = 1, we have: P1 = Lρ(x1, λ1) + γ1 2 ∥x1 − x0∥2 = Lρ(x1, λ1) − Lρ(x1, λ0) + Lρ(x1, λ0) − Lρ(x0, λ0) + Lρ(x0, λ0) + γ1 2 ∥x1 − x0∥2 Lemma 3 ≤ ⟨λ1 − λ0 , F(x1)⟩ − ρσ2 + αβ1 − γ1 2 ∥x1 − x0∥2 + Lρ(x0, λ0) (7) ≤ ρ 2 ∥F(x1)∥2 + 1 2ρ ∥λ1 − λ0∥2 + Lρ(x0, λ0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (44) From (43), it follows that: ρ 6 ∥F(x1)∥2 ≤ ¯U + 1 2ρ ∥λ0∥2 + c0 − ρ 6 ∥F(x1)∥2 − ⟨λ0 , F(x1)⟩ − f(x1) − ρ 6 ∥F(x1)∥2 = ¯U + 1 2ρ ∥λ0∥2 + c0 − ρ 6 ∥F(x1) + 3λ0 ρ ∥2 + 3 2ρ ∥λ0∥2 − f(x1) − ρ 6 ∥F(x1)∥2 (ρ≥3ρ0) ≤ ¯U + 1 2ρ ∥λ0∥2 + c0 + 3 2ρ ∥λ0∥2 − f(x1) − ρ0 2 ∥F(x1)∥2 (2) ≤ ¯U + 2 ρ ∥λ0∥2 + c0 − ¯L (ρ≥1) ≤ ¯U + c0 − ¯L + 2∥λ0∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (45) Using (45) in (44), we obtain: P1 ≤ 3( ¯U + c0 − ¯L + 2∥λ0∥2) + 1 ρ ∥λ1∥2 + 1 ρ ∥λ0∥2 + ¯U + c0 + 1 2ρ ∥λ0∥2 ≤ 4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 27 It then follows that for k = 1, (20) is verified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Now, assume that (20) holds for some k ≥ 1 (induction hypothesis) and we will prove that it continues to hold for k + 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using Lemma 3 together with the definition of Pk, we have: Lρ(xk+1, λk) ≤ Lρ(xk, λk) ≤ Pk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' By using the expression of Lρ, we have that: f(xk+1) + ⟨λk , F(xk+1)⟩ + ρ 2 ∥F(xk+1)∥2 ≤ Pk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Thus, using (7), it follows that: f(xk+1) − ∥λk∥2 2(ρ − ρ0) − (ρ − ρ0)∥F(xk+1)∥2 2 + ρ 2 ∥F(xk+1)∥2 ≤ Pk, which yields the following: f(xk+1) + ρ0 2 ∥F(xk+1)∥2 ≤ Pk + ∥λk∥2 2(ρ − ρ0) ≤Pk + 1 ≤4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 3 = ˆα, where the last two inequalities are due to the induction hypothesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Therefore, xk+1 ∈ S0 ˆα.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using the same arguments as for k = 1, the optimality condition (38) and the fact that xk ∈ S0 ˆα from the induction hypothesis, it follows that: ∥λk+1∥2 ≤ (Mf + βDS)2 σ2 ≤ 2(ρ − ρ0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since xk, xk+1 ∈ S0 ˆα and from (14), we have: Pk+1 − Pk ≤ − γk+1 4 ∥∆xk+1∥2 − γk 4 ∥∆xk∥2 ≤ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Together with the induction hypothesis, we obtain: Pk+1 ≤ Pk≤4 ¯U + 4c0 − 3¯L + 8∥λ0∥2 + 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Finally, (20) is proved, which completes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 6 Using (5), we have: Pk ≥ f(xk) + ρ 2 ∥F(xk)∥2 + ⟨λk , F(xk)⟩ ≥ f(xk) + ρ 2 ∥F(xk)∥2 − ∥λk∥2 2(ρ − ρ0) − ρ − ρ0 2 ∥F(xk)∥2 (20b) ≥ f(xk) + ρ0 2 ∥F(xk)∥2 − 1 (Lemma 2) ≥ ¯L − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It follows that the sequence {Pk}k≥1 is bounded from below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 7 Using the optimality condition (38), we have: ∇f(xk+1) = −∇F(xk)T λk+1 − βk+1(xk+1 − xk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It then follows, by exploiting the definition of Lρ and the properties of the derivative, that: ∇xLρ(xk+1, λk+1) = ∇f(xk+1) + ∇F(xk+1)T � λk+1 + ρF(xk+1) � = � ∇F(xk+1) − ∇F(xk) �T λk+1 + ∇F(xk+1)T ∆λk+1 − βk+1∆xk+1 + ρ∇F(xk+1)T � F(xk+1) − F(xk) − ∇F(xk)∆xk+1 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 28 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara Using basic properties of the Euclidean norm, we further get: ∥∇xLρ(xk+1, λk+1)∥ ≤∥∇F(xk+1) − ∇F(xk)∥∥λk+1∥ + ∥∇F(xk+1)∥∥∆λk+1∥ + βk+1∥∆xk+1∥ + ρ∥∇F(xk+1)∥∥F(xk+1) − F(xk) − ∇F(xk)∆xk+1∥ Ass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='2,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='(41) ≤ MfLF + 2MF βk+1 σ ∥∆xk+1∥ + MF ∥∆λk+1∥ + βk+1∥∆xk+1∥ + 2ρM2 F ∥∆xk+1∥ = � MfLF + (2MF + σ)βk+1 σ + 2ρM2 F � ∥∆xk+1∥ + MF ∥∆λk+1∥ (46) (42) ≤ � MfLF + MF Lf + (3MF + σ)βk+1 σ + 2ρM2 F � ∥∆xk+1∥ + MF σ MfLF + (2MF + σ)βk σ ∥∆xk∥ ≤ � MF σ MfLF + MF Lf + (3MF + σ)βk+1 σ + 2ρM2 F � ∥∆xk+1∥ + � MF σ MfLF + MF Lf + (3MF + σ)βk σ + 2ρM2 F � ∥∆xk∥ Similarly,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' we have: ∥∇λLρ(xk+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' λk+1)∥ = ∥F(xk+1)∥ ≤∥F(xk+1) − F(xk) − ∇F(xk)∆xk+1∥ + 1 ρ ∥∆λk+1∥ Ass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2 ≤ 2MF ∥∆xk+1∥ + 1 ρ ∥∆λk+1∥ (47) (42) ≤ � 2MF + 1 ρ Lf + βk+1 σ � ∥∆xk+1∥ + 1 ρ MfLF + (2MF + σ)βk σ2 ∥∆xk∥ ≤ � 2MF + 1 ρ Lfσ + MfLF + (2MF + σ)βk+1 σ2 � ∥∆xk+1∥ + � 2MF + 1 ρ Lfσ + MfLF + (2MF + σ)βk σ2 � ∥∆xk∥ where the first inequality is obtained from the multipliers update in Step 6 of Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, it follows that: ∥∇Lρ(xk+1, λk+1)∥ ≤ ∥∇xLρ(xk+1, λk+1)∥ + ∥∇λLρ(xk+1, λk+1)∥ ≤ Γ(βk+1)∥xk+1 − xk∥ + Γ(βk)∥xk − xk−1∥, where Γ(βk) = � MF + 1 ρ � MfLF + MF Lf + (3MF + σ)βk σ2 + 2MF (ρMF + 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This proves our claim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 8 By exploiting the definition of P(·) defined in (4), we have that for any k ≥ 1: ∇xP(x, λ, y, γ) = ∇xLρ(x, λ) + γ(x − y), ∇λP(x, λ, y, γ) = ∇λLρ(x, λ) ∇yP(x, λ, y, γ) = γ(y − x) and ∇γP(x, λ, y, γ) = 1 2 ∥x − y∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 29 Hence, ∥∇P(xk+1, λk+1, xk, γk+1)∥ ≤ ∥∇Lρ(xk+1, λk+1)∥ + +2γk+1∥xk+1 − xk∥ + 1 2 ∥∆xk+1∥2 ≤(Γmax + DS + 2¯γ) (∥∆xk+1∥ + ∥∆xk∥) , where the last inequality follows from Lemma 7, (22) and (25).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This proves our claim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 9 (i) From Lemma 5 and Assumption 4, it follows that {uk}k≥1 is bounded and therefore, there exists a convergent subsequence {uk}k∈K such that limk∈K uk = u∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence Ω is nonempty.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, Ω is compact since it is bounded and closed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' On the other hand, for any u∗ ∈ Ω, there exists a sequence of increasing integers K such that limk∈K uk = u∗ and using Lemma 8 and (24), it follows that: ∥∇P(u∗)∥ = lim k∈K ∥∇P(uk)∥ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, u∗ ∈ crit P and 0 ≤ limk→∞ dist(uk, Ω) ≤ limk∈K dist(uk, Ω) = dist(u∗, Ω) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (ii) Since P(·) is continuous and {P(uk) = Pk}k≥1 converges to P ∗, then any subsequence {P(uk) = Pk}k∈K that converges, it converges to the same limit P ∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (iii) Let (x, λ, y, γ) ∈ crit P that is ∇P(x, λ, y, γ) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It then follows that: ∇xP(x, λ, y, γ) = ∇xLρ(x, λ) + γ(x − y) = 0, ∇λP(x, λ, y, γ) = ∇λLρ(x, λ) = 0 ∇yP(x, λ, y, γ) = γ(y − x) = 0 and ∇γP(x, λ, y, γ) = 1 2 ∥x − y∥2 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' With some minor rearrangements, we obtain: ∇f(x) + ∇F(x)T λ = 0, F(x) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence, (x, λ) is a stionary point of (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This concludes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ Proof of Lemma 10 From the boundedness of ∥∆λk+1∥2 derived in (8), we have: ∥∆λk+1∥2 ≤ c1(β)∥∆xk+1∥2 + c2(β)∥∆xk∥2 ≤ max{c1(β), c2(β)} � ∥∆xk+1∥2 + ∥∆xk∥2� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (48) Adding the term ∥∆xk+1∥2 + ∥∆xk∥2 on both sides in (48), we have: ∥zk+1 − zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2 ≤ ∥∆xk+1∥2 + ∥∆λk+1∥2 + ∥∆xk∥2 (48) ≤ � max{c1(β), c2(β)} + 1 �� ∥∆xk+1∥2 + ∥∆xk∥2� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (49) Considering (25), we can then rewrite (14) as follows: Pk+1 − Pk (14) ≤ − ¯ γ 4 � ∥∆xk+1∥2 + ∥∆xk∥2� (49) ≤ − ¯ γ 4 � max{c1(β), c2(β)} + 1 � ∥zk+1 − zk∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (50) Since Pk → P ∗ and {Pk}k≥1 is monotonically decreasing to P ∗, then it follows that the error sequence {Ek}k≥1, is non-negative, monotonically decreasing and converges to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' We distinguish two cases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Case 1: There exists k1 ≥ 1 such that Ek1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, Ek = 0 ∀k ≥ k1 and using (50), we have: ∥zk+1 − zk∥2 ≤ 4 � max{c1(β), c2(β)} + 1 � ¯ γ (Ek − Ek+1) = 0 ∀k ≥ k1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 30 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara From Lemma 5 the sequence {zk}k≥1 is bounded, and thus: ∞ � k=1 ∥∆xk∥ + ∥∆λk∥ = k1 � k=1 ∥∆xk∥ + ∥∆λk∥<∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Case 2: The error Ek > 0 ∀k ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, there exists k1 = k1(ϵ, τ) ≥ 1 such that ∀k ≥ k1 we have dist(uk, Ω) ≤ ϵ, P ∗ < P(uk) < P ∗ + τ and ϕ′(Ek)∥∇P(xk, λk, xk−1, γk)∥ ≥ 1, (51) where ϵ > 0, τ > 0 and ϕ ∈ Ψτ are well defined and correspond to those in Definition 2, recall that P(·) satisfies the K�L property on Ω.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since ϕ is concave, we have ϕ(Ek) − ϕ(Ek+1) ≥ ϕ′(Ek)(Ek − Ek+1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, from (50) and (51) we get: ∥zk+1 − zk∥2 ≤ ϕ′(Ek)∥zk+1 − zk∥2∥∇P(xk, λk, xk−1, γk)∥ ≤ 4 � max{c1(β), c2(β)} + 1 � ¯ γ ϕ′(Ek)(Ek − Ek+1)∥∇P(xk, λk, xk−1, γk)∥ ≤ 4 � max{c1(β), c2(β)} + 1 � ¯ γ � ϕ(Ek) − ϕ(Ek+1) � ∥∇P(xk, λk, xk−1, γk)∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since ∥zk+1 − zk∥2 = ∥∆xk+1∥2 + ∥∆λk+1∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Using the fact that for any a, b, c, d ≥ 0, if a2 + b2 ≤ c × d, then (a + b)2 ≤ 2a2 + 2b2 ≤ 2c × d ≤ c2 + d2 ≤ (c + d)2, it follows that for any θ > 0, we have: ∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 � max{c1(β), c2(β)} + 1 � θ ¯ γ � ϕ(Ek) − ϕ(Ek+1) � + 1 θ ∥∇P(xk, λk, xk−1, γk)∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' (52) Furthermore, we have: ∥∇P(xk, λk, xk−1, γk)∥ ≤ ∥∇Lρ(xk, λk)∥ + +2¯γ∥xk − xk−1∥ (46),(47) ≤ (Γmax + DS + 2¯γ) (∥∆xk∥ + ∥∆λk∥) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, (52) becomes: ∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 � max{c1(β), c2(β)} + 1 � θ ¯ γ � ϕ(Ek) − ϕ(Ek+1) � + Γmax + DS + 2¯γ θ � ∥∆xk∥ + ∥∆λk∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let us now choose θ > 0 so that 0 < Γmax+DS+2¯γ θ < 1 and define the parameter δ0 as: δ0 = 1− Γmax+DS+2¯γ θ > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, by summing up the above inequality from k = ¯k to k = K and using the property: �K k=¯k ∥∆xk∥ = �K k=¯k ∥∆xk+1∥ + ∥∆x ¯k∥ − ∥∆xK+1∥, we get: K � k=¯k ∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 � max{c1(β), c2(β)} + 1 � θ ¯ γδ0 � ϕ(E ¯k) − ϕ(EK+1) � + Γmax + DS + 2¯γ θδ0 � ∥∆x ¯k∥ + ∥∆λ ¯k∥ � − Γmax + DS + 2¯γ θδ0 � ∥∆xK+1∥ + ∥∆λK+1∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Linearized augmented Lagrangian for nonconvex optimization 31 Using the fact that {Ek}k≥k1 is monotonically decreasing and that the function ϕ is positive and increasing, which yields ϕ(Ek) ≥ ϕ(Ek+1) > 0, then: K � k=¯k ∥∆xk+1∥ + ∥∆λk+1∥ ≤ 4 � max{c1(β), c2(β)} + 1 � θ ¯ γδ0 ϕ(E ¯k) + Γmax + DS + 2¯γ θδ0 � ∥∆x ¯k∥ + ∥∆λ ¯k∥ � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' It is clear that the right-hand side of the above inequality is bounded for any K ≥ ¯k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Letting K → ∞, we get that: ∞ � k=¯k ∥∆xk+1∥ + ∥∆λk+1∥ < ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' From Lemma 5, the sequence {(xk, λk)}k≥1 is bounded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Then, it follows that: ¯k � k=1 ∥∆xk∥ + ∥∆λk∥ < ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hence: �∞ k=1 ∥∆xk∥ + ∥∆λk∥ < ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Let m, n ∈ Z+ such that n ≥ m, we have: ∥zn − zm∥ = ∥ n−1 � k=m ∆zk+1∥ ≤ n−1 � k=m ∥∆zk+1∥ ≤ n−1 � k=m ∥∆xk+1∥ + ∥∆λk+1∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Since �∞ k= ∥∆xk+1∥ + ∥∆λk+1∥ < ∞, it follows that ∀ε > 0, ∃N ∈ Z+ such that ∀m, n ∈ Z+ where n ≥ m, we have: ∥zn − zm∥ ≤ ε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This implies that {zk}k≥1 is a Cauchy sequence and converges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Moreover, by Theorem 1, {zk}k≥1 converges to a staionary point of (1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' This concludes our proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ⊓⊔ References 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Attouch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Bolte and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Svaiter, Convergence of descent methods for semi- algebraic and tame problems: proximal algorithms, forward-backward splitting, and regularized Gauss–Seidel methods, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Program.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' A 137, 91–129, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Bertsekas Constrained Optimization and Lagrange Multiplier Methods, Athena Scientific, Belmont, MA 1996.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Bertsekas Convex Optimization Algorithms, Athena Scientific, Belmont, MA 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Bertsekas, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Tsitsiklis, Parallel and Distributed Computation: Numer- ical Methods, Athena Scientific, 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Bolte, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Sabach and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Teboulle, Proximal alternating linearized minimization for nonconvex and nonsmooth problems, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Program.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 146(1–2), 459–494, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Bot and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Nguyen, The proximal alternating direction method of multipliers in the nonconvex setting: convergence analysis and rates, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Op.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 45(2), 682–712, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Boyd, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Parikh, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Chu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Peleato, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Eckstein, Distributed optimization and statistical learning via the alternating direction method of multipliers, Found.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Trends Mach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Learn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 3(1), 1–122, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Cohen, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hallak and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Teboulle, A Dynamic Alternating Direction of Multi- pliers for Nonconvex Minimization with Nonlinear Functional Equality Constraints, Journal of Optimization Theory and Applications, 193, 324–353, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Fan, Comments on “wavelets in statistics: a review, by A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Antoniadis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Ital.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Stat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 6, 131–138, 1997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Fessler, Optimization methods for magnetic resonance image reconstruction, key models and optimization algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' IEEE Signal Process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Mag.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 37(1), 33–40, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 32 L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' El Bourkhissi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Gabay, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Mercier, A dual algorithm for the solution of nonlinear variational problems via finite element approximation, Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2(1), 17–40, 1976.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Glowinski, and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Le Tallec, Augmented Lagrangian and Operator-Splitting Meth- ods in Nonlinear Mechanics, SIAM, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 9, 1989.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Grant and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Boyd, CVX: Matlab Software for Disciplined Convex Programming, version 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='1, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Available: http://cvxr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='com/cvx,Mar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='2014 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Haarhoff, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Buys, A new method for the optimization of a nonlinear function subject to nonlinear constraints, Computer Journal, 13, 178–184, 1970.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hajinezhad and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hong, Perturbed proximal primal-dual algorithm for noncon- vex nonsmooth optimization, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Program.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hestenes, Multiplier and gradient methods, Journal of Optimization Theory and Applications, 4, 303–320, 1969.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hong, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Hajinezhad, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Zhao, Prox-PDA: The proximal primal-dual algorithm for fast distributed nonconvex optimization and learning over networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In: D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Precup, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Teh (eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=') Proceedings of the 34th International Conference on Machine Learning, Proceedings of Machine Learning Research, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 70, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 1529–1538, PMLR 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Jiang, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Lin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Ma and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Zhang, Structured nonconvex and nonsmooth opti- mization: algorithms and iteration complexity analysis, Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Optim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 72(1), 115–157, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Luke, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Sabach, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Teboulle, Optimization on spheres: models and prox- imal algorithms with computational performance comparisons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' SIAM J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Data Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 1(3), 408–445, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Messerer, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Baumg¨artner and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Diehl, Survey of sequential convex programming and generalized Gauss-Newton methods, ESAIM: Proceedings and Surveys, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Necoara and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Kvamme, DuQuad: A toolbox for solving convex quadratic pro- grams using dual (augmented) first order algorithms, 2015 54th IEEE Conference on Decision and Control (CDC), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 2043-2048, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Powell, A method for nonlinear optimization in minimization problems, in Optimization (R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Fletcher, ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' ), Academic Press, 283–298, 1969.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Rockafellar and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Wets, Variational Analysis, Springer, Berlin, 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Royset, Variational Analysis in Modern Statistics, Special Issue Mathematical Programming, Series B, Volume 174, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Royer, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' O’Neill and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content='Wright, A Newton-CG algorithm with complex- ity guarantees for smooth unconstrained optimization, Mathematical Programming, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Shefi and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Teboulle, Rate of convergence analysis of decompositionmethods based on the proximal method of multipliers for convex minimization, SIAM J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Opt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 24(1), 269–297, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Tran-Dinh and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Diehl, Local convergence of sequential convex programming for nonconvex optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' In M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Diehl, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Glineur, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Jarlebring, and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Michiels, editors, Recent advances in optimization and its application in engineering, pages 93–103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Springer-Verl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=', 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' W¨achter and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Biegler, On the Implementation of a Primal-Dual Interior Point Filter Line Search Algorithm for Large-Scale Nonlinear Programming, Math- ematical Programming 106(1), 25-57, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Xie, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Wright, Complexity of Proximal Augmented Lagrangian for Nonconvex Optimization with Nonlinear Equality Constraints, Journal of Scientific Computing, 86, 38, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Yashtini, Convergence and rate analysis of a proximal linearized ADMM for nonconvex nonsmooth optimization, Journal of Global Optimization, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Zhang, Nearly unbiased variable selection under minimax concave penalty, Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Stat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 38, 894–942, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Zhang and Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Luo, A proximal alternating direction method of multiplier for lin- early constrained nonconvex minimization, SIAM J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' Optim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} +page_content=' 30(3), 2272–2302, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/g9E_T4oBgHgl3EQf3Rw9/content/2301.08345v1.pdf'} diff --git a/gdE1T4oBgHgl3EQfzAU4/content/2301.03439v1.pdf b/gdE1T4oBgHgl3EQfzAU4/content/2301.03439v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..64b095f647e7bba9c1c1cf4977528079d6d49738 --- /dev/null +++ b/gdE1T4oBgHgl3EQfzAU4/content/2301.03439v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88d35a0ec2b960a0c0e2d73077abc6ca22309253e409344469cbd9d551fb1f97 +size 3654961 diff --git a/gdE1T4oBgHgl3EQfzAU4/vector_store/index.faiss b/gdE1T4oBgHgl3EQfzAU4/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..4073cc42b7b6dfc7238ed88fa4319d972305d740 --- /dev/null +++ b/gdE1T4oBgHgl3EQfzAU4/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74f9eb06e1b909d516572cf8982b67118a1e6d22975aa330dafbe5340d1e2fcb +size 2293805 diff --git a/gdE1T4oBgHgl3EQfzAU4/vector_store/index.pkl b/gdE1T4oBgHgl3EQfzAU4/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..7bf81f8602da5572428f6bc19d025c38f55fb770 --- /dev/null +++ b/gdE1T4oBgHgl3EQfzAU4/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ac18e54d3276a5a65a6f58da4956424eed66a38644d8b33a101da09bcbdcead +size 87996 diff --git a/gtAzT4oBgHgl3EQfMfvB/content/tmp_files/2301.01134v1.pdf.txt b/gtAzT4oBgHgl3EQfMfvB/content/tmp_files/2301.01134v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..8aaf1a6f23629c500f99d6527a189074ef54a20d --- /dev/null +++ b/gtAzT4oBgHgl3EQfMfvB/content/tmp_files/2301.01134v1.pdf.txt @@ -0,0 +1,995 @@ +Ring That Bell: A Corpus and Method for Multimodal Metaphor +Detection in Videos +Khalid Alnajjar1, Mika Hämäläinen1 and Shuo Zhang2 +1University of Helsinki, Finland +2Bose Corporation, USA +firstname.lastname@{helsinki.fi or bose.com} +Abstract +We present the first openly available multi- +modal metaphor annotated corpus. The corpus +consists of videos including audio and subti- +tles that have been annotated by experts. Fur- +thermore, we present a method for detecting +metaphors in the new dataset based on the +textual content of the videos. +The method +achieves a high F1-score (62%) for metaphor- +ical labels. +We also experiment with other +modalities and multimodal methods; however, +these methods did not out-perform the text- +based model. In our error analysis, we do iden- +tify that there are cases where video could help +in disambiguating metaphors, however, the vi- +sual cues are too subtle for our model to cap- +ture. The data is available on Zenodo. +1 +Introduction +Figurative language is a challenging topic for com- +putational modeling as the meaning of a figurative +expression is non-compositional and typically very +context dependent (see Roberts and Kreuz 1994). +Metaphor is one of the most important figures of +language; it is constantly used in every day lan- +guage (Steen et al., 2010a) to draw comparisons +or to express something difficult and foreign in +more familiar terms. Metaphors can be conven- +tional (Traugott, 1985) and they are often found in +idioms, but at the same time metaphors are used to +create something new (see Kantokorpi et al. 1990). +Given its ubiquitous presence, understanding +metaphors is integral in achieving true natural +language understanding (NLU) in the real world. +Without their successful interpretation, our mod- +els are bound to make mistakes whenever any- +thing is expressed in an indirect or creative fashion. +Metaphors are often very contextual and their suc- +cessful detection and interpretation requires a wide +range of contextual cues that would be captured +in audio (e.g., prosody) and video (e.g., gestures +and actions). Therefore, we believe a multimodal +dataset is a great contribution to metaphor research +within and outside of the field of NLP. +Two important parts of a metaphor are a tenor +and a vehicle (see Richards 1936). For example, in +the metaphor life is a journey, life is the tenor and +journey is the vehicle. How metaphors essentially +operate is that a vehicle is used to give some of its +attributes to the tenor. In the case above, journeys +are long and full of adventure, which means that +these properties are attributed to life in an indirect +fashion. The meaning of a metaphor is never literal +nor compositional, but rather calls for interpretation +on the level of pragmatics (see Rosales Sequeiros +2016). +Meanwhile, multimodality is becoming increas- +ingly important for many tasks (see Castellucci +et al. 2020; Mogadala et al. 2020; Declerk et al. +2020). We believe the availability of multimodal +datasets for a variety of NLP tasks is lacking, and +we hope to contribute to the community with our +multimodal metaphor dataset. +In this paper, we present the first fully open ex- +pert annotated multimodal dataset for metaphor +detection1. In addition, we experiment with uni- +modal and multimodal methods for metaphor detec- +tion. Our results indicate that the text-based model +achieved the best performance. We discuss the re- +sults of our experiments and conduct an extensive +error analysis to shed light on what was learned +successfully by the model and its shortcomings. +Using CC BY licensed videos in our corpus has +been the primary design principle of our data col- +lection so that we can release our corpus without +any restrictions in its entirety. This, we believe, +is more useful for research purposes than a cor- +pus consisting of short video clips to compile with +copyright laws such as the fair use law in the US. +1https://doi.org/10.5281/zenodo.7217991 +arXiv:2301.01134v1 [cs.MM] 15 Dec 2022 + +2 +Related Work +Metaphors have, thus far, been computationally de- +tected using only text. In this section, we describe +some of the recent approaches for textual metaphor +detection, the corpora used to achieve that and +some of the multimodal research conducted on +NLP tasks other than metaphor detection. There +are several takes on metaphor interpretation (Xiao +et al., 2016; Rai et al., 2019; Bar et al., 2020) and +generation (Hämäläinen, 2018; Terai and Sugyo, +2019; Zheng et al., 2019), but we do not describe +them in detail as interpretation is a very different +problem. +There are two corpora currently used for +metaphor detection, +VU Amsterdam (VUA) +Metaphor Corpus (Steen et al., 2010b) and Cor- +pus of Non-Native Written English Annotated for +Metaphor (Beigman Klebanov et al., 2018). Unlike +our corpus, both of these datasets contain textual +modality only. +For textual metaphor detection, Gao et al. (2018) +has used a bi-directional LSTM (long short-term +memory) based model with ELMo embeddings. +Similarly, Liu et al. (2020) have used a bi-LSTM +model with BERT and XLNet for the same task. +Not unlike the previous approaches, Dankers et al. +(2020) has also applied bi-LSTM models compar- +ing ELMo and GloVe embeddings to BERT embed- +dings with global and hierarchical attention models. +Traditional machine learning methods, Logistic +Regression, Linear SVC (Support Vector Classi- +fication) and Random Forest Classifier, have been +used recently with feature engineering to detect +metaphors (Wan et al., 2020). In DeepMet, pro- +posed by Su et al. (2020), a siamese neural network +have been utilized, where textual RoBERTa (Liu +et al., 2019) embeddings are computed from the +context, the token in question and its part-of-speech +and fine-grained part-of-speech. +DeepMet was +the best performing solution for detecting textual +metaphors in the VUA dataset, based on a recent +shared task (Leong et al., 2020). +There are several recent works on multimodal +detection of a variety of linguistic phenomena. For +example, SVMs (Support Vector Machines) with +word embeddings and feature extraction have been +used for multimodal sarcasm detection (Castro +et al., 2019; Alnajjar and Hämäläinen, 2021). Mit- +tal et al. (2020) uses GloVe embeddings, features +extracted from audio and facial recognition system +output to predict emotion in a multimodal dataset. +These multimodal features are fused using a mem- +ory fusion network (MFN) (Zadeh et al., 2018). +Similarly, Li et al. (2021) detect emotion in a mul- +timodal dataset by modeling the problem from the +point of view of the quantum theory. While the +field has seen increasing research on multimodal +NLP (Tsai et al., 2019; Mai et al., 2020; Sahu and +Vechtomova, 2021), no data or model has been +proposed for multimodal metaphor detection. +3 +Our Metaphor Corpus +In this section, we present our video, audio and +textual corpus of manually annotated metaphorical +language. Our selection of the video clips includes +only CC-BY licensed videos on YouTube that have +human authored closed captions in English. The +content of the videos presents mainly real people +talking, which rules out animations and video game +streams. The availability of human authored closed +captioning is important as it speeds up our anno- +tation time and provides us with subtitles that are +already aligned with video and audio. The CC- +BY license was an important selection criterion +because it makes it possible for us to release the +dataset openly. +We used the filters provided by YouTube to +limit our search to videos that were marked as +CC-BY and had closed captioning. However, the +YouTube filter does not distinguish between auto- +matically generated closed captioning and a human +authored one. Fortunately, it is relatively easy to +tell these two apart from each other. Automated +closed captioning tends to appear one word at a +time, whereas human authored closed captioning is +visualized more like traditional subtitles. These cri- +teria greatly reduced the number of eligible videos +to include in our corpus. Apart from these criteria, +we also filtered videos with sensitive and offensive +languages. No further restrictions have been ex- +plicitly placed on the genres or types of videos, as +we do not want to introduce biases for which types +of contents are more likely to contain metaphors. +Therefore, the availability of the metaphors natu- +rally occurring in the corpus is the result of the +ubiquity of the metaphor in everyday language use. +All Youtube queries were conducted in incognito +mode to avoid biased YouTube suggestions based +on our viewing habits. +Figure 1 shows real examples from our corpus +where video can be useful in detecting metaphors. +On the left, the woman wearing a gray shirt is + +sentence +that you can use to really up your game +because while a quick fix can be appetizing and appealing +That’s the street address for your website +you’re ready to give it a shot +Table 1: Example of the annotations for the metaphor detection corpus. +Figure 1: Metaphors made visible in the video through +gestures. +talking about sprinkling keywords and showing a +sprinkling gesture. On the right, the woman wear- +ing the wine red shirt says ring that bell and shows +a bell ringing gesture. +Our corpus consists of 27 YouTube videos with +a total duration of 3 hours, 53 minutes and 47 sec- +onds of video. For comparison, a recently released +multimodal dataset for sarcasm detection (Castro +et al., 2019) has the duration of 3 hours, 40 minutes +and 47 seconds. The videos belong mostly to a +start-up domain and many of them deal with issues +of online visibility for a start-up company. This +domain was a consequence of our selection criteria +for videos. It turns out that YouTube has plenty +of high-quality human close-captioned videos re- +leased under the CC-BY license that relate to this +particular domain. +Our corpus provides linguistics researchers with +the ability to study the use of metaphor in a mul- +timodal setting, something that has gained atten- +tion in their field of science as well (Müller and +Cienki, 2009). This can, indeed, foster a wider +interdisciplinary collaboration leading to a deeper +understanding of the phenomenon. +3.1 +Annotation +Two expert annotators went through the video files +and annotated metaphors by surrounding them with +v tags for vehicles and t tags for tenors. +The +use of experts is motivated by the fact that previ- +ous research has found that non-expert annotators +struggle with metaphors (Hämäläinen and Alnajjar, +2019). +The annotators followed a simple procedure in +annotating the data: +• Is the meaning literal? +• If the meaning of the word is abstract, is it a +dictionary meaning? +• Does the potential metaphor express prag- +matic insincerity? +• If the answer to all of the questions is no, +annotate it as a metaphor. +In other words, if the meaning of a word or a +phrase is not literal, it is annotated as a metaphor. +However, just the mere fact of a word being used +in an abstract way is not enough to mark it as +metaphorical. For example, in the sentence it is +tied to revenue, “tied” is not tagged as a metaphor +just because it is used in a more abstract sense than +the typical concrete sense of tying one’s shoes, for +example. If the abstract meaning of a word ap- +pears in a dictionary, the word is not considered +metaphorical. However, conventional metaphors +that consist of multiple words, and are thus idioms, +are tagged as metaphors. We do not make a distinc- +tion between metaphors and similes. +Pragmatic insincerity (see Grice 1975) is a phe- +nomenon related to sarcasm as one of its precon- +ditions (see Kumon-Nakamura et al. 1995). There +is a certain overlap between metaphors and sarcas- +tic expressions in the sense that both use words in +their non-literal meaning. In order to ensure that +we do not mix these two notions with each other, +it is important to avoid annotating pragmatically +insincere expressions as metaphorical. +Table 1 shows an example of annotations. The +annotations were done directly in the subtitles The +utterances are time stamped and aligned with the +video. In the table, tenors are indicated with +and vehicles with . For deictic tenors, an r +attribute is provided to resolve the deixis by indi- +cating the actual tenor that has appeared earlier in +the conversation. In the examples, game is used +metaphorically to talk about marketing, quick fix is + +TheJourney +you should probably sprinkle those in. +and ring that bellcalled appetizing as though it was something edible +and domain name is contrasted to a physical street +address by direct comparison. Give it a shot is a +conventional metaphor. +All in all, after multiple annotation iterations, +the dataset consists of 304 vehicles and 67 tenors. +This totals to 371 metaphorical expressions. They +vary in length: the shortest tenor is one word, such +as it, while the longest tenor is several words the +discovery of those five noble gases to illuminate +like that. The same goes for vehicles where their +length varies form one word such as dive to multi- +ple words: the history of the internet itself. On a +token level, we have 672 vehicle tokens and 113 +tenor tokens, so altogether 785 metaphorical to- +kens. +In total, 6% of the expressions in the corpus are +metaphorical. While this percentage might appear +low, it is natural and more representative of the real +usage of metaphors in typical conversations which +makes this corpus suitable for building metaphor +detection models applicable for real-world scenar- +ios. +Around 55% of the vehicles are conventional +metaphors and 45% are novel metaphors. How- +ever, it is fairly common that same words appear in +the corpus in a metaphorical and non-metaphorical +sense. In our corpus, there are two videos that +deal with actual cooking, in which many food- +related metaphors appear non-metaphorically, such +as sprinkle those in, said metaphorically about key- +words and a little sprinkle, said non-metaphorically +about sugar. Another example is the use of house +non-metaphorically as in come pick it up at my +house and metaphorically as in think of hosting as +your house, where a metaphorical connection is +drawn between hosting and a house. +3.2 +Data preparation +As YouTube serves files in several different formats +such as webm, mkv and mp4 the first step is to use +FFmpeg2 to convert all videos into mp4 format. We +also use the same tool to clip the video files into +sentence-length clips based on the time stamps in +the subtitles and extract their audio into wav files. +This process yielded 6,565 video and audio clips +that are aligned with text. +We split the datset randomly so that 70% of sen- +tences that contain metaphors and 70% of sentences +that don’t contain any metaphors are used for train- +2https://ffmpeg.org/ +ing, 15 % of both types of sentences for validation +and 15% of both for testing. This way we ensure +that both metaphorical and non-metaphorical sen- +tences are divided proportionally with the same +ratios. These splits are used for all the models. +4 +Metaphor Detection +We experiment with uni- and multi-modal models +for metaphor detection. In this section, we describe +the preprocessing steps applied and the experimen- +tal setups conducted. +4.1 +Preprocessing +For each modality, we make use of the latest ad- +vances in neural network models to capture impor- +tant features that have achieved state-of-the-art re- +sults in various NLP tasks. As metaphor detection +has been conducted solely based on text, we fol- +low the DeepMet approach by Su et al. (2020) and +process the entire textual content using spaCy (Hon- +nibal et al., 2020) to tokenize it and acquire Uni- +versal Dependencies style syntactic trees (Nivre +et al., 2020) and Penn Treebank parts-of-speech +tags (Santorini, 1990). Similarly to the original ap- +proach, all of our textual models predict metaphors +at the token level given the context surrounding it +and its POS tags as input. +We resample the audio to 16kHz. Audio features +are extracted using Wav2Vec2FeatureExtractor pro- +vided by the Transoformers Python library (Wolf +et al., 2020). +Video features are obtained by taking equally- +distributed 16 frames from a clip and then resize +them into 128x171, followed by normalization and +center cropping to 112x112. +4.2 +Textual model +We train two text-only models, both follow the +architecture and approach of DeepMet where we +obtain textual embeddings using RoBERTa (Liu +et al., 2019) and feed them into two transformer +encoding layers which are then combined by apply- +ing global average pooling and concatenation. A +dense fully-connected layer takes in the combined +output of both encoders and predicts whether the +token is metaphorical (c.f., Su et al. 2020 for more +details). +In our first textual model, we train the model +using our corpus, whereas in the second one we +train it using VUA corpus (with a learning rate of + +0.00001, akin the original paper) and later fine-tune +it using our corpus. +4.3 +Audio model +We extend and fine-tune Facebook’s pretrained mul- +tilingual XLSR-Wav2Vec2 large model (Baevski +et al., 2020). The model is trained on Multilingual +LibriSpeech (Pratap et al., 2020), CommonVoice +(Ardila et al., 2020) and Babel (Roach et al., 1996) +for speech recognition. We employ this model to +encode speech into vector representations from raw +audio. +We replace the classification layer of the original +model with a dense fully-connected layer that pro- +duces two outputs, one for each label. Unlike the +textual model, here we classify whether the entire +spoken expression contains a metaphor or not (i.e., +not on a word level). +4.4 +Video model +For our video unimodal model, we incorporate a +pretrained model for human action detection. The +model is based on the 18 layer deep R(2+1)D net- +work (Tran et al., 2018) and it is trained on the +Kinetics-400 (Zisserman et al., 2017) dataset. The +intuition behind using this model is that it was +able to detect actions (e.g., playing organ), ges- +tures (e.g., pointing) and movements (e.g., waving). +Realizing such information is crucial in understand- +ing the context, and would provide further cues for +detecting metaphors. +Similar to the audio model, we substitute the +original classification layer with a fully connected +layer and fine-tune the pretrained model to predict +whether a scene is metaphorical or not. +4.5 +Multimodal metaphor detection +We test out three multimodal metaphor detection +models; 1) text and audio, 2) text and video and +3) text, audio and video. The textual model is the +fine-tuned model using the VUA corpus and our +textual corpus. +In all of the models, the final classification layer +of their sub-models are removed. Unimodal models +are combined by concatenating the weights of their +last layer, which are then fed to a classification +layer. +4.6 +Common configuration +All of the models described above share common +configurations, unless we explicitly indicate other- +wise. Prior to the last classification layer of all of +our mono- and multimodal models, we introduce a +dropout layer (Srivastava et al., 2014) (with a prob- +ability of 20%) to accelerate training, and reduce +internal covariate shift and overfitting. +We use the cross entropy loss function along +with Adam optimizer (Kingma and Ba, 2014; +Loshchilov and Hutter, 2019) to update the weights +and train the models. All the fine-tuned models are +trained with a learning rate of 0.0001 and for 3 full +epochs. +5 +Results +In this section, we follow the evaluation metrics +commonly used for the metaphor detection task by +reporting the precision, recall and F1 scores for the +metaphorical label. +Regarding the textual models, we report three +sets of results, which are for the models trained +on: 1) VUA corpus, 2) our corpus and 3) both +the VUA and our corpus. All the models predict +metaphoricity on the token level. To ensure that +our implementation of the DeepMet approach is +correct, we tested the first model on the VUA test +dataset of the metaphor detection shared task and +achieved an F1-score of 0.68 and 0.73 on all POS +and verb subsets of the data, respectively. These +results are relatively close to the results reported by +the authors. +Table 2 shows the classification results of all +three models on the test set. +The test set con- +tained 90 metaphorical tokens and 6,961 non- +metaphorical tokens. The results indicate that the +textual model trained solely on the VUA dataset +performed poorly on our test set. +In compari- +son, training the model using our metaphor corpus +only resulted in a great increase of correct predic- +tions. Nonetheless, combining both corpora by +fine-tuning the first model with our corpus pro- +duced the winning model, which managed to spot +76% of the metaphorical tokens correctly. +We believe that the huge differences between the +first and second textual models, despite the larger +size of VUA’s training dataset, are due to the dif- +ferences in domains. The VUA corpus contains +academic texts, conversation, fiction, and news +texts, whereas our corpus is dominated by con- +versations on the web and start-ups. It is evident +that by exposing the model to general domains (i.e., +VUA’s corpus) and, thereafter, concentrating it on +the start-up domain, the model was able to identify +the highest number of metaphorical usages. + +Trained on +Precision +Recall +F1-score +VUA +0.04 +0.33 +0.07 +Ours +0.38 +0.63 +0.47 +VUA + Ours +0.53 +0.76 +0.62 +Table +2: +Classification +results +of +the +textual +monomodal models on the test set of our corpus, +for the metaphorical label. +Results from the other models (unimodal or mul- +timodal) that involving audio and video showed +that adding these modalities actually did not help +improving the model - rather, they are detrimental +to the model performance on metaphor detection. +We extend two possible explanations for this failure. +First, it is possible that because the visual and audio +cues of metaphor are subtle, these models failed to +learn from such a small amount of annotated data. +Second, it is unclear that the specific models we +are using for audio and video modalities encode +the information relevant for the metaphor detec- +tion task. For instance, whereas it is impossible to +completely disentangle what exactly the Wav2Vec +model is encoding, we can conjecture that it en- +codes information about phoneme identity consid- +ering it is optimized for the speech recognition task. +Therefore, it may not be entirely surprising that the +Wav2Vec encoding is not useful for the metaphor +detection task because it is adding redundant or +irrelevant information to the model. It is our future +work (or the future work for the community who +utilizes this dataset) to refine our understanding +of the multimodal encoding for the metaphor de- +tection task (for instance, employing a model that +more directly encodes information about speech +prosody from the audio). +5.1 +Error analysis +When looking at the results of the text only model, +we can see that the model identifies metaphors cor- +rectly as metaphors more often than not. There +are some metaphorical tokens in metaphors con- +sisting of multiple words that get classified wrong, +for example, in You could think of hosting as your +house, the tenor hosting and the determinant your +of the metaphorical word house are not identified +as metaphorical, while house is correctly identified. +Another example is the conventional metaphor toot +their own horn, where all other words except for +own are correctly identified as metaphorical. +There are also a fewer number of cases where all +words get identified wrongly as non-metaphorical, +for example, the model did not predict any +metaphorical tokens in It’s where you live, while +in reality it is the tenor and where you live is the +vehicle. Also, individual tenors where the vehicle +comes later get often not recognized such as in Yes, +malware you could think of like, where malware +is the tenor for a vehicle that appears later in the +dialog. +When the tenor and the vehicle co-exist nearby, +the model can get all metaphorical tokens right +such as in It’s kinda like real estate right? where +both the tenor it and the vehicle real estate are cor- +rectly identified. Also many tenorless expressions +are fully recognized correctly as metaphorical, such +as Spreadin’ the love. +There were plenty of cases (61) where the model +predicted a metaphor tag for a token while there +was no metaphor. Curiously, prepositions were of- +ten tagged metaphorical, such as to in ring that bell +to see these episodes first. The actual metaphorical +part ring that bell ends before the preposition to +that has a non-metaphorical meaning in order to. +We can also see that the model was indeed fooled +by cooking terms that were used both metaphor- +ically and non-metaphorically. +In Yeah a little +sprinkle, both a and sprinkle were classified as +metaphors, while the context was about sprinkling +sugar. Another similar case was there’s five noble +gases that illuminate, where noble gases and illumi- +nate were erroneously classified to be metaphorical. +This was clearly due to the tenor in the corpus: the +discovery of those five noble gases to illuminate +like that that contained similar words. It is evident +that the model relies on word similarities more than +reaching to a higher pragmatic representation of the +phenomenon, however, this is not an unexpected +behavior from a machine learning model. +There are also cases where the model detects a +metaphor, that could theoretically be a metaphor, +but is not because of the way it was used in the cor- +pus. For example, the model predicts Give it a go +as metaphorical in the expression button, "Give it a +go.", where people are talking about a button with +a particular text rather than using the expression +metaphorically. Another such an example is flying +in (money flying). Such an expression might be +used metaphorically, but in this case this was a note +for the hearing impaired as money was actually +flying on the video. + +6 +Discussion and Conclusions +In this work, we have only focused on metaphor +as a strictly linguistic phenomenon and we have +built a multimodal dataset where these linguistic +metaphors have been tagged in terms of tenors and +vehicles. However, it is apparent that metaphor is +a phenomenon that occurs on a higher level of our +cognitive capacities than mere language. There are +several cases in our corpus, where we can evidence +the existence of a metaphor but it is never expressed +verbally. For example in Figure 2, money flying +cannot be a metaphor when inspected purely from +the point of view of language and its relation to +the video when money is actually flying in the +scene. However, it is a metaphor on a higher level +in the sense that the entire scene where money was +flying was to indicate someone becoming rich. In +other words, stating a fact that is happening is not +metaphorical if the fact is literally taking place, +however the fact itself might be metaphorical. +Figure 2: Money actually flying on the video. +At the same time, as evidenced by our error anal- +ysis, there are certainly cases where video modality +could help in disambiguating whether something +is said metaphorically or not. For instance, talk- +ing about sprinkling in a kitchen environment (see +Figure 3) is a very strong sign that the word is po- +tentially non-metaphorical. Integrating these weak +cues into a multimodal system is, however, not an +easy task given that the current methods for video +processing are limited in their coverage. +Therefore, in the future, it would be useful to +annotate metaphors also in the other modalities. +Money flying can be a visual metaphor, and so can +a sound effect, and they can exist independently +from each other in different modalities. Perhaps the +reason why our multimodal attempts failed was that +metaphor can be independent of the other modali- +ties. Producing such a dataset where these modal +specific metaphors are also annotated for video and +Figure 3: Sprinkling used in a kitchen in reference to +sugar. +audio is definitely a huge undertaking that requires +research in its own right. +It is clear that our model can detect metaphors +correctly, but also the mistakes it makes highlight +that despite using a large RoBERTa model, the +meaning representation the model has cannot reach +to such a nuanced level as to confidently detect +metaphors. Metaphor is a figurative device that can- +not be explained by semantics, but rather requires +pragmatic inspection. It is not clear based on our re- +search and other contemporary approaches whether +the current word or sentence embedding models are +sufficient to navigate in the depths of pragmatics +and subjective interpretation in any other way than +learning some irrelevant co-occurring phenomena +from a biased corpus. At the same time there is no +such thing as an unbiased corpus, either, given that +bias (and mostly heuristics causing it) is a funda- +mental part of our cognition as human beings. +In this paper, we have presented a new open and +multimodal dataset for metaphor detection. Be- +cause we have focused strictly on CC-BY licensed +videos, we can make the entire dataset available on +Zenodo. In our current work, we have not taken +the context widely into account when predicting +metaphoricity, but rather resorted to a very local +context. The fact that the videos can be published +in full length makes it possible for any future work +to explore different ways of including contextual +cues freely. +Acknowledgments +This work was partially financed by the Society of +Swedish Literature in Finland with funding from +Enhancing Conversational AI with Computational +Creativity, and by the Ella and Georg Ehrnrooth +Foundation for Modelling Conversational Artificial +Intelligence with Intent and Creativity. + +The Journey +JOURNEY +(money flying)Yeah a little sprinkleReferences +Khalid Alnajjar and Mika Hämäläinen. 2021. +¡Qué +maravilla! multimodal sarcasm detection in Span- +ish: +a dataset and a baseline. +arXiv preprint +arXiv:2105.05542. +Rosana Ardila, Megan Branson, Kelly Davis, Michael +Kohler, Josh Meyer, Michael Henretty, Reuben +Morais, Lindsay Saunders, Francis Tyers, and Gre- +gor Weber. 2020. +Common voice: A massively- +multilingual speech corpus. In Proceedings of the +12th Language Resources and Evaluation Confer- +ence, pages 4218–4222, Marseille, France. Euro- +pean Language Resources Association. +Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, +and Michael Auli. 2020. +wav2vec 2.0: A frame- +work for self-supervised learning of speech represen- +tations. Advances in Neural Information Processing +Systems, 33. +Kfir Bar, Nachum Dershowitz, and Lena Dankin. 2020. +Automatic metaphor interpretation using word em- +beddings. arXiv preprint arXiv:2010.02665. +Beata Beigman Klebanov, Chee Wee (Ben) Leong, and +Michael Flor. 2018. A corpus of non-native written +English annotated for metaphor. In Proceedings of +the 2018 Conference of the North American Chap- +ter of the Association for Computational Linguistics: +Human Language Technologies, Volume 2 (Short Pa- +pers), pages 86–91, New Orleans, Louisiana. Asso- +ciation for Computational Linguistics. +Giuseppe Castellucci, Simone Filice, Soujanya Poria, +Erik Cambria, and Lucia Specia, editors. 2020. Pro- +ceedings of the First International Workshop on Nat- +ural Language Processing Beyond Text. Association +for Computational Linguistics, Online. +Santiago Castro, Devamanyu Hazarika, Verónica Pérez- +Rosas, Roger Zimmermann, Rada Mihalcea, and +Soujanya Poria. 2019. Towards multimodal sarcasm +detection (an _obviously_ perfect paper). +In Pro- +ceedings of the 57th Annual Meeting of the Associa- +tion for Computational Linguistics (Volume 1: Long +Papers), Florence, Italy. Association for Computa- +tional Linguistics. +Verna Dankers, +Karan Malhotra, +Gaurav Kudva, +Volodymyr Medentsiy, and Ekaterina Shutova. 2020. +Being neighbourly: Neural metaphor identification +in discourse. In Proceedings of the Second Work- +shop on Figurative Language Processing, pages +227–234, Online. Association for Computational +Linguistics. +Thierry Declerk, Itziar Gonzalez-Dios, and German +Rigau, editors. 2020. Proceedings of the LREC 2020 +Workshop on Multimodal Wordnets (MMW2020). +The European Language Resources Association +(ELRA), Marseille, France. +Ge Gao, Eunsol Choi, Yejin Choi, and Luke Zettle- +moyer. 2018. Neural metaphor detection in context. +In Proceedings of the 2018 Conference on Empiri- +cal Methods in Natural Language Processing, pages +607–613, Brussels, Belgium. Association for Com- +putational Linguistics. +Herbert P Grice. 1975. +Logic and conversation. +In +Speech acts, pages 41–58. Brill. +Mika Hämäläinen. 2018. +Harnessing nlg to create +finnish poetry automatically. +In Proceedings of +the ninth international conference on computational +creativity. Association for Computational Creativity +(ACC). +Mika Hämäläinen and Khalid Alnajjar. 2019. +Let’s +FACE it. Finnish poetry generation with aesthetics +and framing. +In Proceedings of the 12th Interna- +tional Conference on Natural Language Generation, +pages 290–300, Tokyo, Japan. Association for Com- +putational Linguistics. +Matthew Honnibal, Ines Montani, Sofie Van Lan- +deghem, +and Adriane Boyd. 2020. +spaCy: +Industrial-strength Natural Language Processing in +Python. +Mervi Kantokorpi, Lyytikäinen Pirjo, and Viikari Auli. +1990. Runousopin perusteet. Gaudeamus. +Diederik P Kingma and Jimmy Ba. 2014. Adam: A +method for stochastic optimization. arXiv preprint +arXiv:1412.6980. +Sachi Kumon-Nakamura, Sam Glucksberg, and Mary +Brown. 1995. How about another piece of pie: The +allusional pretense theory of discourse irony. Jour- +nal of Experimental Psychology: General, 124(1):3. +Chee Wee (Ben) Leong, Beata Beigman Klebanov, +Chris Hamill, Egon Stemle, Rutuja Ubale, and Xi- +anyang Chen. 2020. +A report on the 2020 VUA +and TOEFL metaphor detection shared task. In Pro- +ceedings of the Second Workshop on Figurative Lan- +guage Processing, pages 18–29, Online. Association +for Computational Linguistics. +Qiuchi Li, Dimitris Gkoumas, Christina Lioma, and +Massimo Melucci. 2021. Quantum-inspired multi- +modal fusion for video sentiment analysis. Informa- +tion Fusion, 65:58–71. +Jerry Liu, Nathan O’Hara, Alexander Rubin, Rachel +Draelos, and Cynthia Rudin. 2020. Metaphor detec- +tion using contextual word embeddings from trans- +formers. +In Proceedings of the Second Workshop +on Figurative Language Processing, pages 250–255, +Online. Association for Computational Linguistics. +Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Man- +dar Joshi, Danqi Chen, Omer Levy, Mike Lewis, +Luke Zettlemoyer, and Veselin Stoyanov. 2019. +Roberta: A robustly optimized bert pretraining ap- +proach. +Ilya Loshchilov and Frank Hutter. 2019. +Decoupled +weight decay regularization. In International Con- +ference on Learning Representations. + +Sijie Mai, Haifeng Hu, and Songlong Xing. 2020. +Modality to modality translation: An adversarial rep- +resentation learning and graph fusion network for +multimodal fusion. In Proceedings of the AAAI Con- +ference on Artificial Intelligence, volume 34, pages +164–172. +Trisha Mittal, Uttaran Bhattacharya, Rohan Chandra, +Aniket Bera, and Dinesh Manocha. 2020. +M3er: +Multiplicative multimodal emotion recognition us- +ing facial, textual, and speech cues. In Proceedings +of the AAAI Conference on Artificial Intelligence, +volume 34, pages 1359–1367. +Aditya Mogadala, Sandro Pezzelle, Dietrich Klakow, +Marie-Francine Moens, and Zeynep Akata, edi- +tors. 2020. +Proceedings of the Second Workshop +on Beyond Vision and LANguage: +inTEgrating +Real-world kNowledge (LANTERN). Association for +Computational Linguistics, Barcelona, Spain. +Cornelia Müller and Alan Cienki. 2009. Chapter 13. +Words, gestures, and beyond: Forms of multimodal +metaphor in the use of spoken language, pages 297– +328. De Gruyter Mouton. +Joakim Nivre, Marie-Catherine de Marneffe, Filip Gin- +ter, Jan Hajic, Christopher D Manning, Sampo +Pyysalo, Sebastian Schuster, Francis Tyers, and +Daniel Zeman. 2020. +Universal dependencies v2: +An evergrowing multilingual treebank collection. In +Proceedings of The 12th Language Resources and +Evaluation Conference, pages 4034–4043. +Vineel Pratap, Qiantong Xu, Anuroop Sriram, Gabriel +Synnaeve, and Ronan Collobert. 2020. +Mls: A +large-scale multilingual dataset for speech research. +ArXiv, abs/2012.03411. +Sunny Rai, Shampa Chakraverty, Devendra K Tayal, +Divyanshu Sharma, and Ayush Garg. 2019. Under- +standing metaphors using emotions. New Genera- +tion Computing, 37(1):5–27. +Ivor Armstrong Richards. 1936. +The Philosophy of +Rhetoric. Oxford University Press, London, United +Kingdom. +Peter Roach, +Simon Arnfield, +W Barry, +J Bal- +tova, Marian Boldea, Adrian Fourcin, W Gonet, +Ryszard Gubrynowicz, E Hallum, Lori Lamel, et al. +1996. Babel: An eastern european multi-language +database. +In Proceeding of Fourth International +Conference on Spoken Language Processing. IC- +SLP’96, volume 3, pages 1892–1893. IEEE. +Richard M Roberts and Roger J Kreuz. 1994. +Why +do people use figurative language? +Psychological +science, 5(3):159–163. +Xose Rosales Sequeiros. 2016. +Metaphor: +Prag- +matics, relevance and cognition. +English Studies, +97(6):656–677. +Gaurav Sahu and Olga Vechtomova. 2021. Adaptive +fusion techniques for multimodal data. In Proceed- +ings of the 16th Conference of the European Chap- +ter of the Association for Computational Linguistics: +Main Volume, pages 3156–3166, Online. Associa- +tion for Computational Linguistics. +Beatrice Santorini. 1990. Part-of-speech tagging guide- +lines for the penn treebank project (3rd revision). +Technical Reports (CIS), page 570. +Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, +Ilya Sutskever, and Ruslan Salakhutdinov. 2014. +Dropout: A simple way to prevent neural networks +from overfitting. Journal of Machine Learning Re- +search, 15(56):1929–1958. +GJ Steen, AG Dorst, JB Herrmann, AA Kaal, and +T Krennmayr. 2010a. Metaphor in usage. Cognitive +Linguistics, 21(4):757–788. +GJ Steen, AG Dorst, JB Herrmann, AA Kaal, T Krenn- +mayr, and T Pasma. 2010b. A method for linguistic +metaphor identification. from mip to mipvu. Con- +verging Evidence in Language and Communication +Research, (14). +Chuandong Su, Fumiyo Fukumoto, Xiaoxi Huang, Jiyi +Li, Rongbo Wang, and Zhiqun Chen. 2020. Deep- +Met: A reading comprehension paradigm for token- +level metaphor detection. In Proceedings of the Sec- +ond Workshop on Figurative Language Processing, +pages 30–39, Online. Association for Computational +Linguistics. +Asuka Terai and Taiki Sugyo. 2019. Construction of +a corpus-based metaphor generation support system +built on japanese literature. In 2019 IEEE 11th Inter- +national Workshop on Computational Intelligence +and Applications (IWCIA), pages 1–6. IEEE. +Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, +Yann LeCun, and Manohar Paluri. 2018. A closer +look at spatiotemporal convolutions for action recog- +nition. In 2018 IEEE/CVF Conference on Computer +Vision and Pattern Recognition, pages 6450–6459. +Elizabeth Closs Traugott. 1985. +‘conventional’and +‘dead’metaphors +revisited. +The +ubiquity +of +metaphor: +Metaphor in language and thought, +pages 17–56. +Yao-Hung Hubert Tsai, Shaojie Bai, Paul Pu Liang, +J Zico Kolter, Louis-Philippe Morency, and Rus- +lan Salakhutdinov. 2019. +Multimodal transformer +for unaligned multimodal language sequences. In +Proceedings of the conference. Association for Com- +putational Linguistics. Meeting, volume 2019, page +6558. NIH Public Access. +Mingyu Wan, Kathleen Ahrens, Emmanuele Chersoni, +Menghan Jiang, Qi Su, Rong Xiang, and Chu-Ren +Huang. 2020. Using conceptual norms for metaphor +detection. In Proceedings of the Second Workshop +on Figurative Language Processing, pages 104–109, +Online. Association for Computational Linguistics. + +Thomas Wolf, Lysandre Debut, Victor Sanh, Julien +Chaumond, Clement Delangue, Anthony Moi, Pier- +ric Cistac, Tim Rault, Rémi Louf, Morgan Funtow- +icz, Joe Davison, Sam Shleifer, Patrick von Platen, +Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, +Teven Le Scao, Sylvain Gugger, Mariama Drame, +Quentin Lhoest, and Alexander M. Rush. 2020. +Transformers: State-of-the-art natural language pro- +cessing. In Proceedings of the 2020 Conference on +Empirical Methods in Natural Language Processing: +System Demonstrations, pages 38–45, Online. Asso- +ciation for Computational Linguistics. +Ping +Xiao, +Khalid +Alnajjar, +Mark +Granroth- +Wilding, Kat Agres, and Hannu Toivonen. 2016. +Meta4meaning: Automatic metaphor interpretation +using corpus-derived word associations. +In Pro- +ceedings of the Seventh International Conference +on Computational Creativity. Sony CSL Paris. +Amir Zadeh, Paul Pu Liang, Navonil Mazumder, +Soujanya Poria, Erik Cambria, and Louis-Philippe +Morency. 2018. Memory fusion network for multi- +view sequential learning. Proceedings of the Thirty- +Second AAAI Conference on Artificial Intelligence. +Danning Zheng, Ruihua Song, Tianran Hu, Hao Fu, +and Jin Zhou. 2019. “love is as complex as math”: +Metaphor generation system for social chatbot. In +Workshop on Chinese Lexical Semantics, pages 337– +347. Springer. +Andrew Zisserman, Joao Carreira, Karen Simonyan, +Will Kay, Brian Zhang, Chloe Hillier, Sudheendra +Vijayanarasimhan, Fabio Viola, Tim Green, Trevor +Back, Paul Natsev, and Mustafa Suleyman. 2017. +The kinetics human action video dataset. + diff --git a/gtAzT4oBgHgl3EQfMfvB/content/tmp_files/load_file.txt b/gtAzT4oBgHgl3EQfMfvB/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..7e7d710d34cf446e1c8b41e7dff059562a5eadc6 --- /dev/null +++ b/gtAzT4oBgHgl3EQfMfvB/content/tmp_files/load_file.txt @@ -0,0 +1,529 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf,len=528 +page_content='Ring That Bell: A Corpus and Method for Multimodal Metaphor Detection in Videos Khalid Alnajjar1, Mika Hämäläinen1 and Shuo Zhang2 1University of Helsinki, Finland 2Bose Corporation, USA firstname.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='lastname@{helsinki.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='fi or bose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='com} Abstract We present the first openly available multi- modal metaphor annotated corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The corpus consists of videos including audio and subti- tles that have been annotated by experts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Fur- thermore, we present a method for detecting metaphors in the new dataset based on the textual content of the videos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The method achieves a high F1-score (62%) for metaphor- ical labels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We also experiment with other modalities and multimodal methods;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' however, these methods did not out-perform the text- based model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In our error analysis, we do iden- tify that there are cases where video could help in disambiguating metaphors, however, the vi- sual cues are too subtle for our model to cap- ture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The data is available on Zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1 Introduction Figurative language is a challenging topic for com- putational modeling as the meaning of a figurative expression is non-compositional and typically very context dependent (see Roberts and Kreuz 1994).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphor is one of the most important figures of language;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' it is constantly used in every day lan- guage (Steen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2010a) to draw comparisons or to express something difficult and foreign in more familiar terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphors can be conven- tional (Traugott, 1985) and they are often found in idioms, but at the same time metaphors are used to create something new (see Kantokorpi et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1990).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Given its ubiquitous presence, understanding metaphors is integral in achieving true natural language understanding (NLU) in the real world.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Without their successful interpretation, our mod- els are bound to make mistakes whenever any- thing is expressed in an indirect or creative fashion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphors are often very contextual and their suc- cessful detection and interpretation requires a wide range of contextual cues that would be captured in audio (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', prosody) and video (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', gestures and actions).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Therefore, we believe a multimodal dataset is a great contribution to metaphor research within and outside of the field of NLP.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Two important parts of a metaphor are a tenor and a vehicle (see Richards 1936).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For example, in the metaphor life is a journey, life is the tenor and journey is the vehicle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' How metaphors essentially operate is that a vehicle is used to give some of its attributes to the tenor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In the case above, journeys are long and full of adventure, which means that these properties are attributed to life in an indirect fashion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The meaning of a metaphor is never literal nor compositional, but rather calls for interpretation on the level of pragmatics (see Rosales Sequeiros 2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Meanwhile, multimodality is becoming increas- ingly important for many tasks (see Castellucci et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mogadala et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Declerk et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We believe the availability of multimodal datasets for a variety of NLP tasks is lacking, and we hope to contribute to the community with our multimodal metaphor dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In this paper, we present the first fully open ex- pert annotated multimodal dataset for metaphor detection1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In addition, we experiment with uni- modal and multimodal methods for metaphor detec- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Our results indicate that the text-based model achieved the best performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We discuss the re- sults of our experiments and conduct an extensive error analysis to shed light on what was learned successfully by the model and its shortcomings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Using CC BY licensed videos in our corpus has been the primary design principle of our data col- lection so that we can release our corpus without any restrictions in its entirety.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This, we believe, is more useful for research purposes than a cor- pus consisting of short video clips to compile with copyright laws such as the fair use law in the US.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='5281/zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='7217991 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='01134v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='MM] 15 Dec 2022 2 Related Work Metaphors have, thus far, been computationally de- tected using only text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In this section, we describe some of the recent approaches for textual metaphor detection, the corpora used to achieve that and some of the multimodal research conducted on NLP tasks other than metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are several takes on metaphor interpretation (Xiao et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2016;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Rai et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Bar et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020) and generation (Hämäläinen, 2018;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Terai and Sugyo, 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Zheng et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019), but we do not describe them in detail as interpretation is a very different problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are two corpora currently used for metaphor detection, VU Amsterdam (VUA) Metaphor Corpus (Steen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2010b) and Cor- pus of Non-Native Written English Annotated for Metaphor (Beigman Klebanov et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Unlike our corpus, both of these datasets contain textual modality only.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For textual metaphor detection, Gao et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2018) has used a bi-directional LSTM (long short-term memory) based model with ELMo embeddings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Similarly, Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2020) have used a bi-LSTM model with BERT and XLNet for the same task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Not unlike the previous approaches, Dankers et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2020) has also applied bi-LSTM models compar- ing ELMo and GloVe embeddings to BERT embed- dings with global and hierarchical attention models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Traditional machine learning methods, Logistic Regression, Linear SVC (Support Vector Classi- fication) and Random Forest Classifier, have been used recently with feature engineering to detect metaphors (Wan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In DeepMet, pro- posed by Su et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2020), a siamese neural network have been utilized, where textual RoBERTa (Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019) embeddings are computed from the context, the token in question and its part-of-speech and fine-grained part-of-speech.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' DeepMet was the best performing solution for detecting textual metaphors in the VUA dataset, based on a recent shared task (Leong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are several recent works on multimodal detection of a variety of linguistic phenomena.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For example, SVMs (Support Vector Machines) with word embeddings and feature extraction have been used for multimodal sarcasm detection (Castro et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Alnajjar and Hämäläinen, 2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mit- tal et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2020) uses GloVe embeddings, features extracted from audio and facial recognition system output to predict emotion in a multimodal dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' These multimodal features are fused using a mem- ory fusion network (MFN) (Zadeh et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Similarly, Li et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2021) detect emotion in a mul- timodal dataset by modeling the problem from the point of view of the quantum theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' While the field has seen increasing research on multimodal NLP (Tsai et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mai et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Sahu and Vechtomova, 2021), no data or model has been proposed for multimodal metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 3 Our Metaphor Corpus In this section, we present our video, audio and textual corpus of manually annotated metaphorical language.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Our selection of the video clips includes only CC-BY licensed videos on YouTube that have human authored closed captions in English.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The content of the videos presents mainly real people talking, which rules out animations and video game streams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The availability of human authored closed captioning is important as it speeds up our anno- tation time and provides us with subtitles that are already aligned with video and audio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The CC- BY license was an important selection criterion because it makes it possible for us to release the dataset openly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We used the filters provided by YouTube to limit our search to videos that were marked as CC-BY and had closed captioning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' However, the YouTube filter does not distinguish between auto- matically generated closed captioning and a human authored one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Fortunately, it is relatively easy to tell these two apart from each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Automated closed captioning tends to appear one word at a time, whereas human authored closed captioning is visualized more like traditional subtitles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' These cri- teria greatly reduced the number of eligible videos to include in our corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Apart from these criteria, we also filtered videos with sensitive and offensive languages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' No further restrictions have been ex- plicitly placed on the genres or types of videos, as we do not want to introduce biases for which types of contents are more likely to contain metaphors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Therefore, the availability of the metaphors natu- rally occurring in the corpus is the result of the ubiquity of the metaphor in everyday language use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' All Youtube queries were conducted in incognito mode to avoid biased YouTube suggestions based on our viewing habits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Figure 1 shows real examples from our corpus where video can be useful in detecting metaphors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' On the left, the woman wearing a gray shirt is sentence that you can use to really up your game because while a quick fix can be appetizing and appealing That’s the street address for your website you’re ready to give it a shot Table 1: Example of the annotations for the metaphor detection corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Figure 1: Metaphors made visible in the video through gestures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' talking about sprinkling keywords and showing a sprinkling gesture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' On the right, the woman wear- ing the wine red shirt says ring that bell and shows a bell ringing gesture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Our corpus consists of 27 YouTube videos with a total duration of 3 hours, 53 minutes and 47 sec- onds of video.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For comparison, a recently released multimodal dataset for sarcasm detection (Castro et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019) has the duration of 3 hours, 40 minutes and 47 seconds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The videos belong mostly to a start-up domain and many of them deal with issues of online visibility for a start-up company.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This domain was a consequence of our selection criteria for videos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' It turns out that YouTube has plenty of high-quality human close-captioned videos re- leased under the CC-BY license that relate to this particular domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Our corpus provides linguistics researchers with the ability to study the use of metaphor in a mul- timodal setting, something that has gained atten- tion in their field of science as well (Müller and Cienki, 2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This can, indeed, foster a wider interdisciplinary collaboration leading to a deeper understanding of the phenomenon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='1 Annotation Two expert annotators went through the video files and annotated metaphors by surrounding them with v tags for vehicles and t tags for tenors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The use of experts is motivated by the fact that previ- ous research has found that non-expert annotators struggle with metaphors (Hämäläinen and Alnajjar, 2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The annotators followed a simple procedure in annotating the data: Is the meaning literal?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' If the meaning of the word is abstract, is it a dictionary meaning?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Does the potential metaphor express prag- matic insincerity?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' If the answer to all of the questions is no, annotate it as a metaphor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In other words, if the meaning of a word or a phrase is not literal, it is annotated as a metaphor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' However, just the mere fact of a word being used in an abstract way is not enough to mark it as metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For example, in the sentence it is tied to revenue, “tied” is not tagged as a metaphor just because it is used in a more abstract sense than the typical concrete sense of tying one’s shoes, for example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' If the abstract meaning of a word ap- pears in a dictionary, the word is not considered metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' However, conventional metaphors that consist of multiple words, and are thus idioms, are tagged as metaphors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We do not make a distinc- tion between metaphors and similes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Pragmatic insincerity (see Grice 1975) is a phe- nomenon related to sarcasm as one of its precon- ditions (see Kumon-Nakamura et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1995).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There is a certain overlap between metaphors and sarcas- tic expressions in the sense that both use words in their non-literal meaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In order to ensure that we do not mix these two notions with each other, it is important to avoid annotating pragmatically insincere expressions as metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Table 1 shows an example of annotations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The annotations were done directly in the subtitles The utterances are time stamped and aligned with the video.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In the table, tenors are indicated with and vehicles with .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For deictic tenors, an r attribute is provided to resolve the deixis by indi- cating the actual tenor that has appeared earlier in the conversation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In the examples, game is used metaphorically to talk about marketing, quick fix is TheJourney you should probably sprinkle those in.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' and ring that bellcalled appetizing as though it was something edible and domain name is contrasted to a physical street address by direct comparison.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Give it a shot is a conventional metaphor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' All in all, after multiple annotation iterations, the dataset consists of 304 vehicles and 67 tenors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This totals to 371 metaphorical expressions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' They vary in length: the shortest tenor is one word, such as it, while the longest tenor is several words the discovery of those five noble gases to illuminate like that.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The same goes for vehicles where their length varies form one word such as dive to multi- ple words: the history of the internet itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' On a token level, we have 672 vehicle tokens and 113 tenor tokens, so altogether 785 metaphorical to- kens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In total, 6% of the expressions in the corpus are metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' While this percentage might appear low, it is natural and more representative of the real usage of metaphors in typical conversations which makes this corpus suitable for building metaphor detection models applicable for real-world scenar- ios.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Around 55% of the vehicles are conventional metaphors and 45% are novel metaphors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' How- ever, it is fairly common that same words appear in the corpus in a metaphorical and non-metaphorical sense.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In our corpus, there are two videos that deal with actual cooking, in which many food- related metaphors appear non-metaphorically, such as sprinkle those in, said metaphorically about key- words and a little sprinkle, said non-metaphorically about sugar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Another example is the use of house non-metaphorically as in come pick it up at my house and metaphorically as in think of hosting as your house, where a metaphorical connection is drawn between hosting and a house.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='2 Data preparation As YouTube serves files in several different formats such as webm, mkv and mp4 the first step is to use FFmpeg2 to convert all videos into mp4 format.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We also use the same tool to clip the video files into sentence-length clips based on the time stamps in the subtitles and extract their audio into wav files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This process yielded 6,565 video and audio clips that are aligned with text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We split the datset randomly so that 70% of sen- tences that contain metaphors and 70% of sentences that don’t contain any metaphors are used for train- 2https://ffmpeg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='org/ ing, 15 % of both types of sentences for validation and 15% of both for testing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This way we ensure that both metaphorical and non-metaphorical sen- tences are divided proportionally with the same ratios.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' These splits are used for all the models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4 Metaphor Detection We experiment with uni- and multi-modal models for metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In this section, we describe the preprocessing steps applied and the experimen- tal setups conducted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='1 Preprocessing For each modality, we make use of the latest ad- vances in neural network models to capture impor- tant features that have achieved state-of-the-art re- sults in various NLP tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' As metaphor detection has been conducted solely based on text, we fol- low the DeepMet approach by Su et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' (2020) and process the entire textual content using spaCy (Hon- nibal et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020) to tokenize it and acquire Uni- versal Dependencies style syntactic trees (Nivre et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020) and Penn Treebank parts-of-speech tags (Santorini, 1990).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Similarly to the original ap- proach, all of our textual models predict metaphors at the token level given the context surrounding it and its POS tags as input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We resample the audio to 16kHz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Audio features are extracted using Wav2Vec2FeatureExtractor pro- vided by the Transoformers Python library (Wolf et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Video features are obtained by taking equally- distributed 16 frames from a clip and then resize them into 128x171, followed by normalization and center cropping to 112x112.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='2 Textual model We train two text-only models, both follow the architecture and approach of DeepMet where we obtain textual embeddings using RoBERTa (Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2019) and feed them into two transformer encoding layers which are then combined by apply- ing global average pooling and concatenation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' A dense fully-connected layer takes in the combined output of both encoders and predicts whether the token is metaphorical (c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', Su et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020 for more details).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In our first textual model, we train the model using our corpus, whereas in the second one we train it using VUA corpus (with a learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='00001, akin the original paper) and later fine-tune it using our corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='3 Audio model We extend and fine-tune Facebook’s pretrained mul- tilingual XLSR-Wav2Vec2 large model (Baevski et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The model is trained on Multilingual LibriSpeech (Pratap et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020), CommonVoice (Ardila et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2020) and Babel (Roach et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 1996) for speech recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We employ this model to encode speech into vector representations from raw audio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We replace the classification layer of the original model with a dense fully-connected layer that pro- duces two outputs, one for each label.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Unlike the textual model, here we classify whether the entire spoken expression contains a metaphor or not (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', not on a word level).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='4 Video model For our video unimodal model, we incorporate a pretrained model for human action detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The model is based on the 18 layer deep R(2+1)D net- work (Tran et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2018) and it is trained on the Kinetics-400 (Zisserman et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2017) dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The intuition behind using this model is that it was able to detect actions (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', playing organ), ges- tures (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', pointing) and movements (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', waving).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Realizing such information is crucial in understand- ing the context, and would provide further cues for detecting metaphors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Similar to the audio model, we substitute the original classification layer with a fully connected layer and fine-tune the pretrained model to predict whether a scene is metaphorical or not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='5 Multimodal metaphor detection We test out three multimodal metaphor detection models;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1) text and audio, 2) text and video and 3) text, audio and video.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The textual model is the fine-tuned model using the VUA corpus and our textual corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In all of the models, the final classification layer of their sub-models are removed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Unimodal models are combined by concatenating the weights of their last layer, which are then fed to a classification layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='6 Common configuration All of the models described above share common configurations, unless we explicitly indicate other- wise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Prior to the last classification layer of all of our mono- and multimodal models, we introduce a dropout layer (Srivastava et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', 2014) (with a prob- ability of 20%) to accelerate training, and reduce internal covariate shift and overfitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We use the cross entropy loss function along with Adam optimizer (Kingma and Ba, 2014;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Loshchilov and Hutter, 2019) to update the weights and train the models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' All the fine-tuned models are trained with a learning rate of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='0001 and for 3 full epochs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 5 Results In this section, we follow the evaluation metrics commonly used for the metaphor detection task by reporting the precision, recall and F1 scores for the metaphorical label.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Regarding the textual models, we report three sets of results, which are for the models trained on: 1) VUA corpus, 2) our corpus and 3) both the VUA and our corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' All the models predict metaphoricity on the token level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' To ensure that our implementation of the DeepMet approach is correct, we tested the first model on the VUA test dataset of the metaphor detection shared task and achieved an F1-score of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='68 and 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='73 on all POS and verb subsets of the data, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' These results are relatively close to the results reported by the authors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Table 2 shows the classification results of all three models on the test set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The test set con- tained 90 metaphorical tokens and 6,961 non- metaphorical tokens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The results indicate that the textual model trained solely on the VUA dataset performed poorly on our test set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In compari- son, training the model using our metaphor corpus only resulted in a great increase of correct predic- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Nonetheless, combining both corpora by fine-tuning the first model with our corpus pro- duced the winning model, which managed to spot 76% of the metaphorical tokens correctly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We believe that the huge differences between the first and second textual models, despite the larger size of VUA’s training dataset, are due to the dif- ferences in domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The VUA corpus contains academic texts, conversation, fiction, and news texts, whereas our corpus is dominated by con- versations on the web and start-ups.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' It is evident that by exposing the model to general domains (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=', VUA’s corpus) and, thereafter, concentrating it on the start-up domain, the model was able to identify the highest number of metaphorical usages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Trained on Precision Recall F1-score VUA 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='04 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='33 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='07 Ours 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='38 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='63 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='47 VUA + Ours 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='53 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='76 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='62 Table 2: Classification results of the textual monomodal models on the test set of our corpus, for the metaphorical label.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Results from the other models (unimodal or mul- timodal) that involving audio and video showed that adding these modalities actually did not help improving the model - rather, they are detrimental to the model performance on metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We extend two possible explanations for this failure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' First, it is possible that because the visual and audio cues of metaphor are subtle, these models failed to learn from such a small amount of annotated data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Second, it is unclear that the specific models we are using for audio and video modalities encode the information relevant for the metaphor detec- tion task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For instance, whereas it is impossible to completely disentangle what exactly the Wav2Vec model is encoding, we can conjecture that it en- codes information about phoneme identity consid- ering it is optimized for the speech recognition task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Therefore, it may not be entirely surprising that the Wav2Vec encoding is not useful for the metaphor detection task because it is adding redundant or irrelevant information to the model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' It is our future work (or the future work for the community who utilizes this dataset) to refine our understanding of the multimodal encoding for the metaphor de- tection task (for instance, employing a model that more directly encodes information about speech prosody from the audio).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='1 Error analysis When looking at the results of the text only model, we can see that the model identifies metaphors cor- rectly as metaphors more often than not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are some metaphorical tokens in metaphors con- sisting of multiple words that get classified wrong, for example, in You could think of hosting as your house, the tenor hosting and the determinant your of the metaphorical word house are not identified as metaphorical, while house is correctly identified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Another example is the conventional metaphor toot their own horn, where all other words except for own are correctly identified as metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are also a fewer number of cases where all words get identified wrongly as non-metaphorical, for example, the model did not predict any metaphorical tokens in It’s where you live, while in reality it is the tenor and where you live is the vehicle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Also, individual tenors where the vehicle comes later get often not recognized such as in Yes, malware you could think of like, where malware is the tenor for a vehicle that appears later in the dialog.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' When the tenor and the vehicle co-exist nearby, the model can get all metaphorical tokens right such as in It’s kinda like real estate right?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' where both the tenor it and the vehicle real estate are cor- rectly identified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Also many tenorless expressions are fully recognized correctly as metaphorical, such as Spreadin’ the love.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There were plenty of cases (61) where the model predicted a metaphor tag for a token while there was no metaphor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Curiously, prepositions were of- ten tagged metaphorical, such as to in ring that bell to see these episodes first.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The actual metaphorical part ring that bell ends before the preposition to that has a non-metaphorical meaning in order to.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' We can also see that the model was indeed fooled by cooking terms that were used both metaphor- ically and non-metaphorically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Yeah a little sprinkle, both a and sprinkle were classified as metaphors, while the context was about sprinkling sugar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Another similar case was there’s five noble gases that illuminate, where noble gases and illumi- nate were erroneously classified to be metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' This was clearly due to the tenor in the corpus: the discovery of those five noble gases to illuminate like that that contained similar words.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' It is evident that the model relies on word similarities more than reaching to a higher pragmatic representation of the phenomenon, however, this is not an unexpected behavior from a machine learning model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are also cases where the model detects a metaphor, that could theoretically be a metaphor, but is not because of the way it was used in the cor- pus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For example, the model predicts Give it a go as metaphorical in the expression button, "Give it a go.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' ", where people are talking about a button with a particular text rather than using the expression metaphorically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Another such an example is flying in (money flying).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Such an expression might be used metaphorically, but in this case this was a note for the hearing impaired as money was actually flying on the video.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 6 Discussion and Conclusions In this work, we have only focused on metaphor as a strictly linguistic phenomenon and we have built a multimodal dataset where these linguistic metaphors have been tagged in terms of tenors and vehicles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' However, it is apparent that metaphor is a phenomenon that occurs on a higher level of our cognitive capacities than mere language.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' There are several cases in our corpus, where we can evidence the existence of a metaphor but it is never expressed verbally.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For example in Figure 2, money flying cannot be a metaphor when inspected purely from the point of view of language and its relation to the video when money is actually flying in the scene.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' However, it is a metaphor on a higher level in the sense that the entire scene where money was flying was to indicate someone becoming rich.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In other words, stating a fact that is happening is not metaphorical if the fact is literally taking place, however the fact itself might be metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Figure 2: Money actually flying on the video.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' At the same time, as evidenced by our error anal- ysis, there are certainly cases where video modality could help in disambiguating whether something is said metaphorically or not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' For instance, talk- ing about sprinkling in a kitchen environment (see Figure 3) is a very strong sign that the word is po- tentially non-metaphorical.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Integrating these weak cues into a multimodal system is, however, not an easy task given that the current methods for video processing are limited in their coverage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Therefore, in the future, it would be useful to annotate metaphors also in the other modalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Money flying can be a visual metaphor, and so can a sound effect, and they can exist independently from each other in different modalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Perhaps the reason why our multimodal attempts failed was that metaphor can be independent of the other modali- ties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Producing such a dataset where these modal specific metaphors are also annotated for video and Figure 3: Sprinkling used in a kitchen in reference to sugar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' audio is definitely a huge undertaking that requires research in its own right.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' It is clear that our model can detect metaphors correctly, but also the mistakes it makes highlight that despite using a large RoBERTa model, the meaning representation the model has cannot reach to such a nuanced level as to confidently detect metaphors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphor is a figurative device that can- not be explained by semantics, but rather requires pragmatic inspection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' It is not clear based on our re- search and other contemporary approaches whether the current word or sentence embedding models are sufficient to navigate in the depths of pragmatics and subjective interpretation in any other way than learning some irrelevant co-occurring phenomena from a biased corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' At the same time there is no such thing as an unbiased corpus, either, given that bias (and mostly heuristics causing it) is a funda- mental part of our cognition as human beings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In this paper, we have presented a new open and multimodal dataset for metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Be- cause we have focused strictly on CC-BY licensed videos, we can make the entire dataset available on Zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In our current work, we have not taken the context widely into account when predicting metaphoricity, but rather resorted to a very local context.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The fact that the videos can be published in full length makes it possible for any future work to explore different ways of including contextual cues freely.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Acknowledgments This work was partially financed by the Society of Swedish Literature in Finland with funding from Enhancing Conversational AI with Computational Creativity, and by the Ella and Georg Ehrnrooth Foundation for Modelling Conversational Artificial Intelligence with Intent and Creativity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The Journey JOURNEY (money flying)Yeah a little sprinkleReferences Khalid Alnajjar and Mika Hämäläinen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' ¡Qué maravilla!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' multimodal sarcasm detection in Span- ish: a dataset and a baseline.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' arXiv preprint arXiv:2105.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='05542.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, Reuben Morais, Lindsay Saunders, Francis Tyers, and Gre- gor Weber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Common voice: A massively- multilingual speech corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the 12th Language Resources and Evaluation Confer- ence, pages 4218–4222, Marseille, France.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Euro- pean Language Resources Association.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' wav2vec 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='0: A frame- work for self-supervised learning of speech represen- tations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Advances in Neural Information Processing Systems, 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Kfir Bar, Nachum Dershowitz, and Lena Dankin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Automatic metaphor interpretation using word em- beddings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' arXiv preprint arXiv:2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='02665.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Beata Beigman Klebanov, Chee Wee (Ben) Leong, and Michael Flor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' A corpus of non-native written English annotated for metaphor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the 2018 Conference of the North American Chap- ter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Pa- pers), pages 86–91, New Orleans, Louisiana.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Asso- ciation for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Giuseppe Castellucci, Simone Filice, Soujanya Poria, Erik Cambria, and Lucia Specia, editors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Pro- ceedings of the First International Workshop on Nat- ural Language Processing Beyond Text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Santiago Castro, Devamanyu Hazarika, Verónica Pérez- Rosas, Roger Zimmermann, Rada Mihalcea, and Soujanya Poria.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Towards multimodal sarcasm detection (an _obviously_ perfect paper).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Pro- ceedings of the 57th Annual Meeting of the Associa- tion for Computational Linguistics (Volume 1: Long Papers), Florence, Italy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computa- tional Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Verna Dankers, Karan Malhotra, Gaurav Kudva, Volodymyr Medentsiy, and Ekaterina Shutova.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Being neighbourly: Neural metaphor identification in discourse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the Second Work- shop on Figurative Language Processing, pages 227–234, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Thierry Declerk, Itziar Gonzalez-Dios, and German Rigau, editors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Proceedings of the LREC 2020 Workshop on Multimodal Wordnets (MMW2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The European Language Resources Association (ELRA), Marseille, France.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Ge Gao, Eunsol Choi, Yejin Choi, and Luke Zettle- moyer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Neural metaphor detection in context.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the 2018 Conference on Empiri- cal Methods in Natural Language Processing, pages 607–613, Brussels, Belgium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Com- putational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Herbert P Grice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1975.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Logic and conversation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Speech acts, pages 41–58.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Brill.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mika Hämäläinen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Harnessing nlg to create finnish poetry automatically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the ninth international conference on computational creativity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Creativity (ACC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mika Hämäläinen and Khalid Alnajjar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Let’s FACE it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Finnish poetry generation with aesthetics and framing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the 12th Interna- tional Conference on Natural Language Generation, pages 290–300, Tokyo, Japan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Com- putational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Matthew Honnibal, Ines Montani, Sofie Van Lan- deghem, and Adriane Boyd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' spaCy: Industrial-strength Natural Language Processing in Python.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mervi Kantokorpi, Lyytikäinen Pirjo, and Viikari Auli.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Runousopin perusteet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Gaudeamus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Diederik P Kingma and Jimmy Ba.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Adam: A method for stochastic optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' arXiv preprint arXiv:1412.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='6980.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Sachi Kumon-Nakamura, Sam Glucksberg, and Mary Brown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1995.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' How about another piece of pie: The allusional pretense theory of discourse irony.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Jour- nal of Experimental Psychology: General, 124(1):3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Chee Wee (Ben) Leong, Beata Beigman Klebanov, Chris Hamill, Egon Stemle, Rutuja Ubale, and Xi- anyang Chen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' A report on the 2020 VUA and TOEFL metaphor detection shared task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Pro- ceedings of the Second Workshop on Figurative Lan- guage Processing, pages 18–29, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Qiuchi Li, Dimitris Gkoumas, Christina Lioma, and Massimo Melucci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Quantum-inspired multi- modal fusion for video sentiment analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Informa- tion Fusion, 65:58–71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Jerry Liu, Nathan O’Hara, Alexander Rubin, Rachel Draelos, and Cynthia Rudin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphor detec- tion using contextual word embeddings from trans- formers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the Second Workshop on Figurative Language Processing, pages 250–255, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Man- dar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Roberta: A robustly optimized bert pretraining ap- proach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Ilya Loshchilov and Frank Hutter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Decoupled weight decay regularization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In International Con- ference on Learning Representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Sijie Mai, Haifeng Hu, and Songlong Xing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Modality to modality translation: An adversarial rep- resentation learning and graph fusion network for multimodal fusion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the AAAI Con- ference on Artificial Intelligence, volume 34, pages 164–172.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Trisha Mittal, Uttaran Bhattacharya, Rohan Chandra, Aniket Bera, and Dinesh Manocha.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' M3er: Multiplicative multimodal emotion recognition us- ing facial, textual, and speech cues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 1359–1367.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Aditya Mogadala, Sandro Pezzelle, Dietrich Klakow, Marie-Francine Moens, and Zeynep Akata, edi- tors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Proceedings of the Second Workshop on Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics, Barcelona, Spain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Cornelia Müller and Alan Cienki.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Chapter 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Words, gestures, and beyond: Forms of multimodal metaphor in the use of spoken language, pages 297– 328.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' De Gruyter Mouton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Joakim Nivre, Marie-Catherine de Marneffe, Filip Gin- ter, Jan Hajic, Christopher D Manning, Sampo Pyysalo, Sebastian Schuster, Francis Tyers, and Daniel Zeman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Universal dependencies v2: An evergrowing multilingual treebank collection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of The 12th Language Resources and Evaluation Conference, pages 4034–4043.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Vineel Pratap, Qiantong Xu, Anuroop Sriram, Gabriel Synnaeve, and Ronan Collobert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mls: A large-scale multilingual dataset for speech research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' ArXiv, abs/2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content='03411.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Sunny Rai, Shampa Chakraverty, Devendra K Tayal, Divyanshu Sharma, and Ayush Garg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Under- standing metaphors using emotions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' New Genera- tion Computing, 37(1):5–27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Ivor Armstrong Richards.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1936.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The Philosophy of Rhetoric.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Oxford University Press, London, United Kingdom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Peter Roach, Simon Arnfield, W Barry, J Bal- tova, Marian Boldea, Adrian Fourcin, W Gonet, Ryszard Gubrynowicz, E Hallum, Lori Lamel, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1996.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Babel: An eastern european multi-language database.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceeding of Fourth International Conference on Spoken Language Processing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' IC- SLP’96, volume 3, pages 1892–1893.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' IEEE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Richard M Roberts and Roger J Kreuz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Why do people use figurative language?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Psychological science, 5(3):159–163.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Xose Rosales Sequeiros.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphor: Prag- matics, relevance and cognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' English Studies, 97(6):656–677.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Gaurav Sahu and Olga Vechtomova.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Adaptive fusion techniques for multimodal data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceed- ings of the 16th Conference of the European Chap- ter of the Association for Computational Linguistics: Main Volume, pages 3156–3166, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Associa- tion for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Beatrice Santorini.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Part-of-speech tagging guide- lines for the penn treebank project (3rd revision).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Technical Reports (CIS), page 570.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Dropout: A simple way to prevent neural networks from overfitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Journal of Machine Learning Re- search, 15(56):1929–1958.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' GJ Steen, AG Dorst, JB Herrmann, AA Kaal, and T Krennmayr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2010a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Metaphor in usage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Cognitive Linguistics, 21(4):757–788.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' GJ Steen, AG Dorst, JB Herrmann, AA Kaal, T Krenn- mayr, and T Pasma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2010b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' A method for linguistic metaphor identification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' from mip to mipvu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Con- verging Evidence in Language and Communication Research, (14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Chuandong Su, Fumiyo Fukumoto, Xiaoxi Huang, Jiyi Li, Rongbo Wang, and Zhiqun Chen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Deep- Met: A reading comprehension paradigm for token- level metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the Sec- ond Workshop on Figurative Language Processing, pages 30–39, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Asuka Terai and Taiki Sugyo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Construction of a corpus-based metaphor generation support system built on japanese literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In 2019 IEEE 11th Inter- national Workshop on Computational Intelligence and Applications (IWCIA), pages 1–6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' IEEE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, Yann LeCun, and Manohar Paluri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' A closer look at spatiotemporal convolutions for action recog- nition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6450–6459.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Elizabeth Closs Traugott.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 1985.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' ‘conventional’and ‘dead’metaphors revisited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The ubiquity of metaphor: Metaphor in language and thought, pages 17–56.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Yao-Hung Hubert Tsai, Shaojie Bai, Paul Pu Liang, J Zico Kolter, Louis-Philippe Morency, and Rus- lan Salakhutdinov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Multimodal transformer for unaligned multimodal language sequences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the conference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Com- putational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Meeting, volume 2019, page 6558.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' NIH Public Access.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Mingyu Wan, Kathleen Ahrens, Emmanuele Chersoni, Menghan Jiang, Qi Su, Rong Xiang, and Chu-Ren Huang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Using conceptual norms for metaphor detection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the Second Workshop on Figurative Language Processing, pages 104–109, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Association for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier- ric Cistac, Tim Rault, Rémi Louf, Morgan Funtow- icz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Rush.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Transformers: State-of-the-art natural language pro- cessing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45, Online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Asso- ciation for Computational Linguistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Ping Xiao, Khalid Alnajjar, Mark Granroth- Wilding, Kat Agres, and Hannu Toivonen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Meta4meaning: Automatic metaphor interpretation using corpus-derived word associations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Pro- ceedings of the Seventh International Conference on Computational Creativity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Sony CSL Paris.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Amir Zadeh, Paul Pu Liang, Navonil Mazumder, Soujanya Poria, Erik Cambria, and Louis-Philippe Morency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Memory fusion network for multi- view sequential learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Proceedings of the Thirty- Second AAAI Conference on Artificial Intelligence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Danning Zheng, Ruihua Song, Tianran Hu, Hao Fu, and Jin Zhou.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' “love is as complex as math”: Metaphor generation system for social chatbot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' In Workshop on Chinese Lexical Semantics, pages 337– 347.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Springer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' Andrew Zisserman, Joao Carreira, Karen Simonyan, Will Kay, Brian Zhang, Chloe Hillier, Sudheendra Vijayanarasimhan, Fabio Viola, Tim Green, Trevor Back, Paul Natsev, and Mustafa Suleyman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} +page_content=' The kinetics human action video dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/gtAzT4oBgHgl3EQfMfvB/content/2301.01134v1.pdf'} diff --git a/h9AzT4oBgHgl3EQfMvvr/content/tmp_files/2301.01139v1.pdf.txt b/h9AzT4oBgHgl3EQfMvvr/content/tmp_files/2301.01139v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..fdf1b15d50877fd3f7f46427e5857c95d27952d1 --- /dev/null +++ b/h9AzT4oBgHgl3EQfMvvr/content/tmp_files/2301.01139v1.pdf.txt @@ -0,0 +1,2785 @@ +Springer Nature 2021 LATEX template +Crystal Chemistry at High Pressure +Katerina P. Hilleke1 and Eva Zurek1* +1Department of Chemistry, State University of New York at +Buffalo, 777 Natural Sciences Complex, Buffalo, 14260-3000, NY, +USA. +*Corresponding author(s). E-mail(s): ezurek@buffalo.edu; +Contributing authors: khilleke@gmail.com; +1 Introduction +Jupiter, the largest planet in our solar system, is a gas giant comprised mostly +of hydrogen and helium. After penetrating its atmosphere, a mixture of hydro- +gen and helium, methane and ammonia, one would find a massive sea of +hydrogen – behaving very differently than one might expect from general chem- +istry classes [1]. The hydrogen layer surrounding Jupiter’s core, under immense +pressure and at high temperature, is metallic. This is just one example of how +the pressure variable is so important in determining chemical and physical +behavior, bearing consequences for the evolution and dynamics of planetary +interiors [2]. +Under pressure, the behavior of the elements as well as the compounds +they form can change drastically from what we, as beings living and learn- +ing at 1 atmosphere, are used to. Hydrogen, as described above, can become +metallic at high pressure, joining its Group I cousins the alkali metals. It may +be perplexing then to learn that the alkali metals lithium [3] and sodium [4, 5] +become semiconducting and insulating, respectively, when squeezed. Potas- +sium starts to behave more like a transition metal [6–8]. When combined with +one another, the unexpected behavior of the elements under pressure makes for +correspondingly curious compounds. Sodium chlorides with vastly divergent +stoichiometries from the typical 1:1 (Figure 1) have been predicted and synthe- +sized [9], stubbornly inert helium atoms form a compound with sodium [10], +and metal superhydrides wherein the hydrogen atoms coalesce into clathrate- +like networks have been reported [11]. Various “rules” for the behavior of +materials at high pressure have been proposed by Prewitt and Downs [12], +1 +arXiv:2301.01139v1 [cond-mat.mtrl-sci] 3 Jan 2023 + +Springer Nature 2021 LATEX template +2 +Article Title +Grochala and co-workers [13], and Zhang et. al. [14], to name a few – and have +undergone progressive revision as more is uncovered about the structures and +materials far below us in the Earth’s core and far away in the planets and stars. +Fig. 1 The familiar crystal structure of NaCl, table salt, (a) is the only known stable +compound of sodium and chlorine at ambient pressure. Under pressures exerted by a dia- +mond anvil cell (b), a variety of additional stoichiometries and crystal structures decorate +the sodium-chlorine phase diagram, including the P4/mmm Na3Cl, P4/m Na3Cl2, Imma +NaCl2, and Pm¯3 NaCl7 [10] structures shown here. +Over the past century, experimental methods have evolved to create pro- +gressively higher pressures in a laboratory setting, allowing us to directly probe +the behavior of materials under extreme conditions. Pressures in the megabar +range can now be routinely achieved, albeit still requiring delicate setups [15– +17]. Diamond anvil cells (DACs) combine the superior hardness of this desired +polymorph of carbon with its optical transparency, enabling the creation of +the highest static pressures and interrogation of the sample via visual and +spectroscopic means. Engineering advances from multistage compression appa- +ratuses [18, 19] to toroidal DACs [20, 21] have driven the experimental ceiling +for static pressures ever higher. Dynamic compression experiments under ramp +or shock conditions, driven by gas guns, lases pulses, or magnetic fields can +reach well into the terapascal regime [22–24], allowing us to study the behavior +of diamond at 5 TPa [25] and iron at conditions thought to be in super-Earths +[26]. Diagnostic techniques for characterizing the resulting substances can be +difficult to implement in both dynamic and static compression, often requiring +theoretical support for their interpretation [27]. +Far cheaper than high-pressure experiments are computations modeling +high-pressure systems. Band structures, phonons, Raman or infra-red spectra +and numerous material properties can all be calculated without so much as +stepping foot into a laboratory. Crystal structure prediction (CSP) techniques, +not weighted down by preconceived, atmospheric pressure-based, notions of +how atoms might appropriately arrange themselves in a unit cell, can identify +which structures can exist under high-pressure conditions [28–30]. The com- +putational exploration of potential energy landscapes will not be discussed + +Na3Cl2 +b +(a) +NaCl +NaCl2 +NaCl7 +Na3CSpringer Nature 2021 LATEX template +Article Title +3 +in our contribution; to learn about methods that can be employed to iden- +tify the global, as well as important local, minima we point the reader to +an excellent chapter in this book “Crystal Structure Prediction” by Andreas +Hermann, Lewis J. Conway and Chris J. Pickard. Exploratory calculations +highlight promising phases for further experimental investigation, but compu- +tation can just as well follow experimental results, elucidating behavior and +filling in the gaps. The resulting feedback loop of experiment and theory has +driven the discovery and characterization of a plethora of phases ranging from +the superhard [31–33] to the superconducting [11, 34–38]. +In the following sections we build a framework for understanding the +behavior of materials at high pressure, starting from the effects of pressure +on the atoms themselves, driving electronic transitions and altering periodic +trends. From there, the various manifestations of high pressure in the solid +state are sorted into categories (which are not mutually exclusive, but rather +illustrative), starting with exotic electronic structures and electrides. We dis- +cuss compounds of the noble gases and those containing elements that are +immiscible at ambient pressure, as well as crystal lattices that contain bizarre +geometrical motifs and bonding configurations. Finally, we survey the effects +of high pressure on superconductivity, a field that has recently undergone a +veritable explosion as high pressure phases toe the line of room-temperature +superconductivity [39]. +2 The atom under pressure +Chemistry describes the interactions between the 118 distinct elements that +are organized in the periodic table, proposed by Mendeleev while classifying +elements according to their chemical properties observed at atmospheric condi- +tions. The trends found within the periodic table can be used to compare atoms +according to their size, the number of electrons surrounding their nuclei, and +to make predictions as to whether those electrons are held tightly or loosely. +Moreover, the periodic table allows students and researchers to predict how +different elements will interact with one another: will they form compounds, +emulsions or alloys, or will they be unreactive? If reactivity is suspected, the +periodic table can be used to guess if the bonds are covalent, ionic, or (usu- +ally) somewhere in between. Across the periodic table, trends in properties +such as atomic radii, electronegativity, and oxidation state can be mapped, +leading one to conclude that fluorine, the most electronegative element, will +gain electrons in a binary phase thereby achieving an F− configuration with +a filled valence shell. On the other hand, cesium, as the least electronega- +tive element (neglecting francium, whose miniscule half-life renders it basically +experimentally irrelevant), typically assumes an oxidation state of 0 or +1.1 +Yet at high pressure, several LinCs phases have been predicted [44] where Cs +attains unusual formal oxidation states thought to be in excess of -2 due to +1When dissolved in amines or ethers, Na, K, Rb, and Cs can assume the unusual oxidation state +of -1. [40] These so-called alkalides are thought to form ion-pairs with solvated metal cations. [41– +43] + +Springer Nature 2021 LATEX template +4 +Article Title +substantial electron transfer from lithium to cesium. How can the stability of +these unintuitive stoichiometries and their resulting electronic structures be +rationalized? +Let us consider electronegativity a little more. Although Pauling’s [45] is +the most widely adopted, a number of metrics have been used to produce scales +of electronegativity for the elements. In Pauling’s formulation, electronega- +tivity differences between pairs of atoms A and B are calculated from the +homo- and heteronuclear bond dissociation energies, then referenced against +the electronegativity of H being set to 2.1 (later adjusted to 2.2). In this +regime, the electronegativities of fluorine, lithium, and cesium are 3.98, 0.98, +and 0.79, respectively. Several other scales have been proposed, including Mul- +liken’s “absolute electronegativity” [46] being the average of the first ionization +energy and electron affinity of an atom [47]. Dong et al. modify the Mulliken +definition for elements under high pressure taking as a reference the homoge- +neous electron gas rather than the vacuum [48]. Allen’s electronegativities are +derived from the average energies of the valence electrons in the atom [49], +and a closely related scheme has recently been proposed by Rahm et al [50– +52], where electronegativity is calculated as the average of the electron binding +energies of ground state valence electrons – approximately translated to the +Allen scale if averaging over valence electrons alone. Broadly speaking, the +common factor of importance in all of these is the collection of atomic orbital +energies and the differences between them. Under pressure, those change. +The prototypical model for understanding the quantized behavior of energy +levels in a confined system is the particle in a box. The resulting energy levels +for a particle of mass m in a box of width L are given by En = ℏ2n2π2 +2mL2 , where +n is the principal quantum number. To consider the effects of pressure on +this model, we could simply make the box smaller by reducing the width L, +which has the effect of increasing the En. Thus, energy levels (orbital energies) +will increase under pressure. The complicating factor is that the rate of this +increase is not the same for each orbital because the number of radial nodes +plays a role. The peak density of a 4s orbital is further from the nucleus +than a 4p orbital, since the electrons occupying the 4s orbital must maintain +orthogonality to those in the 1s, 2s, and 3s orbitals, while those in the 4p +orbital must only contend with the 3p and 2p orbitals (this analogy may be +extended to the 4d and 4f shells). With more electron density being further +from the nucleus, the electrons in the 4s orbital will feel the effects of pressure +more strongly than those in the 4p orbital. This is illustrated schematically +in Figure 2a and 2b. At certain levels of confinement, electronic ns → np, +ns −→ (n−1)d, and (n−1)d −→ (n−2)f transitions can become favorable. This +is the reason why pressure drives rearrangements of the orbital energies of an +atom, with ensuing electronic transitions. (In section 3.2, we will explore the +opportunities presented by another sort of destination orbital, this time one +not centered on an atom.) +In compressed lithium, the ground state electronic configuration transitions +from 1s22s1 → 1s22p1, while cesium undergoes a 6s1 → 5d1 pressure induced + +Springer Nature 2021 LATEX template +Article Title +5 +transformation. The exact pressures at which these electronic changes occur +depend, of course, on the chemical environment of the lithium or cesium atoms +– and in calculations, on the theoretical methodology. In Cs, the s → d tran- +sition is thought to drive the transformation to the complex Cs-III structure +with 84 atoms in the unit cell that is stable near 4.2 GPa [53–58], while in +lithium the pressure at which the s-p mixing occurs is believed to be somewhat +higher, in the megabar to multimegabar range, dependent on the chemical +environment and level of theory [57, 59–61]. Another example includes potas- +sium, whose complex phase diagram under pressure (see, e.g. Figure 2c,d), has +been in part attributed to this s → d electronic transition. [62] The energies +of core orbitals can also become relevant; for example, in Cs-VI the 5p bands +hybridize with the 6s [63, 64], making them accessible for chemical interac- +tions, and the core orbitals of K have also been proposed to be key to its +structural diversity. [62] Overall, one effect of pressure on the alkali metals is +that their electronegativities undergo quite a rearrangement, ending up with +cesium being more electronegative than lithium [48, 57]. From this perspective, +the formation of cesium anions in the LinCs phases begins to make sense. +Fig. 2 External pressure on an atom raises the energy of atomic orbitals as they are con- +strained to a smaller space, but the rate of this increase differs between s, p, and d orbitals +(a), (b), favoring s → d transitions for many of the alkali metals, including potassium. +Potassium adopts the bcc structure at 0 GPa (c) but at higher pressures takes on a series +of complex structures including the incommensurately modulated host-guest tI19 phase [65] +(d). +Among the d-block similar reorderings are predicted to take place, with +the group 10 metals Ni, Pd, and Pt preferring d10 closed-shell configurations, +as compared to the s2d8 favored at ambient conditions, while the group 11 +and 12 metals become electron donors [48]. The former transition is associated +with a spike in the estimated chemical hardness – obtained as half of the +HOMO-LUMO gap – of Ni, Pd, and Pt, reaching values comparable to some +of the noble gases. This is contrary to the general trend where the hardness of + +a) +b) +c) bcc K (O GPa) +d) tI19 K (20 GPa) +Incommensurately modulated +host-guest framework +Ambient pressure +orbital energies +External pressure +alters orderingSpringer Nature 2021 LATEX template +6 +Article Title +most of the elements in the periodic table decreases with pressure as energy +levels become closer to one another [48]. The resulting changes in the relative +hardness of pairs of elements can lead to changes in compound stability arising +from HSAB (hard-soft acid-base) arguments, and the appearance of strange +multicenter bonding manifolds in certain high pressure phases have been linked +to general increases in softness [66]. +From transition-metal-like behavior in the s-block to relative inertness in +the formerly-d8-transition metals, atoms under compression can behave very +differently from their ambient-pressure selves, and the consequences for mate- +rials under pressure are far-reaching. We will now explore some of the wild +and wonderful structures and phenomena that result. +3 The crystal under pressure +At ambient pressure a majority of the solid, metallic elements of the periodic +table adopt very simple, symmetric, structures that are close-packed. The most +stable geometries are those that minimize the free energy. However, at low tem- +peratures the entropic contributions between different solid phases are typically +negligible, so computational studies often employ the enthalpy to determine the +structures that are preferred. With increasing pressure the enthalpy, consist- +ing of the internal energy and pressure-volume terms (H = U +PV ), becomes +dominated by the PV term. It would be natural, therefore, to imagine that +close-packed structures with increased coordination numbers become preferred +at high pressures. The reality, however, does not coincide with our expecta- +tions. For example, within cesium the nearest neighbor coordination number +first increases from 8 (bcc) to 12 (fcc), then decreases to about 10 (Cs-III), +8 (Cs-IV) and finally increases again to 10/11 (Cs-V) and 12 (Cs-VI). These +structural transitions are believed to be driven by the previously discussed +pressure-induced s → d valence electronic transition within the constituent +atoms, which causes the interatomic distances to become smaller compared to +the ranges of the wavefunction [53]. +Moving beyond elemental crystal structures, several compounds with seem- +ingly bizarre stoichiometries (at least from the perspective of minds that +experience a 1 atmosphere reality) have been predicted and/or synthesized +under pressure. The familiar combination of sodium and chlorine, table salt and +prototypical ionic compound with a 1:1 ratio, is not the only stable crystalline +structure in the Na-Cl phase diagram. At least two unique stoichiometries, +Na3Cl and NaCl3, were synthesized, and several others were predicted to +become stable when squeezed [9]. Noble gases xenon, argon, and helium are +active components of solid compounds that have been synthesized [10, 67, 68], +and very hydrogen-rich compounds such as YH9, LaH10, and CaH6 that +are high-temperature superconductors with superconducting critical temper- +atures (Tcs) approaching room temperature have been made [11, 37, 69–72]. +These metal superhydrides are materials-by-design success stories inspired by +theoretical predictions [34, 35, 73]. + +Springer Nature 2021 LATEX template +Article Title +7 +In the following sections, we explore the plethora of exciting materials that +can be created using high pressure, with all their intriguing structural and +behavioral phenomena. While their stability and existence can be traced to the +pressure driven electronic rearrangements of the constituent atoms described +in Section 2, the manifestations of these rearrangements can take many forms. +Below, we describe a variety of illustrative phases sorted into a series of cate- +gories, but by necessity these categories will, at times, overlap. Nevertheless, +all exemplify the ramifications of high pressure on solid-state chemistry. +3.1 Electronic structure +At atmospheric pressure, the stoichiometries of many inorganic solid-state +compounds can be predicted from the most common oxidation states of their +constituent elements. Usually, alkali metals and alkaline earth metals possess +oxidation states of +1 and +2 respectively, so when combined with the O2− ion +one would expect Na2O and MgO, as well as K2O and SrO to form. The noble +gases are mostly inert, the p-block is amenable to forming covalently bonded +networks, and while a wide variety of oxidation states, which correspond to +various filled or half-filled orbitals, are available to several transition metals, +Zn, Cd, and Hg steadily persist in maintaining their d10 configurations. +The consequences of the orbital energy shifts discussed in Section 2 mean +that several of these “rules” no longer apply at high pressures, and elements can +adopt unusual oxidation states in compounds with unexpected stoichiometries. +The LinCs phases [44] used to illustrate the effects of pressure on electroneg- +ativity provide one such example. Above 70 GPa, the very lithium rich Li5Cs +phase is predicted to become stable, joined at higher pressures by Li3Cs, Li4Cs, +and LiCs [44]. Remarkably, the calculated Bader charges on the Cs atoms in +these stoichiometries are all more negative than -1, and since calculated Bader +charges frequently underestimate formal oxidation states, in these compounds +Cs may attain a formal oxidation state that is potentially lower than -2. While +alkali metal anions (alkalides) have previously been captured at ambient pres- +sures with cryptands [74], they achieve charges only up to -1 with the additional +electron going into the ns orbital. In the case of the LinCs phases, however, +pressure induces significant Li 2s → 2p and Cs 6s → 5d electronic transitions, +with the latter increasing progressively with higher Li content, thereby facil- +itating the acceptance of electron density by the typically unoccupied Cs 5d +orbitals. +For K, Cs, and Rb the pressure-driven ns → (n − 1)d transitions led to +predictions of transition-metal like behavior [6, 75] in the formation of inter- +metallic compounds with actual transition metals [8, 76–78]. Within some of +these compounds, the transition metal elements assume exotic electronic con- +figurations, as in the case of the predicted potassium iridide K3Ir (Figure 3a) +containing the Ir3− anion [79]. The Ir 5d orbital becomes fully occupied as +a result of electron transfer from K, echoed in the later predicted Rb3Ir and +Cs3Ir phases [80]. K3Ir and Rb3Ir share the Pmnm Cu3Ti structure type, while + +Springer Nature 2021 LATEX template +8 +Article Title +Cs3Ir adopts the P21/m Ni3Ta type, which consist of Ir@M8 and M@M8 dis- +torted cubes, and Ir@M12 distorted cuboctahedra respectively. In combination +with Li under high pressure, Au displays a similar ability to adopt a signif- +icantly negative formal charge of less than -3 in the predicted phases Li4Au +and Li5Au – where electrons donated from Li are placed into the empty Au 6p +orbitals [81], which are less destabilized than the Li 2s or 2p under pressure. +Fig. 3 High pressure compounds where the transition metal atoms adopt curious electronic +configurations in (a) K3Ir [79] (Cu3Ti type), with Ir@K8 and K@K8 distorted cubes and +iridide I3− anions, and (b) HgF4 [82], in which d8 configurations on the Hg lead to square +planar geometries. +Pressure can also promote chemical interactions with core or semi-core +orbitals, as in the case of HgF3 and HgF4 [82]. These stoichiometries are pre- +dicted to become stable above 73 and 38 GPa, respectively, and above 200 GPa +HgF4 is computed to decompose into HgF3 and F2. The I4m symmetry HgF4 +crystal possesses square planar HgF4 units typical of a d8 organometallic com- +plex, with the Electron Localization Function (ELF) [83] confirming covalent +Hg-F interactions. To form the four Hg-F bonds, not only the Hg 6s but also +two of the semicore 5d electrons are required. In HgF3, the Fm¯3m structure +(which distorts below 100 GPa to C2/m symmetry) involves a fluorite-type +HgF+ +2 lattice stuffed with F− ions, leaving Hg with a d9 configuration. A +series of predicted CsFn phases, in which the Cs 5p electrons participate in +Cs-F covalent bonds, are another example of the pressure-induced activation +of core electrons [84]. Their crystal structures display motifs resembling the +isoelectronic [XeFn]− molecules – for example, Fdd2 CsF5 contains planar +pentagonal CsF5 units, similar to the [XeF5]− anion. With increasing fluorine +content, the formal oxidation state on cesium reaches values greater than +1. +3.2 High Pressure Electrides +Electrides are solids where the electrons, localized on non nuclear-centric sites, +behave as anions [85]. They are conceptually related to solvated electrons, in + +a) K3Ir (Cu3Ti type) +b) HgF4 +Iridide Ir3- +d8 Hg(IV)Springer Nature 2021 LATEX template +Article Title +9 +which the excess electrons can be thought to occupy cavities in the fluid [86], +as well as alkalide liquids [41, 42] or alkalide solids where alkali metal anions +fill the interstitial voids [74]. Although many types of electride families are +known at atmospheric conditions, including those that are organic, inorganic, +intermetallic and those where the electron localization is restricted to various +dimensions or possesses topological properties, herein, we restrict the discus- +sion to high pressure electrides: systems where the electron localization occurs +as a response to compression [87]. +Though the formation of high pressure electrides has been rationalized in +many ways, including pressure induced orbital rehybridizations [4, 88, 89], +and multicenter bond formation [90, 91], herein we focus on a simple model +proposed by Miao and Hoffmann [87, 92]. As atoms in a solid compound are +compressed, raising their orbital energies, the electrons in the highest-energy +orbitals can vacate the atom entirely and occupy the interstices of the crys- +tal lattice instead. To understand why this might occur, Miao and Hoffmann +pointed out that orbitals can be ascribed to these voids, which can be thought +of as interstitial quasiatoms (ISQs). At ambient pressure, the ISQ energies +are higher than the atom-centered ones. However, unlike the atom-centered +orbitals, those of the ISQ do not experience the repulsive effect caused by the +core electrons, and their increase in energy with pressure will be less than the +atom-centered orbitals. When the ISQ orbital energies fall below the atom- +centered ones they will be occupied, thereby localizing the valence electrons +in the interstitial regions. These electrons, detached from the nuclei, serve as +anions and the corresponding compounds are called electrides. +For several simple metals, high-pressure phases identified as electrides via +calculations have been subsequently studied experimentally. In sodium, Neaton +and Ashcroft posited that under pressures high enough to induce overlap of the +2p orbitals, a combination of Pauli repulsion and core orthogonality constraints +would drive the valence electrons away from the ionic cores, to localize in the +crystalline interstices instead. This electronic redistribution would result in a +metal-to-insulator transition [88]. Later CSP calculations – with experimental +confirmation of an optically transparent, wide-bandgap insulating phase in the +same publication – proposed an insulating hP4 phase with P63/mmc sym- +metry to become stable above 260 GPa. [4] The new hP4 phase was in fact +experimentally observed at pressures as low as 200 GPa, but the discrepancy +was ascribed to a combination of thermal effects as well as the preferential +stabilization of metallic states by the computational method employed. Mov- +ing to higher pressures and temperatures, evidence for the hP4 phase has +been obtained in shock-ramp experiments [5]. However, in situ X-ray diffrac- +tion (XRD) revealed peaks that could not be attributed to hP4 between +240-325 GPa at temperatures in the thousands-of-degrees Kelvin. Consistent +with these dynamic compression experiments, calculations showed that the +free energy of a P63/m symmetry phase that is a topological electride was +lower than that of hP4 at these pressures and temperatures [93]. At higher + +Springer Nature 2021 LATEX template +10 +Article Title +pressures yet, ca. 15.5 TPa, sodium is predicted to adopt a curious, metal- +lic cI 24 electride phase consisting of Na12 icosahedra [94]. In the insulating +hP4 structure that dominates much of the high-pressure landscape in sodium, +highlighted in Figure 4a, the atoms occupy the Ni sites of the Ni2In structure +type with the ISQs on the In sites, in line with the treatment of this phase as +(Na+)2E2− (where E2− denotes a doubly-occupied ISQ). In fact, several A2X +alkali metal chalcogenides adopt the antifluorite Fm¯3m crystal structure at +ambient conditions, but eventually transition to the Ni2In structure type under +pressure [95, 96]. Potassium also adopts the hP4 phase when squeezed [62, 97]. +Lithium presents another example of complex structural and electronic +behavior under pressure, as first postulated by Neaton and Ashcroft who +suggested it may adopt an insulating, paired ground state [59]. Subsequent +experiments revealed that Li assumes the same semimetallic cI16 [60] struc- +ture found in Na [98]. At higher pressures, Li transitions to a number of unique +phases, such as those with orthorhombic C-centered lattices and 88, 40, and +24 atoms in their unit cells which have been observed [99]. One of these, oC40 +with the Aba2 space group, is an electride displaying especially interesting +behavior [100]. In this phase, ISQs occupy three separate symmetry-distinct +sites, two that are doubly occupied (EII) and the third singly occupied (EI), so +its primitive cell can be considered as Li20EII +8 EI +4. The EI-EI distance remains +roughly constant at a short 1.3 ˚A from 50-80 GPa, with an elevated elec- +tron density found between the ISQs. Crystal Orbital Hamilton Population +(COHP) [101], ELF, and projected density of states (PDOS) analyses all indi- +cate bonding character between the EI sites, and examination of the Γ-point +band-decomposed charge densities revealed bonding and antibonding states +analogous to the σg and σ∗ +u orbitals in H2. In fact, computations have shown +that ISQs can form covalent, ionic and metallic bonds with atoms as well as +with other ISQs [87, 89, 100, 102, 103]. For this reason, Miao has espoused +the idea that ISQs may be thought of as a chemical element and placed above +helium in the periodic table under pressure [104]. +The proclivity of the elements to ISQ formation has been investigated by +comparing the energies of their orbitals calculated at different pressures using +a He confinement model with those of an ISQ 1s orbital [92]. Unsurprisingly, +Li and Na were found to favor ISQ formation at relatively low pressures, +with Mg, Al, In, and Tl predicted to follow suit at higher pressures. Among +the heavier alkali metals, the energies of the valence s orbitals were found to +rapidly increase in energy relative to that of the ISQ 1s – but as previously +discussed, these elements are also susceptible to a pressure induced electronic +ns → (n − 1)d transition. Within cesium, for example, the increased d occu- +pation, already noted by Sternheimer in 1950 [53], was invoked to explain the +curious structure of Cs-IV adopted at 4.3 GPa, where the Cs atoms possess +a coordination number of 8 [55]. This decrease in coordination number with +increasing pressure appeared so counterintuitive to Linus Pauling that he pre- +sented an alternative structure solution assuming cubic symmetry and invoking +icosahedral clusters [105]. Pauling’s hypothesis turned out to be incorrect. + +Springer Nature 2021 LATEX template +Article Title +11 +Fig. 4 High pressure electride formation is observed in a number of complex high-pressure +phases of the alkali metals, including both the insulating (a) hP4 phase of Na [4] (Ni2In struc- +ture type) where ISQs containing two electrons occupy the In sites, and (b) semiconducting +oC40 Li, containing three inequivalent ISQ sites of which two (black) are doubly occupied, +and the third (pink) is singly occupied so a bond is formed between nearest neighbors, as +highlighted by the lines that join them. +Importantly, von Schnering and Nesper [106] recognized that the Cs atoms +in the Cs-IV structure occupied the Th sites of the ThSi2 lattice – and that +the valence electron density exhibited maxima not near the Cs atoms but at +the Si positions of ThSi2 and between the Si-Si bonds, dubbing this phase +an electride. This I41/amd structure is also assumed by K and Rb under +pressure [107–109]. +If the high-pressure phase behavior of the alkali metals were not complex +enough, Na, K, and Rb all adopt different versions of an incommensurately +modulated host-guest lattice (similar to the W5Si3 type), often referred to +as the tI19 structure, a model of which is illustrated for K in Figure 2d. [4, +65, 107, 110–114] All three share the same host lattice, but display different +periodicity in the guest lattice. In a study using commensurate approximants +to model the electronic structure, electron localization in the interstitial spaces +was found, with some highly localized basins as well as a 1D channel of electron +density lying in channels of the host structure [114]. Another study proposed +that electrides in the heavy alkali metals could be stabilized via ferromagnetic +ordering [75]. +Numerous predicted binary and ternary systems under pressure also behave +as electrides, including two Li3Fe phases (with P6/mmm and P4/mbm sym- +metries) [115], P4/mbm Na3Fe [116], and a superconducting Y3Si phase [117]. +In a range of LinI stoichiometry phases predicted above 50 GPa, ISQs form +within the interstices between I-centered Li polyhedra – but higher pressures +drive electron density back from the ISQs to the iodine atoms, filling the 5p +and eventually the 5d orbitals, skipping the 6s, in line with disfavoring s orbital +occupation under pressure [103]. Finally, several electride phases have been + +two-electron ISO +a) hP4 Na +b) oC40 Li +one-electron ISQ +ISOSpringer Nature 2021 LATEX template +12 +Article Title +identified involving the famously unreactive noble gases - including a particu- +larly surprising case [10] where the noble gas is crucial to the stability of the +synthesized structure. +3.3 Compounds of noble gases +Where did all the xenon go? +Relative to the abundance of Ar and Kr in the Earth’s atmosphere, the +amount of Xe is strikingly lower than it should be, a problem known as the +“missing xenon paradox”. Geoscientists have explained this discrepancy in +many different ways [118–120], but a growing body of evidence suggests that +the Xe did not escape, and instead it has been incorporated into the miner- +als found within the Earth. At Earth’s core pressures, both the atomic orbital +energies and the relative electronegativies of the elements, including Xe and +the Fe and Ni that comprise the majority of the core, are significantly per- +turbed [48, 50]. Therefore, it should not be a surprise that their reactivity +differs from our 1 atmosphere expectations. +Prior to the advent of widespread CSP, computational investigations con- +cluded that Xe incorporation into Fe and Ni would not occur, at least not +to a large extent [121, 122]. These studies, which relied on the assumption +that the Xe-metal alloys adopted crystal lattices similar to those of the ele- +mental metals, turned out to be incorrect. Later CSP-based studies predicted +the emergence of stable Xe-Fe and Xe-Ni compounds above 250 and 200 GPa, +respectively [123], with Pm¯3m XeFe3 (AuCu3-type) and Pmmn XeNi3 (based +on Xe@Ni12 cuboctahedra) having the lowest enthalpies of formation, although +P¯62m XeFe5 and XeNi5, as well as P21/m XeNi6 also appeared on the con- +vex hull. Confirming these predictions, experimental studies have synthesized +XeNi3 and Xe(Fe/Ni)3 phases at high pressure, although with slightly differ- +ent structures than those predicted. This includes Pm¯3m for XeNi3, either as +an ordered AuCu3 [67] or disordered CrNi3 alloy [68], and a mixture of fcc +and Pmmn-symmetry phases for XeFe3 [68]. In these systems, the Fe and Ni +atoms behave as oxidants, accepting 5p electron density from Xe [68, 123], in +agreement with the predicted increase in electronegativity differences between +Xe and the transition metals at high pressure [50]. Xe2FeO2 and XeFe3O6, +both involving substantial Fe-O and Xe-O bonding, have also been computed +to be stable at pressures relevant to the Earth’s core [124]. The high-pressure +ArNi phase, in which some Ni 3d electron density is transferred to the Ar 4s, +inducing a magnetic moment on the Ni, has been synthesized [125]. +CSP calculations have also predicted stable compounds containing Xe, +or other noble gas (NG) elements, and Mg above 125-250 GPa +[102]. This +includes Mg-Xe and Mg-Kr phases, which adopt structures based on stacked +square lattices of Mg and the NGs in different patterns, ranging from the +CsCl type (Pm¯3m) to more complex P4/nmm or I4/mmm arrangements for +MgNG and Mg2NG stoichiometries. Compounds of Mg with Ar, on the other +hand, were found to favor hexagonal arrangements such as anti-NiAs type +MgAr (P63/mmc). In these compounds the energies of the metal 3s orbitals + +Springer Nature 2021 LATEX template +Article Title +13 +increase precipitously in comparison to the outer shell d orbitals of the noble +gases inducing Mg 3s → NG d orbital transfer. The ELF of Mg2NG (NG=Xe, +Kr, and Ar) phases shows an additional interesting feature: ISQ formation. +This occurs far below the pressures at which elemental Mg is predicted to +form an electride [89, 92]. Two reasons have been used to explain this phe- +nomenon [102]. First, far fewer ISQ sites – concomitantly occupying less space +– relative to the elemental Mg electride are necessary to accept the displaced +valence electrons of Mg, as many of them are transferred to the NG atoms +instead. In addition, the NG atoms promote the formation of larger interstitial +spaces in the structure, stabilizing the ISQ at lower pressures. Under moder- +ate pressures of ca. 50-300 GPa, the energies of the outer shell Xe d orbitals +are similar to those of the ISQ 1s, although with higher pressure they become +lower in energy, congruent with the gradual ISQ 1s → NG d electron transfer +with increasing pressure up to 600 GPa [102]. +Several other stable compounds of Xe have been predicted at high pres- +sures, including XeO, XeO2, and XeO3 [126], as well as Xe3O2, Xe2O, and +Xe7O2 [127], while Xe2O5 and Xe3O2 have both been experimentally observed +at pressures lower than 100 GPa. [128] Krypton oxide, KrO, has been predicted +as well [129], as have xenon nitrides [130, 131] and carbides [132]. Fluorides of +argon [133], krypton [134], and xenon [135] – with Xe-Xe dimers cropping up +in Xe2F and XeF – have all been predicted. Several xenon chlorides including +XeCl, XeCl2, and XeCl4, with the former two being metastable by 10 GPa and +reaching the convex hull by 60 GPa have been computationally studied [136]. +When they are combined with Li, the noble gases Ar [137] and Xe [138] are +predicted to behave as anions, with the Li 2s orbital rising above the Xe and Kr +outer shell d orbitals. Several cesium xenides are predicted to be stable at high +pressures, many adopting alternate colorings of a distorted bcc lattice [139]. +There is experimental evidence for the formation of a phase mixing Xe with +water ice at conditions expected for planets such as Uranus and Neptune [140]. +Helium is famously the most inert element at ambient pressure by virtue of +its closed-shell electronic configuration, zero electron affinity and large ioniza- +tion potential. Nonetheless, a number of stable helium-containing compounds +have recently been predicted at high pressure, including those with iron [141], +ammonia [142], water [143, 144], nitrogen [145, 146], and even with other noble +gases [147, 148] (the van der Waals compound NeHe2, a Laves phase in the +MgZn2 structure has been experimentally observed [149]). +A particularly noteworthy example is provided by Na2He [10], an electride +phase with a fluorite-like lattice (Figure 5a) in which every Na8 cube that does +not contain a He atom is instead occupied by an electron pair (Figure 5b), +so that the phase can be expressed as (Na+)2(E2−)He. Although He does +not participate in any bond formation, its presence is nevertheless a crucial +stabilizing force in this phase, which has been successfully synthesized above +113 GPa [10]. A subsequent computational study showed that the He atoms act +as inert “spacers” to reduce the Madelung repulsion resulting from the unequal +amounts of cations and anions in the parent (Na+)2(E2−) phase [150]. Reaction + +Springer Nature 2021 LATEX template +14 +Article Title +Fig. 5 Conventional unit cell of helium-containing Na2He [10] in the Fm¯3m space group +(a), an electride with localized electron pairs occupying octahedral vacancies (b). The mech- +anism by which He incorporation serves to stabilize the structure is illustrated in (c), which +shows schematically how He insertion reduces electrostatic repulsions involved in the A2B +(Na2E) stoichiometry. +enthalpies for helium in combination with ionic AB, A2B and AB2 compounds +revealed that He incorporation was generally favored when the cation:anion +ratios were unequal such as in MgF2 and Li2O, but not for AB-type phases such +as LiF or MgO, in line with the prediction of successful He incorporation into +certain alkali oxides and sulfides [151]. Helium placement in the ionic lattices +tends to separate ions of similar charge as shown schematically in Figure 5c. An +FeO2He phase in which the Fe and O atoms form a fluorite lattice and the He +atoms occupy the remaining Fe8 cubes (isopointal to Na2EHe) was predicted +to be stable above 120 GPa [152], with He appearing to play the same role of +spacing agent. This mechanism allows even the most inert noble gases to play +an active role in stabilizing compounds at high pressure, all without forming +a single chemical bond. +3.4 Miscibility under pressure +The noble gases obtained their names due to their general lack of reactiv- +ity, but under ambient conditions, numerous combinations of elements resist +mixing to form alloys or stoichiometric compounds. The proclivity or reluc- +tance of a pair of elements towards mixing has been explained in a variety of +ways, resulting in predictive rules including those of Hume-Rothery and co- +workers [153], Miedema’s model [154, 155], Darken and Gurry’s maps [156], +and more [157–159]. As we will soon see, pressure turns out to be a useful +variable that can alter the (im)miscibility of two or more elements. +Consider, for example, magnesium and iron, whose large size and small +electronegativity differential at ambient pressure makes compound forma- +tion intractable [160]. According to Miedema’s rules, compound formation is +favored by large electronegativity differences and similar charge densities [154]. +The electronegativity difference between the two elements greatly increases +under pressure [48, 57] – and because Mg is more compressible, its radius + +a) Na2He +b) Na2EHe +AB stoichiometry +electrostatic attraction +no spacer needed +A2B stoichiometry +attraction + repulsion +spacer needed +Na>EHel +111 +directionSpringer Nature 2021 LATEX template +Article Title +15 +approaches that of Fe when squeezed, thereby increasing the miscibility of the +two elements. As a result, stable Mg-Fe compounds have been computation- +ally [161, 162] and experimentally [163] studied under pressure. The higher +compressibility of K has also been found to favor compound formation with +transition metals such as Ag, even at pressures below which K is anticipated +to undergo an ns → (n − 1)d transition [7]. +Another case-in-point of pressure induced reactivity are the Li-Be alloys +predicted to be stable in the megabar regime [164]. By 20 GPa LiBe2 reaches +the Li-Be convex hull, where it is joined by LiBe4 (shown in Figure 6a) +and LiBe by 80 GPa. At 100 GPa the latter trades its place on the convex +hull with Li3Be. Alignment of strong diffraction peaks with 2kF (twice the +free-electron Fermi wavevector) is suggestive of stabilization through a Fermi +surface-Brillouin zone interaction mechanism, which has been used to explain +the particular stabilities (and electron counts which make them so) of Hume- +Rothery electron phases [165, 166]. Furthermore, at around 82 GPa, an odd +feature emerges in the DOS curve of P21/m LiBe: the base of the valence band +appears as a step-like function, remains flat for ∼4 eV, and sharply increases +once more before more complex features take over. This is linked to a distinct +separation – made possible by the pressure-induced increase in the electroneg- +ativity difference between Li and Be – between high- and low-electron density +planes associated with the Be and Li atoms, respectively, leading to quasi- +2D-like behavior in a geometrically 3D structure. Stabilization through Fermi +sphere interaction with higher zones (referred to as Jones zones from the Mott- +Jones formulation of this mechanism) was invoked to propose a high-pressure +NaAl phase in the NaTl structure type [167] just above 12 GPa. +Another element that does not undergo compound formation with Li at +ambient conditions is Fe. Just above 40 GPa, however, Li3Fe (P6/mmm) and +LiFe (Fd¯3m, NaTl-type) phases are computed to lie on the convex hull [115]. +Some interstitial electron localization appears in Li3Fe, both in P6/mmm +symmetry as well as the P4/mbm symmetry computed to prevail just above +60 GPa which is shown in Figure 6b. Both Li3Fe phases are host-guest lat- +tices, with the Fe atoms lying in larger hexagonal (P6/mmm) and heptagonal +(P4/mbm) channels, and the electron localization is found within the smaller +triangular and square channels. Na3Fe is predicted to adopt this same P4/mbm +phase between 120 and 300 GPa [116]. An additional Li3Fe2 phase with C2/m +symmetry appears on the convex hull by 80 GPa, involving Fe zigzag chains +in combination with alternating Li linear or armchair chains. +Recently, a host of bismuth-containing phases have been found to be sta- +bilized at high pressure. Bismuth is a component in numerous topologically +nontrivial materials [170, 171] and superconducting systems at ambient pres- +sure [172–174], yet under these conditions does not form stable compounds +with many of the transition metals. In combination with the exotic bulk prop- +erties of bismuth, the various magnetic or otherwise electronically nontrivial +properties of the transition metals make for a tantalizing combination in high +pressure experiments. Under pressure, the estimated electronegativity of Bi + +Springer Nature 2021 LATEX template +16 +Article Title +Fig. 6 High pressure enables the mixing of elements which do not form compounds or +alloys otherwise. These include (a) LiBe4, a layered compound of Li and Be [164], (b) Li3Fe, +a host-guest compound with Fe atoms as the guests to a Li-based host lattice [115], and (c) +CuBi [168, 169], one representative of the many bismuth-containing compounds which have +been found at high pressure. CuBi has been experimentally observed. +drops precipitously in comparison to most of the d-block, rendering it more +reactive [48, 57]. Indeed, above 32 GPa, FeBi2 [175, 176] was observed to +form in DAC experiments in the I4/mcm Al2Cu structure type shared by +NiBi2 [177] and MnBi2 [178] (both are stabilized by high pressure although +other Ni-Bi and Mn-Bi phases are accessible at atmospheric pressure), as well +as certain high-pressure transition metal pnictides [179–181]. A second phase, +FeBi3 with Cmcm symmetry, has also been predicted to lie on the convex hull +between 36 and 39 GPa, but this narrow stability range likely hinders synthetic +accessibility [176]. +In fact, high-pressure high-temperature methods have been used to syn- +thesize a wide variety of bismuth-containing compounds including CoBi3 [182, +183], which adopts the Pnma NiBi3 structure type, becomes stable by 5 GPa, +and is a superconductor with a Tc just below 0.5 K. Synthesized binaries +of Cu and Bi include Cu11Bi7 at 3 GPa [184] and CuBi (Figure 6c) at +6 GPa [168, 169], as well as a I4/mmm Cu2Bi phase above 50 GPa [185], which + +a) LiBe4 (P6/mmm) +c) CuBi (Pmma) +b) Li3Fe (P4/mbm)Springer Nature 2021 LATEX template +Article Title +17 +is possibly overtaken by a Cu7Bi2 phase [177]. Of the second-row transition +metals, MoBi2, also in the Al2Cu structure type, has been synthesized above +35 GPa, while evidence for a Mo-Bi bcc-type alloy appeared above 5 GPa [186]. +The Linear Approximation to Enthalpy (LAE), a tool for rapid and com- +putationally cheap evaluation of formation enthalpies, was used to explore +the high-pressure stability of structures in binary ambient-immiscible sys- +tems [177]. In concert with the minima hopping CSP method, several new +phases were found to be stabilized by 50 GPa: PbAs, Si3Al, SiAl, SiAl3, BiSn3 +– yet another bismuthide – In3Fe, Hg3In, HgIn, HgIn3, Hg3Sn, ReSn3, ReBr3, +ReGa, and ReGa3. Only a limited range of stoichiometries (A3B, AB, and AB3) +was sampled, encouraging further investigation into each of these systems – +but now there is preliminary data to suggest fertile ground. +3.5 Geometries and bonding +In the previous sections, we have explored curious electronic interactions made +possible by external pressure and compound formation between unexpected +species. Here, we shift our focus to the particular geometrical arrangements +that emerge in materials under pressure. With higher density, atoms are forced +into closer proximity – the possibility of electride formation notwithstanding – +which can promote multicenter bonding in both electron-poor and electron-rich +contexts as coordination numbers increase [13]. When electron-precise species +are closely bunched under compression, electron-deficient multicenter bonding +can emerge as the constituent electrons are needed to span more bonding +interactions. +Bond symmetrization is a frequent secondary consequence of compression: +an asymmetric fragment forced to occupy a progressively smaller space has +less room for asymmetry. Often, this leads to a collapse into a symmetric +and multicentered bonding regime, as was predicted for water ice under pres- +sure by Pauling [187]. He suggested that the intermolecular hydrogen bonds +between adjacent water molecules would shorten with pressure [187], eventu- +ally becoming equivalent with the intramolecular O-H bonds, as illustrated +in Figure 7a. This prediction was verified experimentally upon the discovery +of ice X, where the lone pairs of the oxygen atoms are used to form addi- +tional covalent bonds, rendering the oxygen atoms tetrahedrally coordinated +by hydrogens in a diamond-like network [188, 189]. Pressure-induced hydro- +gen bond symmetrization has also been observed in computations on hydrogen +halide systems such as HF, HCl and HBr [190], as well as in the record-breaking +superconductor Im¯3m H3S [191]. +Small homoatomic clusters alien to the 1 atmosphere pressure-trained mind +are found or predicted for other elements in high pressure crystal structures. +The wide structural variety has been explained by the increased stabiliza- +tion of homonuclear bonds as compared to more polar or ionic bonds under +pressure [104], favoring single-element clustering. An example of a compound +containing novel homonuclear motifs is Pnma NaCl3 (Figure 7b), computed +to be stable from 20 to 48 GPa, featuring a linear Cl− +3 anion reminiscent of + +Springer Nature 2021 LATEX template +18 +Article Title +the more familiar triiodide I− +3 [9]. Another such motif is the pentazolate N− +5 +ring, which can store more energy than the related azide anion N− +3 , but is +challenging to synthesize at ambient pressure [192, 193]. This species, ubiqui- +tous in high-pressure phases, is predicted to be a constituent of LiN5 [194–197] +– a phase that has been successfully quenched to ambient conditions after +synthesis at 45 GPa [198] – to sodium pentazolates NaN5 and Na2N5 [199], +CsN5 [200], CuN5 [201], MgN10 and BeN10 [202], ZnN10 [203], BaN5 and +BaN10 [204], SnN20 [205], and IrN7 [206]. Polynitrogen chains feature in many +proposed high-pressure compounds of Cs [200], Fe [207], Zn [203], Ba [204], +Sn [205], Cd [208], Gd [209], and Ta [210], the last of which has been experi- +mentally realized. High pressure also facilitates the formation of silicon clusters +in predicted phases including Si4 squares in CaSi [211], extended networks and +clathrate-like cages in silicides such as CsSi6 [212], MgSi5 [213], and several +lithium silicide compounds [214]. +Fig. 7 Geometrical and bonding adaptations at high pressures. With pressure (a), cova- +lent O-H and intermolecular hydrogen bonds between separate water molecules in the ice +VIII phase equilibrate to yield the symmetric ice X phase [188, 189]. Pressure coincides +with the appearance of strange motifs including (b) linear trichloride anions in Pnma +NaCl3 [9], (c) clusters of Ge dumbbells (inset) in I4/mmm BaGe3 [215, 216], and (d) Li8H +“superatom-like” building blocks (inset) in Abm2 Li5H [217]. Except for Li5H, all have been +experimentally realized. + +b) NaCl3 (Pnma) +a) Bond symmetrization in ice +differentiated +symmetric +O-H bonds +ice VIII (I41/amd +ice X (Pn3m) +)BaGe3 (I4/mmm) +d) Li5H (Abm2) +Ge2 dumbbell +Li8H +clusters +"superatom"Springer Nature 2021 LATEX template +Article Title +19 +Tetrel clusters comprise a family of polar intermetallic I4/mmm symme- +try compounds formed from alkaline earth or rare earth metals and group +14 tetrels in a 1:3 ratio. An example of these isotypic compounds, BaGe3, is +shown in Figure 7c. The clusters within it may be described as tetrel dumb- +bells condensed into cubes, which are capped on four equatorial faces by +additional dumbbells shared with a neighboring cube, forming a loose three- +dimensional network. This structure has been experimentally observed in Ca, +Y, and Lu silicides [218] (and later identified in a CSP investigation of the +Y-Si system [117]), and a related distorted I¯42m BaSi3 phase has also been +synthesized [219]. Alkaline earth trigermanides CaGe3 [220], SrGe3 [221], and +BaGe3 [215, 216] have also been found to adopt this structure. The distribution +of electrons within the clusters aligns with two-center two-electron bonds along +the tetrel dumbbells, with multicenter interactions between the tetrel and rare +earth/alkaline earth [215, 218, 222]. Superconductivity has been measured in +some of these compounds, albeit at low temperatures [216, 218], augmented +by predictions from first principles calculations [104, 117, 222]. Additionally, +the stability conferred by the strong covalently bonded networks permits some +of these phases, synthesized at high-temperatures and high-pressures, to be +recovered at ambient conditions [219, 222]. A similar example is presented by +elemental carbon – diamond is its ground state at high-pressures, but due to +the immense strength of its sp3 covalently bonded network, the energetic bar- +rier for its transition to the lower-enthalpy allotrope graphite is too high and +it persists “forever” under ambient conditions. Furthermore, laser-driven ramp +compression studies of carbon to 2 TPa have found that carbon stubbornly +maintains the diamond structure well beyond its predicted high-pressure sta- +bility limits [223], as the barriers to breaking the sp3 bonds remain large under +pressure. +Another example of unique clusters predicted to form only at high pressures +are found within a family of lithium subhydrides [217]. Computations uncov- +ered two nearly isoenthalpic Li5H phases that had the most negative enthalpies +of formation. Both were built of Li8H units that behaved as superatoms anal- +ogous to similar units in synthesized Rb9O2 and Cs11O3 suboxides [224]. One +of these, with Abm2 symmetry, is shown in Figure 7d. The Li8H cluster, a dis- +torted bicapped trigonal antiprism of Li encapsulating a single H atom, has +one electron in excess of the closed-shell octet and thus behaves as a superalkali +atom. +In addition to the well-known H2 molecular units and H− hydridic species, +hydrogen atoms can form other distinct clusters. One of these, the trihydrogen +cation, H+ +3 , is in fact one of the most abundant species in the universe – but it +is also largely relegated to interstellar space and the atmospheres of gas giant +planets [225]. High pressure crystal chemistry offers another opportunity. The +halogen polyhydride Cc H5Cl [226–228], predicted to become stable above +100 GPa, contains slightly distorted H+ +3 clusters with H-H distances of 0.74, +0.97 and 1.01 ˚A [226]. By 300 GPa the three distances converge to 0.87-0.88 ˚A, +yielding a nearly-perfect equilateral triangle. With sufficient pressure, this H+ +3 + +Springer Nature 2021 LATEX template +20 +Article Title +unit interacts with a neighboring H2 molecule forming a twisted bowtie-like +loosely interacting H+ +5 motif [226, 228]. Metastable predicted H2F, H3F, and +H5F species, as well as H5Br also contain this triangular H+ +3 cation [227], as +does the metastable P1 LiF4H4 [229]. +With two extra electrons, the trihydride anion, H− +3 , prefers a linear +arrangement involving a three-center four-electron bond. Quantum chemical +calculations have shown that the ground state geometry of the isolated tri- +hydride anion possesses one H-H bond that is substantially longer than the +other (2.84 vs. 0.75 ˚A), with the transition state between the H-H· · ·H and +H· · ·H-H configurations corresponding to the symmetric case [230]. Neverthe- +less, certain predicted high-pressure hydrides of the heavy alkali metals K [66], +Rb [231], and Cs [232] (as well as the alkaline earth metal Ba [233]) feature an +H− +3 anion symmetrized via pressure. Synthesized NaH7 is thought to contain +an asymmetric linear H− +3 motif [234]. Linear H− +3 units are also predicted to +appear in various indium [235] and lithium polyhydrides [236]. +Scandium polyhydrides, meanwhile, are predicted to feature five-membered +rings of hydrogen atoms in various arrangements. In I41/md ScH9, which lies +on the convex hull around 300 GPa [237], strips of edge-sharing H5 pentagons +are stacked perpendicular to one another along the c axis, linked by ver- +tices. The strips are separated by additional H atoms in molecular H2 units. +Around 250 GPa, ScH10 adopts a Cmcm structure in which H5 pentagons +are grouped into sets of three, sharing edges and a single common vertex. +This phase is nearly isoenthalpic with another ScH10 structure with the same +(H5)3 “pentagraphenelike” clusters but arranged in P63/mmc symmetry [238]. +ELF demonstrates bonding within the H5 units [237, 238]. This same penta- +graphenelike structure is expected to lie on the Lu-H convex hull at 300 GPa +and to be very near the Hf-H and Zr-H hulls [238]. For higher hydrogen con- +tents yet, ScH12 is predicted to be built of stacked strips of edge-sharing H5 +pentagons spaced by Sc [237]. +Indeed, the first, most simple, element does not like to be outdone! One +more class of high-pressure hydrogen rich materials whose prediction sparked +tremendous experimental synthetic efforts are the so-called “metal super- +hydrides”. The reason why scientists have pursued them in earnest is the +prediction, verified by recent experiments, of conventional superconductivity +at temperatures approaching those experienced in a cold room (the Tc of LaH10 +is about 10 ◦F – January in Siberia), or a crisp fall day (the Tc of C-S-H is +about 60 ◦F), albeit still at very high pressures! CSP investigations into the +high-pressure Ca-H system revealed a curious Im¯3m CaH6 phase (Figure 8a) +stable above 150 GPa in which the Ca atoms were arranged in a bcc lattice +and the H atoms condensed into a sodalite-like H24 framework [73]. All H-H +distances were equivalent, 1.24 ˚A at 150 GPa, and ELF analysis confirmed +weak covalent bonding between the H atoms. This phase, a good metal, was +predicted to exhibit large electron-phonon coupling, and indeed first principles + +Springer Nature 2021 LATEX template +Article Title +21 +calculations estimated a superconducting transition temperature Tc of 220- +235 K at 150 GPa. Subsequent synthetic exploration led to measurement of Tc +over 200 K at 160-170 GPa for CaH6 [71, 72]. +Fig. 8 Clathrate-like metal hydrides predicted then synthesized at high pressures. Their +remarkable superconducting properties are tied to strong electron-phonon coupling. Sev- +eral metal-hydrogen stoichiometries adopt these so-called “superhydride” motifs, including +Im¯3m MH6, exemplified by CaH6 [71–73], (b) P63/mmc MH9, exemplified by YH9 [35, 69], +and (c) Fm¯3m MH10, exemplified by LaH10 [34, 35, 37, 70]. +The computational discovery of CaH6 was shortly followed up by fruit- +ful theoretical investigations into related metal-hydrogen systems, turning up +isostructural phases for Mg [239], Sc [237, 240, 241], Y [34, 35, 242], Pu [243], +Tb [244], Eu [245], and Pm-Lu [246]. Structures that are distortions of this +high symmetry phase have been predicted as well. This includes a tetragonally- +distorted I4/mmm ZrH6 variant [247], along with an R¯3m phase in which +opposite hexagonal faces of the H24 cubic sodalite framework are opened for +SrH6 [248–250] and LaH6 [34, 35], although the latter may not lie on the +convex hull. Imm2 BaH6 [233], which contains some of the H− +3 trihydride +anions explored above, can be thought of as a highly fragmented version of +the sodalite framework. The role of distortions of the high-symmetry Im¯3m +structure adopted by CaH6 tends to reduce the density of states at the Fermi +level, EF , thereby also lowering Tc. The origin of such distortions has recently +been investigated using the lens of DFT-Chemical Pressure [251]. +Higher hydrogen content allows for other clathrate-like arrangements of +hydrogen, from P63/mmc YH9 [35] (Figure 8b) to Fm¯3m LaH10 [34, 35] +(Figure 8c). Distorted versions of these two structures have also been predicted, +including C2/m CaH9 [252] and P1 Eu4H36 [245], as well as R¯3m CaH10 [252] +and AcH10 [253]. In the case of LaH10, quantum anharmonic effects were found +to be key in stabilizing the Fm¯3m structure over less symmetric variants [254]. +Other more complex clathrate-like hydrogenic frameworks have been predicted +as well. One example is a diamond-like lattice of Mg@H28 clusters interca- +lated with Li@H18 units, which comprise the Fd¯3m Li2MgH16 phase. This +compound is an example of a “hot” superconductor whose estimated Tc, +351 K at 300 GPa, is well above room temperature [255]. Such materials are + +a) CaH6 (Im3m) +b)YH9 (P63/mmc) +C) LaH10 (Fm3m)Springer Nature 2021 LATEX template +22 +Article Title +under intense speculation and investigation for their promise towards achieving +room-temperature superconductivity, as described in Section 4.3 below. +4 Superconductivity +The 1911 discovery of a phenomenon in which a substance’s resistivity can +plummet to zero [256] sparked countless investigations and resulted in a Nobel +prize for Heike Kamerlingh Onnes, as well as a number of future Nobel prizes +(directly or indirectly). The mechanism of superconductivity, and the search +for new superconducting materials, has fascinated scientists for over a cen- +tury. A key parameter for superconductors is the critical temperature, Tc, +below which a material becomes superconducting. For a number of illustra- +tive superconducting materials, Tc is plotted against the pressure at which +they are stable in Figure 9. Of course, practical applications of superconduc- +tivity are limited if temperatures very near 0 K are required, and for some +decades the highest known Tc values lingered in the low twenties [257, 258]. +This sparked debate regarding a possible natural “cap” on superconductivity +around these temperatures [259]. However, a family of cuprates whose super- +conducting mechanism has yet to be explained were the first to break the +liquid nitrogen barrier [260], achieving Tcs over 160 K in the case of pressurized +HgBa2Cam−1CumO2m+2+δ [261]. +The 2001 discovery of superconductivity up to 39 K in MgB2 [263] pro- +vided experimental evidence that non-cuprate phases could be promising +superconductors, despite the fact that they deviated from the collection of +empirical rules enumerated by B. T. Matthias in the 1950s and 1960s [272]. +These ranged from favorable valence electron concentrations to a general +distrust for theorists. MgB2 belongs, with early and long-term record hold- +ers Nb3Sn and Nb3Ge, as well as the clathrate-like hydrides discussed in +Section 3.5, to the family of “conventional” superconductors whose mecha- +nism is thought to follow the theory propounded by Bardeen, Cooper, and +Schrieffer in 1957 [273, 274]. This discovery revitalized interest in conventional +superconductors [275], leading researchers to wonder what the trajectory of +superconductivity research would have looked like had the 1957 measurements +of the heat capacity of MgB2 [276] captured the discontinuity that appears +upon the superconducting transition. +Within BCS theory, two electrons of opposite momentum and spin that are +within ±ℏωcut (an energy in line with the phonon energies) of the Fermi surface +may, at long distances, overcome Coulombic repulsion and experience a net +attractive potential when the lattice is polarized through phonon vibrations. +This forms a Cooper pair, a composite boson of weakly interacting species. +Thermal energy can break the Cooper pairs, with Tc describing the temper- +ature at which this occurs and the superconducting state is destroyed. From + +Springer Nature 2021 LATEX template +Article Title +23 +Fig. 9 Critical superconducting temperature, Tc, plotted against pressure for selected mate- +rials. Data points based on experimental measurements of Tc are plotted with filled circles, +while those estimated via theoretical calculations are plotted with empty circles. Elemen- +tal Tcs are given in brown, intermetallics in green, cuprates (belonging to the non-BCS +unconventional family) in orange, and hydrides in blue. The boiling point of liquid nitro- +gen and room temperature are provided to guide the eye, with a star marking the “holy +grail” of room-temperature superconductivity at ambient pressure. Data was collected from +references [34, 35, 37, 70, 255, 256, 258, 260–271]. +this construction, the Tc for a material can be estimated by +kBTc = 1.14ℏω exp +� −1 +NF V +� +(1) +where ω is the average phonon energy, NF is the single spin electronic den- +sity of states at EF , and V is the pairing potential between two electrons +resulting from the electron-phonon interaction. This suggests that high Tc is +correlated with a large NF (a feature potentially tunable via judicious doping), +strong coupling between electrons and phonons, and high phonon frequencies. +A frequently used semiempirical formula to estimate the Tc of a conventional +superconductor is the Allen-Dynes modified McMillan equation [277–279]: +T c = ωln +1.2 exp +� +− +1.04(1 + λ) +λ − µ∗(1 + 0.62λ) +� +. +(2) + +500 +Li2MgH16 +o Computed only +450 +Experimentally observed +350 +Elements +Cuprates +Intermetallics Hydrides +oYH10 +300 +C-S-H +(K) +Room temperature +YH9 +250 +LaHi0 +200 +H3S +O LaBeH: +HgBaCaCuO +150 +OKB2Hs +OLaBHs +100 +YBaCuO +PHn +Liquid nitrogen +CeH9 +50 +MgB2 +-Nb3Ge +Ca +1Hg +Nb +0 +0 +50 +100 +150 +200 +250 +300 +Pressure (GPa)Springer Nature 2021 LATEX template +24 +Article Title +where λ is the electron-phonon coupling constant, ωln is the logarithmic aver- +age phonon frequency and µ∗ is the Coulomb repulsion parameter, which is +typically treated semiempirically. An approximate – and illustrative – formula +to estimate λ was proposed by Hopfield as [280] +λ = NF ⟨I2⟩ +M⟨ω2⟩ +(3) +where ⟨I2⟩ are the electron-phonon matrix elements averaged over the Fermi +surface, M is the atomic mass, and ⟨ω2⟩ the mean phonon frequency. To +increase λ, then, NF and the electron-phonon matrix elements should be +increased. Converse to expectations from Equation 1 and Equation 2, where +the average phonon energy was directly proportional to Tc, here an increase +in ⟨ω2⟩ serves to decrease λ. In the denominator of Equation 3, ⟨ω2⟩ and M +will naturally tend to counteract one another, as increases in atomic mass +lead to softer phonon frequencies. In fact, evidence suggests that Tc frequently +increases at the edge of dynamic instability, where soft phonons promote strong +coupling [281]. +In the following sections, we describe the effect of pressure on the propensity +for superconductivity of the elements, hydrogen in particular. We end by dis- +cussing families of hydrogen-rich phases that are extremely promising towards +achieving the once-distant goal of room-temperature superconductivity (albeit +potentially only at very high pressures). +4.1 The elements +Most of the elements in the periodic table can be superconducting given +the right conditions. This includes, so far without exception, rather low +temperatures. However, external pressure can affect the Tc, or even induce +superconductivity in some elements [282–285]. In fact, calcium at ambient +pressure is not a superconductor and achieves Tc = 29 K at 216 GPa [264]. +In comparison, the highest elemental Tc at ambient pressure is 9.2 K for +niobium [262], while only lithium among the alkali metals [286], and only +beryllium [287] among the alkaline earth metals are known to superconduct +at ambient pressure, both well under 0.1 K. Of the fifty-four known supercon- +ducting elements of the periodic table, only thirty-one are superconductors at +ambient pressure. +Early indications regarding the ability of pressure to either enhance or +suppress superconductivity [288–290] were less than encouraging. For simple +metals whose electronic structure aligns with a mostly free-electron model, +such as Zn, Cd, Hg, and the group 13 metals, applied pressure serves to sup- +press what superconductivity is present at ambient pressure [289, 290]. In such +free-electron-like metals, the effect of pressure is to broaden electronic bands +and increase phonon frequencies due to the stiffer lattice. Band broadening +reduces the electronic density of states at EF , while a stiffer lattice is correlated + +Springer Nature 2021 LATEX template +Article Title +25 +with weaker coupling between electrons and phonons, both effects being detri- +mental to the superconductivity of a system. At ambient pressure, the alkali +metals behave as free-electron metals, but as we have seen in Section 3.2, with +pressure their electronic structure rapidly diverges from these expectations. +For Li, Tc is highly pressure-dependent and reliant on complex crystal +chemistry [291–293]. Like all of the alkali metals, it adopts a bcc crystal +structure at ambient pressure, which in short order transitions to the fcc struc- +ture. Up to 8 GPa, the increase in pressure is reflected in an increase in +phonon frequencies, typical for a stiffening lattice, but with further pressure +the phonons become softer and just above 30-40 GPa, imaginary modes related +to a structural instability appear. This pressure corresponds to another phase +transition, this time to the hR1 (R¯3m) structure, and shortly thereafter to +the cI16. Maxima in Tc are associated with the onset of dynamical instability, +as the very soft phonon motions strongly bolster the electron-phonon cou- +pling [281, 294]. From its mostly spherical character at 0 GPa, where lithium +adopts a bcc crystal structure, the pressure-induced 2s → 2p electronic tran- +sition drives an increasingly anisotropic Fermi surface featuring hot spots of +especially strong coupling [295], losing free-electron-like behavior and lead- +ing to Fermi surface nesting (FSN) [296–298]. Phonon softening, in particular +along the Γ → K path, is accompanied by enhancement of the electron-phonon +coupling [299, 300], with the result that Tc grows from practically zero to a +maximum of ∼20 K. This value is among the higher elemental Tcs, as a result +of the pressure-induced electronic transitions in lithium. Following the struc- +tural transition to the cI16 phase, lithium remains superconducting but Tc +decreases due to a reduction of the FSN and concomitant smaller electron- +phonon coupling [299, 301, 302]. At high enough pressures, lithium undergoes +a metal-semiconductor transition – and eventually goes back to being a (poor) +metal upon transitioning to the oC24 (Cmca) phase [303], but one nonetheless +predicted to be superconducting with an estimated Tc of 14 K at 200 GPa [304]. +In cesium [305, 306] and rubidium [110, 307], the onset of superconductiv- +ity with pressure is associated with the oC16 (Cs-V and Rb-IV) structures, +alongside the ns → (n − 1)d transition [307]. An increase in d-character in +the electronic states at EF is correlated with higher Tc in the transition met- +als [308, 309] and applies here as well to the heavier alkali metals – which, as +we have seen, behave under pressure as transition metals themselves. +4.2 Hydrogen +Vitaly Ginzburg, awarded the Nobel Prize in Physics in 2003 for his work in +superconductivity and superfluidity, formulated a list of, in his view, the thirty +most pressing problems for physics in the 21st century [310, 311]. Following +controlled nuclear fusion, the second and third items on this list were high- +temperature (room-temperature) superconductivity and metallic hydrogen. +These problems are not unrelated. +In 1926, J. D. Bernal proposed that under sufficient pressure hydrogen +would transition to a metallic state, but it took nearly a decade for pen to + +Springer Nature 2021 LATEX template +26 +Article Title +be put to paper by Wigner and Huntington [312]. Their 1935 suggestion that +hydrogen could be metallized by 25 GPa, estimated using a series of assump- +tions regarding crystal structure and compressibility, proved to be an immense +underestimate. In 1968 Neil Ashcroft explicitly linked the quest for metal- +lic hydrogen with the quest for superconductors with higher Tcs, with the +suggestion that metallic hydrogen itself would be quite a fantastic supercon- +ductor [313]. Hydrogen, being the lightest element, can possess the highest +frequencies (as a diatomic molecule) and experience a large electron-phonon +coupling due to the lack of screening by core electrons. Moreover, in the metal- +lic state its DOS at EF is thought to be quite high, making for a very attractive +material. +The pressure required to metallize hydrogen, however, is in the multi- +megabar range. Claims of metallic or semimetallic hydrogen have been made +for DAC experiments at very high pressures and low temperatures, toeing the +line of the practical limits of these techniques [314–316]. Complicating the +picture, different experiments used different measures to characterize hydro- +gen’s transition to metallicity, from vibrational spectroscopic techniques such +as IR [316], optical measurements such as reflectance and opacity [314], and +resistivity measurements [315], as well as different scales to calibrate pres- +sure. At times, this led to seemingly contradictory results, prompting questions +regarding experimental accuracy and reproducibility [317–319]. Ab initio cal- +culations taking into account the quantum fluctuations of the hydrogen nuclei, +however, can reconcile some of these differences, finding in the C2/c-24 high- +pressure phase closing of the electronic gap (and transition to metallicity) +before the closing of the optical gap (and transition from transparency to +reflectivity) [320]. +Additionally, the impractically low Tcs of ambient pressure materials are +then traded for a much higher Tc in pure metallic hydrogen, but at imprac- +tically high pressures! In fact, ab initio modeling suggests progressive jumps +in Tc with a transition from the molecular (estimated Tc = 356 K near +500 GPa) to the atomic phase (Tc increasing to 481 K ca. 700 GPa), and +with an atomic-atomic phase transition at ∼1-1.5 TPa driving up λ and +resulting in an immense estimated Tc= 764 K [321]. To address the hydrogen +metallization problem, Ashcroft proposed another strategy – instead of pure +hydrogen, hydrogen-rich metallic alloys could be targeted as putative super- +conductors [322]. The presence of additional atoms in the hydrogen matrix +would confer a chemical precompression, thereby lowering the external pres- +sure required to reach the metallic state. Under ambient conditions, the crystal +structures adopted by metal hydrides are largely subject to the dictates of +balanced oxidation states, hence alkali metal hydrides assume the rock salt +structure, and hydrides of +2 metals favor fluorite or Co2Si structures, while +trivalent metal hydrides tend towards the BiF3 structure, and so on [323, 324]. +Much higher hydrogen content would be needed for such hydrogen-rich alloys, +as suggested by Ashcroft, and furthermore many of the resulting structures +might differ greatly from anything observed at ambient conditions. Defying the + +Springer Nature 2021 LATEX template +Article Title +27 +recommendations of Matthias, the simultaneous and serendipitous advances +in CSP methods meant that theoreticians were well poised to answer this call! +4.3 Clathrate-like hydrides +They were successful. As described above, calculations on the high-pressure +Ca-H system located the CaH6 phase described in Section 3.5 [73], in which +Ca atoms are embedded into a hydrogenic sodalite-like clathrate framework. +The strong electron-phonon coupling predicted for Im¯3m CaH6 can be traced +to breathing and rocking modes of the square H4 units of the sodalite frame- +work [73]. The molecular orbital diagram for such an H4 square has, above +a filled bonding state, a half-occupied degenerate non-bonding state. Assum- +ing full ionization (integrated charges within atomic basins according to the +Quantum Theory of Atoms in Molecules indicate roughly 1 electron per Ca +is transferred to the hydrogen network), these orbitals accept the roughly 1/3 +electron per H transferred from the Ca atom. This favors symmetry-breaking +Jahn-Teller distortions – key contributions to the electron-phonon coupling – +that lift the degeneracy. +Similar clathrate-like hydrides, as outlined in Section 3.5, were rapidly +predicted in several other systems. Im¯3m YH6 has been synthesized with a +measured Tc of 224 K at 160 GPa [325]. Hydrides with even higher hydrogen +content have been synthesized as well, including a P63/mmc YH9 phase with +a Tc of 262 K at ca. 180 GPa [69] (a subsequent study reported a slightly lower +Tc of 243 K at 200 GPa [326]), and an isotypic CeH9 phase whose Tc has only +been predicted (57 K at 88 GPa and up to 100 K at 130 GPa [35], or 105- +117 K by 200 K [267]). The relatively low pressure required to stabilize CeH9 +has been ascribed to strong chemical precompression from the delocalized Ce +4f electrons [327]. The reported Tcs for Fm¯3m LaH10 (250 and 260 K at 170 +and 185 GPa, respectively [37, 70]), are in line with theoretical predictions of +257-274 K at 250 GPa [34]. Isotypic Fm¯3m YH10 is computed to be a room +temperature superconductor with a Tc of 305-327 K at 250 GPa [34]. However, +YH10 has thus far eluded synthetic efforts. Partial doping with lanthanum +appears to be one strategy to stabilize YH10: a series of ternary (La/Y)H10 +phases have been experimentally observed, with measured Tc = 253 K [38]. +Although the Tcs of many clathrate-like hydrides are stunning, these +phases will surely decompose well above atmospheric pressures! CeH9 is +remarkable for the comparatively low, sub-megabar, pressures at which it +maintains dynamic stability [267]. In an attempt to preserve the loosely- +bound hydrogenic clathrate frameworks that are associated with such strong +electron-phonon coupling to lower pressures, one promising strategy involves +the addition of a third element in an attempt to further chemically precom- +press the hydrogenic lattices. The Fm¯3m LaBH8 phase, which is predicted to +maintain dynamic stability down to 40 GPa [269], can be derived from LaH10 +by removing two hydrogen atoms per formula unit and placing boron atoms +into the center of H8 cubes that are empty in LaH10 [271, 328]. LaBH8 has an +estimated Tc of 126 K at 50 GPa [269] – and the isostructural LaBeH8 phase + +Springer Nature 2021 LATEX template +28 +Article Title +is predicted to achieve a Tc of 183 K at 20 GPa [271]. Other XYH8 phases +have been proposed, with a variety of possible elemental combinations ripe for +tuning stability and properties [271]. A second possibility is afforded by the +XY2H8 phases that can be constructed by leaving the H8 cubes empty, but +stuffing the center of H4 tetrahedra instead. KB2H8 (dynamically stable to +12 GPa [270]) and LaC2H8 (dynamically stable down to 70 GPa [329]) are two +representatives of this structural arrangement, with estimated Tcs of 134-146 +and 69 K, respectively. +Key to the success of the clathrate-like hydrides is the maintenance of +loosely-coordinated networks of hydrogen, rather than condensation into H2 +molecules. The effect of H-H interatomic distances on superconductivity can +be seen in the MH4 hydrides, which adopt the I4/mmm structure shared with +ThCr2Si2 [330, 331]. Hydrogen occupies two inequivalent sites in the ThCr2Si2 +structure – the apical Ha (Wyckoff position 4e, Si) and basal Hb (4d, Cr). A +plethora of metal hydrides have been predicted or synthesized in this structure +type under pressure, including Ca [73, 252, 332] and Sr [248, 249], Sc [237, +240, 241], Y [34, 242], and Zr [247] and rare earths La [34], Ce [35, 267, 333], +Pr [35, 334, 335], Pu [243], Tb [244], Eu [245], Nd [35, 336], and Th [337, 338], +making systematic study enticing and useful [331]. With metal oxidation states +ranging from +2 to +4, the formulas of these compounds can be written as +Mx+(H− +b )2(Ha)(x−2)− +2 +, with hydridic Hb and a range of charges possible on the +Ha atoms. The Tcs of these phases are correlated with the length of the Ha- +Ha contacts, which can behave anywhere from covalently bound H2 units to +fully dissociated hydridic anions depending on the metal atom – similar to the +behavior of the X-X bond in ThCr2Si2-type AB2X2 phases [339]. The size of +the metal atom can be relevant, as larger atoms will stretch the Ha-Ha contact +through purely steric interactions, but more important is the valency of the +metal atom. Electron transfer from the electropositive metal into the Ha-Ha +motif directly populates the H2 σ∗ +u antibonding orbitals, but is also driven by +a Kubas-like two-pronged mechanism of H2 σg → M d donation, and M d → +H2 σ∗ +u back-donation. With enough H2 σ∗ +u population, the Ha atoms behave +in a hydridic fashion lowering the Tc, as seen in ZrH4 [247] and ThH4 [337]. +Donation of sufficient electron density to weaken, but not fully break, the Ha- +Ha bonding interaction results in a much higher DOS at EF and enhanced Tc, +as in YH4 [242]. +4.4 Covalent hydrides +The first hydride to top the charts, as it were, was not of the metal clathrate- +like family, but instead came from attempts to metallize H2S. Theory identified +an H2S compound that was computed to possess a Tc of 80 K at 160 GPa [340]. +Experimental confirmation followed shortly thereafter, finding a phase with +Tc < 100 K, but in the process a higher-temperature preparation method +yielded a sample with a Tc of 203 K at 150 GPa [266]. A few years before, +synthetic exploration into the (H2S)2H2 stoichiometry found a phase stabi- +lized by pressure-induced hydrogen bonding above 3.5 GPa [341]. This inspired + +Springer Nature 2021 LATEX template +Article Title +29 +CSP investigations of the H3S stoichiometry, which found an R3m phase with +Tc = 155-166 K at 130 GPa [342]. By 180 GPa this structure transitioned to +one with Im¯3m symmetry (Figure 10a) for which the estimated Tc was 191- +204 K at 200 GPa. Serendipitously the experimental [266] and theoretical [342] +manuscripts appeared at nearly the same time. Subsequent XRD studies sup- +ported the identification of the 203 K superconductor as Im¯3m H3S [343], +though other structures have been proposed [344–349]. +A host of studies on the H3S superconductor have followed, exploring the +isotope effect, role of anharmonicity, and possible quantum effects [191, 350– +359]. The inclusion of quantum nuclear motions lowered the pressure where the +less symmetric R3m phase was predicted to transition to the Im¯3m structure +with symmetric H-S bonds into the range of pressures where high Tcs had been +measured [191]. One of the most striking features of the electronic structure +of H3S are a pair of van Hove singularities bracketing EF [360, 361]. Shifting +the position of EF , potentially by doping, could increase the number of states +that can participate in the electron-phonon coupling mechanism, and therefore +increase the Tc of the system. +Doping is a common strategy used for precise tuning of EF , and com- +putations using the virtual crystal approximation (VCA) suggested that the +addition of a little bit of phosphorus, carbon, or silicon could raise the Tc +into the room-temperature regime > 280 K [362–364]. In the VCA, alchem- +ical pseudoatoms are constructed from weighted averages of the component +atom potentials. The resulting chemical chimeras, however, cannot accurately +model the local structural and electronic effects that arise when one atom +is replaced with an entirely different element. This throws off, in particular, +the very dynamical response properties that one must calculate carefully to +obtain reasonable estimates of Tc. Additional studies based on actual doped +H3S models constructed as supercells have sought to explore the local effects +of doping [365–368], although the calculation of dynamical properties of the +requisite large unit cells can be prohibitively expensive. +One particularly promising system involved the addition of carbon to the +H3S lattice by way of methane intercalation [369, 372, 373]. Stoichiometries +that are a linear combination of CH4 and H3S (the most simple of which is +CSH7) proved especially interesting. They yielded a variety of dynamically +stable (although energetically metastable) structures, which differed in the ori- +entation of the methane molecules encapsulated in the H3S lattice. Some of the +highest Tcs predicted for these phases were 181 K at 100 GPa for I¯43m [372], +and 181-194 K for the R3m symmetry structure [369] shown in Figure 10b. +Independently, photochemical synthesis in the C-S-H system yielded the +first report of room-temperature superconductivity, achieving a Tc of 288 K +at 267 GPa [268]. This report has inspired a slew of follow-up work and much +debate [371, 374, 375]. XRD analysis performed at pressures below the pur- +ported room-temperature superconducting transition [370, 376] is consistent +with the Al2Cu geometry (as well as an orthorhombic Pnma variant) associ- +ated with CH4-H2 [377] and H2S-H2 [341] (Figure 10c). This may suggest an + +Springer Nature 2021 LATEX template +30 +Article Title +Fig. 10 The covalent hydride Im¯3m H3S [266] (a) represents a breakthrough in high- +Tc conventional superconductivity. Peaks in the electronic DOS near EF have prompted +numerous investigations on doped versions of H3S, discovering phases such as the methane- +intercalated (CH4)H3S = CSH7 [369] (b), among many others. A recent synthesis of a +room-temperature superconductor consisting of carbon, sulfur, and hydrogen generated even +more momentum, with diffraction analysis performed at pressures below those where the +room-temperature superconducting transition were observed, revealing the presence of a +phase based on the I4/mcm Al2Cu-like structure adopted by the van der Waals (H2S)H2 +phase [341, 370, 371] (c). +overall stoichiometry of [(CH4)2H2]x[(H2S)2H2]y for the room-temperature C- +S-H superconductor, although subsequent phase transitions at higher pressure +to the high-Tc superconducting phase cannot be ruled out. In fact, additional +studies indicate just such a structural transition occurs to form the room- +temperature superconducting phase, with indications of methane signatures +in the Raman spectra [371]. As was the case for the binary H3S system, it +appears that a panoply of metastable phases may be accessible by slight vari- +ations on synthetic procedure, in particular on carbon content [378], offering +plenty of space for further experimental and theoretical discoveries. +In addition to sulfur-based covalent hydrides, phosphorus hydrides have +sparked interest after compression of a phosphine (PH3) sample yielded a +material that became superconducting at 30 K at 83 GPa, increasing to 103 K +at 207 GPa [265]. The structure and composition of the responsible phase or +phases was unclear, prompting an array of follow-up studies levying CSP tech- +niques to identify plausible compounds [379–383]. Pressure was found to drive +the decomposition of phosphine into a variety of products with stoichiome- +tries including PH, PH2, PH3, and more. A predicted C2/m PH3 phase [380] +featuring P-P bonds (in contrast to the H3S superconductor, which has no +S-S bonding) was estimated to be superconducting below 83 K at 200 GPa, +in line with the experimental values. Another study suggested that multiple +metastable decomposition products of phosphine, including those with PH and +PH2 stoichiometries, might in combination be responsible for the observed +superconductivity [382]. PH2 phases with C2/m and I4/mmm symmetries, +differing by a tilt in the component H-P-H moieties, had estimated Tcs of +76 and 70 K, respectively [381]. Later, another set of PH2 phases were pro- +posed consisting of simple cubic layers of phosphorus capped with hydrogen + +a) H3S (Im3m) +b)(CH4)H3S(R3m +C)(H2S)2H2(I4/mcm)Springer Nature 2021 LATEX template +Article Title +31 +atoms and further intercalated with H2 molecules acting as Coulombic spac- +ing agents [383]. At 80 GPa, these structures had estimated Tcs ca. 30 K, +similar to the values that were measured. Raman spectroscopic measurements +provided evidence for phosphine dimerization coupled with dehydrogenation +under pressure, yielding compositions such as P2H4 and P3H6 [384, 385]. In +these phases, low temperatures were required to maintain stability at multi- +megabar pressures. The Tc of a predicted C2/m P4H6 phase was estimated to +be 67 K at 200 GPa [385]. +The plethora of metastable P-H compounds under pressure has prompted +computational investigations into ternary systems containing phosphorus and +hydrogen. Above 250 GPa, an R¯3 LiP2H14 phase, consisting of P@H9 clusters +spaced by Li atoms as well as isolated H atoms, achieves an estimated Tc +of 169 K at 230 GPa (where it is metastable) [386]. Pm¯3 LiPH6, a colored +variant of the A15 crystal structure adopted by intermetallic superconductors +Nb3Ge [258] and Nb3Sn [257], has an estimated Tc of 150-167 K at 200 GPa +(where it is metastable) [387]. In the S-P-H system, obviously tantalizing for +its connection to the H3S superconductor as well as to phosphine derivatives, +relatively low Tcs were predicted for phases on the high-pressure convex hulls, +but low-lying metastable structures based on phosphorus substitution into the +Im¯3m H3S lattice were promising, including Im¯3m S7PH24, which had an +estimated Tc of 183 K at 200 GPa [388]. +5 Conclusion +Although the entirety of the lived human experience resides within a vastly +narrow pressure range, the universe is not so simple. The chemistry we know +at 1 atmosphere is not the chemistry of Jupiter, Saturn, or even the center of +our own planet Earth. Starting from the periodic table itself, the ramifications +of pressure are rapidly found to alter elemental behavior and, consequently, +how the elements interact with one another to form new and bizarre phases. +Potassium, in its guise as a “transition metal”, enjoys all manner of new chem- +ical interactions – in compound formation and in the wildly complex electride +elemental structures it adopts. Cesium can become anionic, and helium takes +an active role in stabilizing a network of sodium and interstitial quasiatoms. +Strange geometrical and bonding motifs from clusters to networks abound. +Yet not only are the structures of phases – electronic and crystalline – +molded by high pressure, but high-pressure studies have revolutionized the +search for high-temperature superconductivity. Pressure induces superconduc- +tivity in a plethora of elements, and drives the formation of phases containing +structural motifs whose atomic vibrations can be strongly coupled to the +underlying electronic structure. From the clathrate-like LaH10 to the covalent +H3S – and the intensely-discussed CSH room-temperature superconductor – +the playing field of high-pressure materials is a promising one for the future. + +Springer Nature 2021 LATEX template +32 +Article Title +Acknowledgments +We acknowledge the NSF (DMR-1827815) for financial support. This mate- +rial is based upon work supported by the U.S. Department of Energy, Office +of Science, Fusion Energy Sciences under Award Number DE-SC0020340 to +E.Z. K.P.H. thanks the US Department of Energy, National Nuclear Security +Administration, through the Chicago-DOE Alliance Center under Coopera- +tive Agreement Grant No. DE-NA0003975 for financial support. We thank +Giacomo Scilla for his help in editing and preparing the manuscript. +References +[1] Guillot, T.: The Interiors of Giant Planets: Models and Outstanding +Questions. Annu. Rev. Earth Planet Sci. 33, 493–530 (2005) +[2] Zaghoo, M., Silvera, I.F.: Conductivity and dissociation in liquid metallic +hydrogen and implications for planetary interiors. Proc. Natl. Acad. Sci. +USA 114(45), 11873–11877 (2017) +[3] Matsuoka, T., Shimizu, K.: Direct observation of a pressure-induced +metal-to-semiconductor transition in lithium. Nature 458(7235), 186– +189 (2009) +[4] Ma, Y., Eremets, M., Oganov, A.R., Xie, Y., Trojan, I., Medvedev, S., +Lyakhov, A.O., Valle, M., Prakapenka, V.: Transparent dense sodium. +Nature 458(7235), 182–185 (2009) +[5] Polsin, D.N., Lazicki, A., Gong, X., Burns, S.J., Coppari, F., Hansen, +L.E., Henderson, B.J., Huff, M.F., McMahon, M.I., Millot, M., Paul, R., +Smith, R.F., Eggert, J.H., Collins, G.W., Rygg, J.R.: Structural com- +plexity in ramp-compressed sodium to 480 GPa. Nat. Commun. 13, 2354 +(2022) +[6] Bukowinski, M.S.T.: The Effect of Pressure on the Physics and Chem- +istry of Potassium. Geophys. Res. Lett. 3, 491–494 (1976) +[7] Atou, T., Hasegawa, M., Parker, L.J., Badding, J.V.: Unusual Chemical +Behavior for Potassium under Pressure: Potassium-Silver Compounds. +J. Am. Chem. Soc. 118, 12104–12108 (1996) +[8] Parker, L.J., Atou, T., Badding, J.V.: Transition Element-Like Chem- +istry for Potassium Under Pressure. Science 273, 95–97 (1996) +[9] Zhang, W., Oganov, A.R., Goncharov, A.F., Zhu, Q., Boulfelfel, +S.E., Lyakhov, A.O., Stavrou, E., Somayazulu, M., Prakapenka, V.B., +Konˆopkov´a, Z.: Unexpected stable stoichiometries of sodium chlorides. +Science 342(6165), 1502–1505 (2013) + +Springer Nature 2021 LATEX template +Article Title +33 +[10] Dong, X., Oganov, A.R., Goncharov, A.F., Stavrou, E., Lobanov, S., +Saleh, G., Qian, G.-R., Zhu, Q., Gatti, C., Deringer, V.L., Dron- +skowski, R., Zhou, X.-F., Prakapenka, V.B., Konˆopkov´a, Z., Popov, I.A., +Boldyrev, A.I., Wang, H.-T.: A stable compound of helium and sodium +at high pressure. Nature Chem. 9, 440–445 (2017) +[11] Geballe, Z.M., Liu, H., Mishra, A.K., Ahart, M., Somayazulu, M., Meng, +Y., Baldini, M., Hemley, R.J.: Synthesis and stability of lanthanum +superhydrides. Angew. Chem. Int. Ed. 57(3), 688 (2018) +[12] Prewitt, C.T., Downs, R.T.: High-pressure crystal chemistry. Rev. +Mineral 37, 283–318 (1998) +[13] Grochala, W., Hoffmann, R., Feng, J., Ashcroft, N.W.: The chemical +imagination at work in very tight places. Angew. Chem. Int. Ed. 46(20), +3620–3642 (2007) +[14] Zhang, L., Wang, Y., Lv, J., Ma, Y.: Materials discovery at high +pressures. Nature Rev. Mater. 2, 17005 (2017) +[15] Jayaraman, A.: Diamond anvil cell and high-pressure physical investiga- +tions. Rev. Mod. Phys. 55, 65–108 (1983) +[16] Bassett, W.A.: Diamond anvil cell, 50th birthday. High Press. Res. 29, +163–186 (2009) +[17] Li, B., Ji, C., Yang, W., Wang, J., Yang, K., Xu, R., Liu, W., Cai, Z., +Chen, J., Mao, H.-k.: Diamond anvil cell behavior up to 4 Mbar. Proc. +Natl. Acad. Sci. USA 115, 1713–1717 (2018) +[18] Dubrovinsky, L., Dubrovinskaia, N., Prakapenka, V.B., Abakumov, +A.M.: Implementation of micro-ball nanodiamond anvils for high- +pressure studies above 6 Mbar. Nature Comm. 3, 1163 (2012) +[19] Dubrovinskaia, N., Dubrovinsky, L., Solopova, N.A., Abakumov, A., +Turner, S., Hanfland, M., Bykova, E., Bykov, M., Prescher, C., +Prakapenka, V.B., Petitgirard, S., Chuvashova, I., Gasharova, B., +Mathis, Y.-L., Ershov, P., Snigireva, I., Snigirev, A.: Terapascal static +pressure generation with ultrahigh yield strength nanodiamond. Sci. +Adv. 2, 1600341 (2016) +[20] Dewaele, A., Loubeyre, P., Occelli, F., Marie, O., Mezouar, M.: Toroidal +diamond anvil cell for detailed measurements under extreme static +pressures. Nature Comm. 9, 2913 (2018) +[21] Jenei, Z., O’Bannon, E.F., Weir, S.T., Cynn, H., Lipp, M.J., Evans, W.J.: +Single crystal toroidal diamond anvils for high pressure experiments + +Springer Nature 2021 LATEX template +34 +Article Title +beyond 5 megabar. Nature Comm. 9, 3563 (2018) +[22] Duffy, T.S., Smith, R.F.: Ultra-High Pressure Dynamic Compression of +Geological Materials. Front. Earth. Sci. 7, 1–20 (2019) +[23] Hansen, L.E., Fratanduono, D.E., Zhang, S., Hicks, D.G., Suer, T., +Sprowal, Z.K., Huff, M.F., Gong, X., Henderson, B.J., Polsin, D.N., +Zaghoo, M., Hu, S.X., Collins, G.W., Rygg, J.R.: Melting of magnesium +oxide up to two terapascals using double-shock compression. Phys. Rev. +B 104, 014106 (2021) +[24] Fratanduono, D.E., Millot, M., Braun, D.G., Ali, S.J., Fernandez- +Pa˜nella, A., Seagle, C.T., Davis, J.-P., Brown, J.L., Akahama, Y., Kraus, +R.G., Marshall, M.C., Smith, R.F., O’Bannon, E.F., McNaney, J.M., +Eggert, J.H.: Establishing gold and platinum standards to 1 terapascal +using shockless compression. Science 372, 1063–1068 (2021) +[25] Smith, R.F., Eggert, J.H., Jeanloz, R., Duffy, T.S., Braun, D.G., Pat- +terson, J.R., Rudd, R.E., Biener, J., Lazicki, A.E., Hamza, A.V., Wang, +J., Braun, T., Benedict, L.X., Celliers, P.M., Collins, G.W.: Ramp +compression of diamond to five terapascals. Nature 511, 330–333 (2014) +[26] Kraus, R.G., Hemley, R.J., Ali, S.J., Belof, J.L., Benedict, L.X., Bernier, +J., Braun, D., Cohen, R.E., Collins, G.W., Coppari, F., Desjarlais, M.P., +Fratanduono, D., Hamel, S., Krygier, A., Lazicki, A., Mcnaney, J., Millot, +M., Myint, P.C., Newman, M.G., Rygg, J.R., Sterbentz, D.M., Stewart, +S.T., Stixrude, L., Swift, D.C., Wehrenberg, C., Eggert, J.H.: Measur- +ing the melting curve of iron at super-Earth core conditions. Science +375(6577), 202–205 (2022) +[27] Zurek, E., Grochala, W.: Predicting crystal structures and properties of +matter under extreme conditions via quantum mechanics: the pressure +is on. Phys. Chem. Chem. Phys. 17, 2917–2934 (2015) +[28] Pickard, C.J., Needs, R.J.: Structures at high pressure from random +searching. Phys. Status Solidi B 246(3), 536–540 (2009) +[29] Oganov, A.R., Lyakhov, A.O., Valle, M.: How Evolutionary Crystal +Structure Prediction Works – and Why. Acc. Chem. Res. 44(3), 227–237 +(2011) +[30] Wang, Y., Ma, Y.: Perspective: Crystal structure prediction at high +pressures. J. Chem. Phys. 140(4), 040901 (2014) +[31] Oganov, A.R., Chen, J., Gatti, C., Ma, Y., Ma, Y., Glass, C.W., Liu, Z., +Yu, T., Kurakevych, O.O.O., Solozhenko, V.L.: Ionic high-pressure form +of elemental boron. Nature 457, 863–867 (2009) + +Springer Nature 2021 LATEX template +Article Title +35 +[32] Zhang, M., Liu, H., Li, Q., Gao, B., Wang, Y., Li, H., Chen, C., Ma, +Y.: Superhard BC3 in Cubic Diamond Structure. Phys. Rev. Lett. 114, +015502 (2015) +[33] Avery, P., Wang, X., Oses, C., Gossett, E., Proserpio, D.M., Toher, C., +Curtarolo, S., Zurek, E.: Predicting superhard materials via a machine +learning informed evolutionary structure search. npj Comput. Mater. 5, +89 (2019) +[34] Liu, H., Naumov, I.I., Hoffmann, R., Ashcroft, N.W., Hemley, R.J.: +Potential high-Tc superconducting lanthanum and yttrium hydrides at +high pressure. Proc. Natl. Acad. Sci. U.S.A. 114, 6990–6995 (2017) +[35] Peng, F., Sun, Y., Pickard, C.J., Needs, R.J., Wu, Q., Ma, Y.: Hydro- +gen clathrate structures in rare earth hydrides at high pressures: possible +route to room-temperature superconductivity. Phys. Rev. Lett. 119, +107001 (2017) +[36] Kruglov, +I.A., +Kvashnin, +A.G., +Goncharov, +A.F., +Oganov, +A.R., +Lobanov, S.S., Holtgrewe, N., Jiang, S., Prakapenka, V.B., Greenberg, +E., Yanilkin, A.V.: Uranium polyhydrides at moderate pressures: Pre- +diction, synthesis, and expected superconductivity. Sci. Adv. 4, 9776 +(2009) +[37] Drozdov, A.P., Kong, P.P., Minkov, V.S., Besedin, S.P., Kuzovnikov, +M.A., Mozaffari, S., Balicas, L., Balakirev, F.F., Graf, D.E., Prakapenka, +V.B., Greenberg, E., Knyazev, D.A., Tkacz, M., Eremets, M.I.: Super- +conductivity at 250 K in lanthanum hydride under high pressures. Nature +569(7757), 528–531 (2019) +[38] Semenok, D.V., Troyan, I.A., Ivanova, A.G., Kvashnin, A.G., Kruglov, +I.A., Hanfland, M., Sadakov, A.V., Sobolevskiy, O.A., Pervakov, K.S., +Lyubutin, I.S., Glazyrin, K.V., Giordano, N., Karimov, D.N., Vasiliev, +A.L., Akashi, R., Pudalov, V.M., Oganov, A.R.: Superconductivity at +253 K in lanthanum-yttrium ternary hydrides. Mater. Today 48, 18–28 +(2021) +[39] Pickett, W., Eremets, M., Heil, C., Hemley, R., Pierleoni, C., Dias, R., +Amsler, M., Ma, Y., Kolmogorov, A., Oganov, A., Pickard, C.J., Bi, T., +Gross, H., Errea, I., Arita, R., Margine, R., Gonnelli, R., Valenti, R., Xie, +S., Boeri, L., Hirschfeld, P., Hennig, R., Profeta, G., Sanna, A., Zurek, +E.: The 2021 room-temperature superconductivity roadmap. J. Phys. +Condens. Mat. 34, 183002 (2022). https://doi.org/10.1088/1361-648x/ +ac2864 +[40] Dye, J.L.: Electrides: Early Examples of Quantum Confinement. Acc. +Chem. Res. 42, 1564–1572 (2009) + +Springer Nature 2021 LATEX template +36 +Article Title +[41] Zurek, E.: Alkali Metals in Ethylenediamine: A Computational Study of +the Optical Absorption Spectra and NMR Parameters of [M(en)+ +3 ·M −] +Ion-Pairs. J. Am. Chem. Soc. 133, 4829–4839 (2011) +[42] Riedel, R., Seel, A.G., Malko, D., Miller, D.P., Sperling, B.T., Choi, H., +Headen, T.F., Zurek, E., Porch, A., Kucernak, A., Pyper, N.C., Edwards, +P.P., Barrett, A.G.M.: Superalkali-Alkalide Interactions and Ion Pairing +in Low-Polarity Solvents. J. Am. Chem. Soc. 143, 3934–3943 (2021) +[43] Abella, L., Philips, A., Autschbach, J.: Ab initio molecular dynam- +ics study of sodium NMR chemical shifts in the methylamine solution +of [Na+[2.2.2]cryptand Na−]. Phys. Chem. Chem. Phys. 23, 339–346 +(2021) +[44] Botana, J., Miao, M.-S.: Pressure-stabilized lithium caesides with cae- +sium anions beyond the −1 state. Nat. Commun. 5(1), 1–8 (2014) +[45] Pauling, L.: The Nature of the Chemical Bond. IV. The Energy of Single +Bonds and the Relative Electronegativity of Atoms. J. Am. Chem. Soc. +54, 3570–3582 (1932) +[46] Pearson, R.G.: Absolute electronegativity and absolute hardness of Lewis +acids and bases. J. Am. Chem. Soc. 107, 6801–6806 (1985) +[47] Mulliken, R.S.: A New Electroaffinity Scale; Together with Data on +Valence States and on Valence Ionization Potentials and Electron +Affinities. J. Chem. Phys. 2, 782–793 (1934) +[48] Dong, X., Oganov, A.R., Cui, H., Zhou, X.-F., Wang, H.-T.: Electroneg- +ativity and chemical hardness of elements under pressure. Proc. Natl. +Acad. Sci. 119, 2117416119 (2022) +[49] Allen, L.C.: Electronegativity Is the Average One-Electron Energy of +the Valence-Shell Electrons in Ground-State Free Atoms. J. Am. Chem. +Soc. 111, 9003–9014 (1989) +[50] Rahm, M., Zeng, T., Hoffmann, R.: Electronegativity Seen as the +Ground-State Average Valence Electron Binding Energy. J. Am. Chem. +Soc. 141, 342–351 (2019). https://doi.org/10.1021/jacs.8b10246 +[51] Rahm, M., Erhart, P., Cammi, R.: Relating atomic energy, radius, and +electronegativity through compression. Chem. Sci. 12, 2397–2403 (2021) +[52] Racioppi, S., Rahm, M.: In-Situ Electronegativity and the Bridging of +Chemical Bonding Concepts. Chem. Eur. J. 27, 18156–18167 (2021) +[53] Sternheimer, R.: On the Compressibility of Metallic Cesium. Phys. Rev. + +Springer Nature 2021 LATEX template +Article Title +37 +78, 235–243 (1950) +[54] Louie, S.G., Cohen, M.L.: Electronic structure of cesium under pressure. +Phys. Rev. B 10, 3237–3245 (1974) +[55] Takemura, K., Minomura, S., Shimomura, O.: X-Ray Diffraction Study +of Electronic Transitions in Cesium under High Pressure. Phys. Rev. +Lett. 49, 1772–1775 (1982) +[56] Connerade, J.P., Semaoune, R.: Atomic compressibility and reversible +insertion of atoms into solids. J. Phys. B: At. Mol. Opt. Phys. 33, 3467– +3484 (2000) +[57] Rahm, M., Cammi, R., Ashcroft, N.W., Hoffmann, R.: Squeezing All +Elements in the Periodic Table: Electron Configuration and Electroneg- +ativity of the Atoms under Compression. J. Am. Chem. Soc. 141, +10253–10271 (2019) +[58] Zurek, E., Jepsen, O., Andersen, O.K.: Muffin-Tin Orbital Wannier- +Like Functions for Insulators and Metals. ChemPhysChem 6, 1934–1942 +(2005) +[59] Neaton, J.B., Ashcroft, N.W.: Pairing in dense lithium. Nature 400, +141–144 (1999) +[60] Hanfland, M., Syassen, J., Christensen, N.E., Novikov, D.L.: New high- +pressure phases of lithium. Nature 408, 174–178 (2000) +[61] Naumov, I.I., Hemley, R.J., Hoffmann, R., Ashcroft, N.W.: Chemical +bonding in hydrogen and lithium under pressure. J. Chem. Phys. 143, +064702 (2015) +[62] Degtyareva, V.F.: Potassium under pressure: Electronic origin of com- +plex structures. Solid State Sci. 36, 62–72 (2014) +[63] Takemura, K., Christensen, N.E., Novikov, D.L., Syassen, K., Schwarz, +U., Hanfland, M.: Phase stability of highly compressed cesium. Phys. +Rev. B 61, 14399–14404 (2000) +[64] McMahan, A.K.: Alkali-metal structures above the s-d transition. Phys. +Rev. B 29, 5982–5985 (1984) +[65] McMahon, M.I., Nelmes, R.J., Schwarz, U., Syassen, K.: Composite +incommensurate K-III and a commensurate form: Study of a high- +pressure phase of potassium. Phys. Rev. B 74, 140102 (2006) +[66] Hooper, J., Zurek, E.: High pressure potassium polyhydrides: a chemical +perspective. J. Phys. Chem. C 116(24), 13322–13328 (2012) + +Springer Nature 2021 LATEX template +38 +Article Title +[67] Dewaele, A., Pepin, C.M., Geneste, G., Garbarino, G.: Reaction between +nickel or iron and xenon under high pressure. High Pressure Research +37, 137–146 (2017) +[68] Stavrou, E., Yao, Y., Goncharov, A.F., Lobanov, S.S., Zaug, J.M., Liu, +H., Greenberg, E., Prakapenka, V.B.: Synthesis of Xenon and Iron-Nickel +Intermetallic Compounds at Earth’s Core Thermodynamic Conditions. +Phys. Rev. Lett. 120, 096001 (2018) +[69] Snider, E., Dasenbrock-Gammon, N., McBride, R., Wang, X., Mey- +ers, N., Lawler, K.V., Zurek, E., Salamat, A., Dias, R.P.: Synthesis of +Yttrium Superhydride Superconductor with a Transition Temperature +up to 262 K by Catalytic Hydrogenation at High Pressures. Phys. Rev. +Lett. 126, 117003 (2021) +[70] Somayazulu, M., Ahart, M., Mishra, A.K., Geballe, Z.M., Baldini, M., +Meng, Y., Struzhkin, V.V., Hemley, R.J.: Evidence for Superconductivity +above 260 K in Lanthanum Superhydride at Megabar Pressures. Phys. +Rev. Lett. 122, 027001 (2019) +[71] Ma, L., Wang, K., Xie, Y., Yang, X., Wang, Y., Zhou, M., Liu, H., Yu, X., +Zhao, Y., Wang, H., Liu, G., Ma, Y.: High-temperature superconducting +phase in clathrate calcium hydride CaH6 up to 215 K at a pressure of +172 GPa. Phys. Rev. Lett. 128, 167001 (2022) +[72] Li, Z., Zhang, C., Wang, X., Zhang, S., Jia, Y., Feng, S., Lu, K., Zhao, +J., Zhang, J., Min, B., Long, Y., Yu, R., Wang, L., Ye, M., Zhang, Z., +Prakapenka, V., Chariton, S., Ginsberg, P.A., Bass, J., Yuan, S., Liu, +H., Jin, C.: Superconductivity above 200 K discovered in superhydrides +of calcium. Nature Commun. 13, 2863 (2022) +[73] Wang, H., John, S.T., Tanaka, K., Iitaka, T., Ma, Y.: Superconduc- +tive sodalite-like clathrate calcium hydride at high pressures. Proc. Natl. +Acad. Sci. USA 109(17), 6463–6466 (2012) +[74] Dye, J.L.: Compounds of Alkali Metal Anions. Angew. Chem. Int. Ed. +18, 587–598 (1979) +[75] Pickard, C.J., Needs, R.J.: Predicted Pressure-Induced s-Band Ferro- +magnetism in Alkali Metals. Phys. Rev. Lett. 107, 087201 (2011) +[76] Hasegawa, M., Atou, T., Badding, J.V.: High-Pressure Synthesis of an +Alkali Metal-Transition metal Laves Phase: KAg2. J. Solid State Chem. +130, 311–315 (1997) +[77] Lee, K.K.M., Jeanloz, R.: High-pressure alloying of potassium and iron: +Radioactivity in the Earth’s core? Geophys. Res. Lett. 30 (2003) + +Springer Nature 2021 LATEX template +Article Title +39 +[78] Adeleke, A.A., Stavrou, E., Adeniyi, A.O., Wan, B., Gou, H., Yao, Y.: +Two good metals make a semiconductor: A potassium-nickel compound +under pressure. Phys. Rev. B 102, 134120 (2020) +[79] Brgoch, J., Hermus, M.: Pressure-Stabilized Ir3− in a Superconducting +Potasium Iridide. J. Phys. Chem. C 120, 20033–20039 (2016) +[80] Lotfi, S., Brgoch, J.: Predicting pressure-stabilized alkali metal iridides: +A-Ir (A=Rb, Cs. Comput. Mater. Sci. 158, 124–129 (2019) +[81] Yang, G., Yang, Y., Peng, F., Bergara, A., Ma, Y.: Gold as a 6p-Element +in Dense Lithium Aurides. J. Am. Chem. Soc. 138, 4046–4052 (2016) +[82] Botana, J., Wang, X., Hou, C., Yan, D., Lin, H., Ma, Y., Miao, M.: +Mercury under Pressure acts as a Transition Metal: Calculated from +First Principles. Angew. Chem. Int. Ed. 54, 9280–9283 (2015) +[83] Savin, A.: The Electron Localization Function (ELF) and Its Relatives: +Interpretations and Difficulties. J. Mol. Struc.-Theochem. 727(1-3), 127– +131 (2005) +[84] Miao, M.-s.: Caesium in high oxidation states and as a p-block element. +Nat. Chem. 5(10), 846–852 (2013) +[85] Liu, C., Nikolaev, S.A., Wen, R., Burton, L.A.: Electrides: a review. J. +Mater. Chem. C 8, 10551–10567 (2020) +[86] Zurek, E., Edwards, P.P., Hoffmann, R.: A Molecular Perspective on +Lithium-Ammonia Solutions. Angew. Chem. Int. Ed. 48, 8198–8232 +(2009) +[87] Miao, M., Hoffmann, R.: High-Pressure Electrides: The Chemical Nature +of Interstitial Quasiatoms. J. Am. Chem. Soc. 137, 3631–3637 (2015) +[88] Neaton, J.B., Ashcroft, N.W.: On the constitution of sodium at higher +densities. Phys. Rev. Lett. 86, 2830–2833 (2001) +[89] Li, P., Gao, G., Wang, Y., Ma, Y.: Crystal Structures and Exotic Behav- +ior of Magnesium under Pressure. J. Phys. Chem. C 114, 21745–21749 +(2010) +[90] Dong, X., Oganov, A.R.: Electrides and Their High-Pressure Chemistry. +In: Correlations in Condensed Matter Under Extreme Conditions, pp. +69–84. Springer, Cham (2017). Chap. 6 +[91] Modak, P., Verma, A.K.: Pressure induced multi-centre bonding and +metal-insulator transition in PtAl2. Phys. Chem. Chem. Phys. 21, +13337–13346 (2019) + +Springer Nature 2021 LATEX template +40 +Article Title +[92] Miao, M.-S., Hoffmann, R.: High pressure electrides: A predictive +chemical and physical theory. Acc. Chem. Res. 47(4), 1311–1317 (2014) +[93] Wang, B., Hilleke, K.P., Wang, X., Polsin, D.N., Zurek, E.: Topolog- +ical Electride Phase of Sodium at High Pressures and Temperatures. +arXiv (2022). https://doi.org/10.48550/ARXIV.2205.06251. https:// +arxiv.org/abs/2205.06251 +[94] Li, Y., Wang, Y., Pickard, C.J., Needs, R.J., Wang, Y., Ma, Y.: Metallic +Icosahedron Phase of Sodium at Terapascal Pressures. Phys. Rev. Lett. +114, 125501 (2015) +[95] Vegas, A., Grzechnik, A., Hanfland, M., M¨uhle, C., Jansen, M.: Antiflu- +orite to Ni2In-type phase transition in K2S at high pressures. Solid State +Sci. 4, 1077–1081 (2002) +[96] Santamaria-Perez, D., Vegas, A., Muehle, C., Jansen, M.: High-pressure +experimental study on Rb2S: antifluorite to Ni2In-type phase transitions. +Acta Crystallogr. B 67, 109–115 (2011) +[97] Marques, M., Ackland, G.J., Lundegaard, L.F., Stinton, G., Nelmes, +R.J., McMahon, M.I., Contreras-Garcia, J.: Potassium under pressure: +A Pseudobinary Ionic Compound. Phys. Rev. Lett. 103, 115501 (2009) +[98] McMahon, M.I., Gregoryanz, E., Lundegaard, L.F., Loa, I., Guillaume, +C., Nelmes, R.J., Kleppe, A.K., Amboage, M., Wilhelm, H., Jeph- +coat, A.P.: Structure of sodium above 100 GPa by single-crystal x-ray +diffraction. Proc. Natl. Acad. Sci. 104, 17297–17299 (2007) +[99] Guillaume, C.L., Gregoryanz, E., Degtyareva, O., McMahon, M.I., Han- +fland, M., Evans, S., Guthrie, M., Sinogeikin, S.V., Mao, H.-k.: Cold +melting and solid structures of dense lithium. Nature Phys. 7, 211–214 +(2011) +[100] Miao, M., Hoffmann, R., Botana, J., Naumov, I.I., Hemley, R.J.: Quasi- +molecules in Compressed Lithium. Angew. Chem. Int. Ed. 56, 972–975 +(2017) +[101] Dronskowski, R., Bl¨ochl, P.E.: Crystal Orbital Hamilton Populations +(COHP). Energy-Resolved Visualization of Chemical Bonding in Solids +Based on Density-Functional Calculations. J. Phys. Chem. 97, 8617– +8624 (1993) +[102] Miao, M.-s., Wang, X., Brgoch, J., Spera, F., Jackson, M.G., Kresse, G., +Lin, H.: Anionic chemistry of noble gases: formation of Mg–NG (NG= +Xe, Kr, Ar) compounds under pressure. J. Am. Chem. Soc. 137(44), +14122–14128 (2015) + +Springer Nature 2021 LATEX template +Article Title +41 +[103] Botana, J., Brgoch, J., Hou, C., Miao, M.: Iodine Anions beyond -1: +Formation of LinI (n=2-5) and Its Interaction with Quasiatoms. Inorg. +Chem. 55, 9377–9382 (2016) +[104] Miao, M., Sun, Y., Zurek, E., Lin, H.: Chemistry under high pressure. +Nat. Rev. Chem. 4, 508–527 (2020) +[105] Pauling, L.: Principles determining the structure of high-pressure metals: +The structures of cesium(IV) and cesium(V). Proc. Natl. Acad. Sci. USA +86, 1431–1433 (1989) +[106] von Schnering, H.G., Nesper, R.: How Nature Adapts Chemical Struc- +tures to Curved Surfaces. Angew. Chem. Int. Ed. 26, 1059–1080 +(1987) +[107] Olijnyk, H., Holzapfel, W.B.: Phase transitions in K and Rb under +pressure. Phys. Lett. A 99, 381–383 (1983) +[108] Ma, Y., Oganov, A.R., Xie, Y.: High-pressure structures of lithium, +potassium, and rubidium predicted by an ab initio evolutionary algo- +rithm. Phys. Rev. B 78, 014102 (2008) +[109] Lundegaard, L.F., Marques, M., Stinton, G., Ackland, G.J., Nelmes, +R.J., McMahon, M.I.: Observation of the oP8 crystal structure in +potassium at high pressure. Phys. Rev. B 80, 020101 (2009) +[110] Schwarz, U., Grzechnik, A., Syassen, K., Loa, I., Hanfland, M.: +Rubidium-IV: A High Pressure Phase with Complex Crystal Structure. +Phys. Rev. Lett. 83, 4085–4088 (1999) +[111] McMahon, M.I., Rekhi, S., Nelmes, R.J.: Pressure Dependent Incom- +mensuration in Rb-IV. Phys. Rev. Lett. 87, 055501 (2001) +[112] McMahon, M.I., Nelmes, R.J.: High-pressure structures and phase +transformations in elemental metals. Chem. Soc. Rev. 35, 943–963 +(2006) +[113] Gregoryanz, E., Lundegaard, L.F., McMahon, M.I., Guillaume, C., +Nelmes, R.J., Mezouar, M.: Structural Diversity of Sodium. Science 320, +1054–1057 (2008) +[114] Woolman, G., Robinson, V.N., Marques, M., Loa, I., Ackland, G.J., +Hermann, A.: Structural and electronic properties of the alkali metal +incommensurate phases. Phys. Rev. Mater. 2, 053604 (2018) +[115] Zhou, Y., Xu, Q., Zhu, C., Li, Q., Liu, H., Wang, H., Tse, J.S.: Predicted +lithium-iron compounds under high pressure. RSC Adv. 6, 66721–66728 + +Springer Nature 2021 LATEX template +42 +Article Title +(2016) +[116] Zhou, Y., Wang, H., Zhu, C., Liu, H., Tse, J.S., Ma, Y.: Prediction +of Host-Guest Na-Fe Intermetallics at High Pressure. Inorg. Chem. 55, +7026–7032 (2016) +[117] Zhang, J., Chen, G., Liu, H.: Stable Structures and Superconductivity +in a Y-Si System under High Pressure. J. Phys. Chem. Lett. 12, 10388– +10393 (2021) +[118] Pepin, R.O., Porcelli, D.: Origin of Noble Gases in the Terrestrial +Planets. Rev. Mineral. Geochem. 47, 191–246 (2002) +[119] Shcheka, S.S., Keppler, H.: The origin of the terrestrial noble-gas +signature. Nature 490, 531–534 (2012) +[120] Marty, B.: The origins and concentrations of water, carbon, nitrogen, +and noble gases on Earth. Earth & Planet. Sci. Lett. 313-314, 56–66 +(2012) +[121] Caldwell, W.A., Nguyen, J.H., Pfrommer, B.G., Mauri, F., Louie, S.G., +Jeanloz, R.: Structure, Bonding, and Geochemistry of Xenon at High +Pressures. Science 277, 930–933 (2012) +[122] Lee, K.K.M., Steinle-Neumann, G.: High-pressure alloying of iron and +xenon: “Missing” Xe in the Earth’s core? J. Geophys. Res. 111, 02202 +(2006) +[123] Zhu, L., Liu, H., Pickard, C.J., Zou, G., Ma, Y.: Reactions of xenon with +iron and nickel are predicted in the Earth’s inner core. Nat. Chem. 6(7), +644–648 (2014) +[124] Peng, F., Song, X., Liu, C., Li, Q., Miao, M., Chen, C., Ma, Y.: Xenon +iron oxides predicted as potential Xe hosts in Earth’s lower mantle. Nat. +Commun. 11(1), 1–7 (2020) +[125] Adeleke, A.A., Kunz, M., Greenberg, E., Prakapenka, V.B., Yao, Y., +Stavrou, E.: A High-Pressure Compound of Argon and Nickel: Noble +Gas in the Earth’s Core? ACS Earth Space Chem. 3, 2517–2524 (2019) +[126] Zhu, Q., Jung, D.Y., Oganov, A.R., Glass, C.W., Gatti, C., Lyakhov, +A.O.: Stability of xenon oxides at high pressures. Nat. Chem. 5(1), 61–65 +(2013) +[127] Hermann, A., Schwerdtfeger, P.: Xenon suboxides stable under pressure. +J. Phys. Chem. Lett. 5(24), 4336–4342 (2014) +[128] Dewaele, A., Worth, N., Pickard, C.J., Needs, R.J., Pascarelli, S., + +Springer Nature 2021 LATEX template +Article Title +43 +Mathon, O., Mezouar, M., Irifune, T.: Synthesis and stability of xenon +oxides Xe2O5 and Xe3O2 under pressure. Nature Chem. 8, 784–790 +(2016) +[129] Zaleski-Ejgierd, P., Lata, P.M.: Krypton oxides under pressure. Sci. Rep. +6, 18938 (2016) +[130] Peng, F., Wang, Y., Wang, H., Zhang, Y., Ma, Y.: Stable xenon nitride +at high pressures. Phys. Rev. B 92(9), 094104 (2015) +[131] Howie, R.T., Turnbull, R., Binns, J., Frost, M., Dalladay-Simpson, +P., Gregoryanz, E.: Formation of xenon-nitrogen compounds at high +pressure. Sci. Rep. 6, 34896 (2016) +[132] Bovornratanaraks, T., Tsuppayakorn-aek, P., Luo, W., Ahuja, R.: +Ground-state structure of semiconducting and superconducting phases +in xenon carbides at high pressure. Sci. Rep. 9, 2459 (2019) +[133] Kurzyd�lowski, D., Zaleski-Ejgierd, P.: Ground-state structure of semi- +conducting and superconducting phases in xenon carbides at high +pressure. Phys. Chem. Chem. Phys. 18, 2309–2313 (2016) +[134] Kurzyd�lowski, D., So�ltysiak, M., D˙zoleva, A., Zaleski-Ejgierd, P.: High- +Pressure Reactivity of Kr and F2-Stabilization of Krypton in the +4 +Oxidation State. Crystals 7, 329 (2017) +[135] Peng, F., Botana, J., Wang, Y., Ma, Y., Miao, M.: Unexpected Trend in +Stability of Xe-F Compounds under Pressure Driven by Xe-Xe Covalent +Bonds. J. Phys. Chem. Lett. 7, 4562–4567 (2016) +[136] Zarifi, N., Liu, H., Tse, J.S., Zurek, E.: Crystal Structures and Electronic +Properties of Xe-Cl Compounds at High Pressure. J. Phys. Chem. C +122, 2941–2950 (2018) +[137] Li, X., Hermann, A., Peng, F., Lv, J., Wang, Y., Wang, H., Ma, Y.: +Stable Lithium Argon compounds under high pressure. Sci. Rep. 5, 16675 +(2015) +[138] Liu, Z., Botana, J., Miao, M., Yan, D.: Unexpected Xe anions in XeLin +intermetallic compounds. EPL 117, 26002 (2017) +[139] Zhang, S., Bi, H., Wei, S., Wang, J., Li, Q., Ma, Y.: Crystal Structures +and Electronic Properties of Cesium Xenides at High Pressures. J. Phys. +Chem. C 119, 24996–25002 (2015) +[140] Sanloup, C., Bonev, S.A., Hochlaf, M., Maynard-Casely, H.E.: Reactivity +of xenon with ice at planetary conditions. Phys. Rev. Lett. 110(26), + +Springer Nature 2021 LATEX template +44 +Article Title +265501 (2013) +[141] Monserrat, R., Martinez-Canales, M., Needs, R.J., Pickard, C.J.: Helium- +Iron Compounds at Terapascal Pressures. Phys. Rev. Lett. 121, 015301 +(2018) +[142] Liu, C., Gao, H., Hermann, A., Wang, Y., Miao, M., Pickard, C.J., Needs, +R.J., Wang, H.-T., Xing, D., Sun, J.: Plastic and Superionic Helium +Ammonia Compounds under High Pressure and High Temperature. +Phys. Rev. X 10, 021007 (2020) +[143] Liu, C., Yao, Y., Klug, D.D.: Stable structures of He and H2O at high +pressure. Phys. Rev. B 91, 014102 (2015) +[144] Liu, C., Gao, H., Wang, Y., Needs, R.J., Pickard, C.J., Sun, J., Wang, +H.-T., Xing, D.: Multiple superionic states in helium-water compounds. +Nat. Phys. 15, 1065–1070 (2019) +[145] Vos, W.L., Finger, L.W., Hemley, R.J., Hu, J.Z., Mao, H.K., Schouten, +J.A.: A high-pressure van der Waals compound in solid nitrogen-helium +mixtures. Nature 358, 46–48 (1992) +[146] Li, Y., Feng, X., Liu, H., Hao, J., Redfern, S.A.T., Lei, W., Liu, D., +Ma, Y.: Route to high-energy density polymeric nitrogen t-N via He-N +compounds. Nat. Commun. 9, 722 (2018) +[147] Wang, Y., Zhang, J., Liu, H., Yang, G.: Prediction of the Xe-He binary +phase diagram at high pressures. Chem. Phys. Lett. 640, 115–118 (2015) +[148] Cazorla, C., Errandonea, D., Sola, E.: High-pressure phases, vibrational +properties, and electronic structure of Ne(He)2 and Ar(He)2: A first- +principles study. Phys. Rev. B 80, 064105 (2009) +[149] Loubeyre, P., Jean-Louis, M., LeToullec, R., Charon-G´erard, L.: High +pressure measurements of the He-Ne binary phase diagram at 296 K: +Evidence for the stability of a stoichiometric Ne(He)2 solid. Phys. Rev. +Lett. 70, 178–181 (1993) +[150] Liu, Z., Botana, J., Hermann, A., Valdez, S., Zurek, E., Yan, D., Lin, H., +Miao, M.: Reactivity of He with ionic compounds under high pressure. +Nature Comm. 9, 951 (2018) +[151] Gao, H., Sun, J., Pickard, C.J., Needs, R.J.: Prediction of pressure- +induced stabilization of noble-gas-atom compounds with alkali oxides +and alkali sulfides. Phys. Rev. Mater. 3, 015002 (2019) +[152] Zhang, J., Lv, J., Li, H., Feng, X., Lu, C., Redfern, S.A.T., Liu, H., + +Springer Nature 2021 LATEX template +Article Title +45 +Chen, C., Ma, Y.: Rare Helium-Bearing Compound FeO2He Stabilized +at Deep-Earth Conditions. Phys. Rev. Lett. 121, 255703 (2018) +[153] Hume-Rothery, W., Raynor, G.V.: The structure of metals and alloys, +4th edn. Institute of Metals, London (1962) +[154] Miedema, A.R.: A simple model for alloys, I. Rules for the alloying +behavious of transition metals. Philips Tech. Rev. 33, 149–160 (1973) +[155] Miedema, A.R.: A simple model for alloys, II. The influence of ionicity on +the stability and other physical properties of alloys. Philips Tech. Rev. +33, 196–202 (1973) +[156] Darken, L.S., Gurry, R.W.: Physical chemistry of metals. McGraw-Hill, +New York (1953) +[157] Chelikowsky, J.R.: Solid solubilities in divalent alloys. Phys. Rev. B 19, +686–701 (1979) +[158] Alonso, J.A., Simozar, S.: Prediction of solid solubility in alloys. Phys. +Rev. B 22, 5583–5589 (1980) +[159] Wang, Z., Huang, Y., Yang, Y., Wang, J., Liu, C.T.: Atomic-size effect +and solid solubility of multicomponent alloys. Scr. Mater. 94, 28–31 +(2015) +[160] Dubrovinskaia, N., Dubrovinsky, L., McCammon, C.: Iron-magnesium +alloying at high pressures and temperatures. J. Phys. Condens. Matter +16, 1143 (2004) +[161] Gao, P., Su, C., Shao, S., Wang, S., Liu, P., Liu, S., Lv, J.: Iron- +magnesium compounds under high pressure. New J. Chem. 43, 17403– +17407 (2019) +[162] Fang, Y., Sun, Y., Wang, R., Zheng, F., Wu, S., Wang, C.-Z., Wentz- +covitch, R.M., Ho, K.-M.: Unconventional iron-magnesium compounds +at terapascal pressures. Phys. Rev. B 104, 144109 (2021) +[163] Dubrovinskaia, +N., +Dubrovinsky, +L., +Kantor, +I., +Crichton, +W.A., +Dmitriev, V., Prakapenka, V., Shen, G., Vitos, L., Ahuja, R., Johansson, +B., Abrikosov, I.A.: Beating the Miscibility Barrier between Iron Group +Elements and Magnesium by High-Pressure Alloying. Phys. Rev. Lett. +95, 245502 (2005) +[164] Feng, J., Hennig, R.G., Ashcroft, N.W., Hoffmann, R.: Emergent reduc- +tion of electronic state dimension in dense ordered Li-Be alloys. Nature +451, 445–448 (2008) + +Springer Nature 2021 LATEX template +46 +Article Title +[165] Degtyareva, V.F.: Simple metals at high pressures: the Fermi sphere- +Brillouin zone interaction model. Phys.-Usp. 49, 369–388 (2006) +[166] Berger, R.F., Walters, P.L., Lee, S., Hoffmann, R.: Connecting the Chem- +ical and Physical Viewpoints of What Determines Structure: From 1-D +Chains to γ-Brasses. Chem. Rev. 111, 4522–4545 (2011) +[167] Feng, J., Hoffmann, R., Ashcroft, N.W.: Double-diamond NaAl via pres- +sure: Understanding structure through Jones zone activation. J. Chem. +Phys. 132, 114106 (2010) +[168] Guo, K., Akselrud, L., Bobnar, M., Burkhardt, U., Schmidt, M., Zhao, +J.-T., Schwarz, U., Grin, Y.: Weak Interactions under Pressure: hp-CuBi +and Its Analogues. Angew. Chem. Int. Ed. 56, 5620–5624 (2017) +[169] Clarke, S., Amsler, M., Walsh, J.P.S., Yu, T., Wang, Y., Meng, Y., +Jacobsen, S., Wolverton, C., Freedman, D.E.: Creating binary Cu-Bi +Compounds via High-Pressure Synthesis: A Combined Experimental and +Theoretical Study. Chem. Mater. 29, 5276–5285 (2017) +[170] Chen, Y.L., Analytis, J.G., Chu, J.-H., Liu, Z.K., Mo, S.-K., Qi, X.L., +Zhang, H.J., Lu, D.H., Dai, X., Fang, Z., Zhang, S.C., Fisher, I.R., Hus- +sain, Z., Shen, Z.-X.: Experimental Realization of a Three-Dimensional +Topological Insulator, Bi2Te. Science 325, 178–181 (2009) +[171] Liu, Z.K., Zhou, B., Zhang, Y., Wang, Z.J., Weng, H.M., Prabhakaran, +D., Mo, S.-K., Shen, Z.X., Fang, Z., Dai, X., Hussain, Z., Chen, Y.L.: +Discovery of a Three-Dimensional Topological Dirac Semimetal, Na3Bi. +Science 343, 864–867 (2014) +[172] Reynolds, J.M., Lane, C.T.: Superconducting Bismuth Alloys. Phys. +Rev. 79, 405 (1950) +[173] Matthias, B.T., A., J., Geballe, T.H., Andres, K., Corenzwit, E.: Many +More Superconducting Bismuth Phases. Phys. Rev. Lett. 17, 640–643 +(1966) +[174] Hor, Y.S., Williams, A.J., Checkelsky, J.G., Roushan, P., Seo, J., Xu, +Q., Zandbergen, H.W., Yazdani, A., Ong, N.P., Cava, R.J.: Supercon- +ductivity in CuxBi2Se3 and its Implications for Pairing in the Undoped +Topological Insulator. Phys. Rev. Lett. 104, 057001 (2010) +[175] Walsh, J.P.S., Clarke, S.M., Meng, Y., Jacobsen, S.D., Freedman, D.E.: +Discovery of FeBi2. ACS Cent. Sci. 6, 867–871 (2016) +[176] Amsler, M., Naghavi, S.S., Wolverton, C.: Prediction of superconducting +iron-bismuth intermetallic compounds at high pressure. Chem. Sci. 8, + +Springer Nature 2021 LATEX template +Article Title +47 +2226 (2017) +[177] Amsler, M., Hegde, V.I., Jacobsen, S.D., Wolverton, C.: Exploring the +High-Pressure Materials Genome. Phys. Rev. X 8, 041021 (2018) +[178] Walsh, J.P.S., Clarke, S.M., Puggioni, D., Tamerius, A.D., Meng, Y., +Rondinelli, J.M., Jacobsen, S.D., Freedman, D.E.: MnBi2: A Metastable +High-Pressure Phase in the Mn-Bi System. Chem. Mater. 31, 3083–3088 +(2019) +[179] Takizawa, H., Uheda, K., Endo, T.: A new ferromagnetic polymorph of +CrSb2 synthesized under high pressure. J. Alloys Compd. 287, 145–149 +(1999) +[180] Wu, X., Steinle-Neumann, G., Qin, S., Kanzaki, M., Dubrovinsky, L.: +Pressure-induced phase transitions of AX2-type iron pnictides: an ab +initio study. J. Phys.: Condens. Matter 21, 185403 (2009) +[181] Poffo, C.M., Souza, S.M., Trichˆes, D.M., de Lima, J.C., Grandi, T.A., +Polian, A., Gauthier, M.: Structural and optical studies of FeSb2 under +high pressure. Physica B 407, 4686–4694 (2012) +[182] Schwarz, U., Tence, S., Janson, O., Koz, C., Krellner, C., Burkhardt, +U., Rosner, H., Steglich, F., Grin, Y.: CoBi3: A Binary Cobalt-Bismuth +Compound and Superconductor. Angew. Chem. Int. Ed. 52, 9853–9857 +(2013) +[183] Tenc´e, S., Janson, O., Krellner, C., Rosner, H., Schwarz, U., Grin, +Y., Steglich, F.: CoBi3- the first binary compound of cobalt with bis- +muth: high-pressure synthesis and superconductivity. J. Phys.: Condens. +Matter 26, 395701 (2014) +[184] Clarke, S.M., Walsh, J.P.S., Amsler, M., Malliakas, C.D., Yu, T., +Goedecker, S., Wang, Y., Wolverton, C., Freedman, D.E.: Discovery +of a Superconducting Cu-Bi Intermetallic Compound by High-Pressure +Synthesis. Angew. Chem. Int. Ed. 128, 13644–13647 (2016) +[185] Amsler, M., Wolverton, C.: Dense superconducting phases of copper- +bismuth at high pressure. Phys. Rev. M 1, 031801 (2017) +[186] Altman, A.B., Tamerius, A.D., Koocher, N.Z., Meng, Y., Pickard, C.J., +Walsh, J.P.S., Rondinelli, J.M., Jacobsen, S.D., Freedman, D.E.: Compu- +tationally Directed Discovery of MoBi2. J. Am. Chem. Soc. 143, 214–222 +(2021) +[187] Pauling, L.: The Nature of the Chemical Bond vol. 260. Cornell +University Press, Ithaca, NY (1960) + +Springer Nature 2021 LATEX template +48 +Article Title +[188] Goncharov, A.F., Struzhkin, V.V., Somayazulu, M.S., Hemley, R.J., +Mao, H.K.: Compression of ice to 210 gigapascals: Infrared evidence for a +symmetric hydrogen-bonded phase. Science 273(5272), 218–220 (1996) +[189] Bernasconi, M., Silvestrelli, P., Parrinello, M.: Ab initio infrared absorp- +tion study of the hydrogen-bond symmetrization in ice. Phys. Rev. Lett. +81(6), 1235 (1998) +[190] Zhang, L., Wang, Y., Zhang, X., Ma, Y.: High-pressure phase transitions +of solid HF, HCl, and HBr: An ab initio evolutionary study. Phys. Rev. +B 82(1), 014108 (2010) +[191] Errea, I., Calandra, M., Pickard, C.J., Nelson, J.R., Needs, R.J., Li, +Y., Liu, H., Zhang, Y., Ma, Y., Mauri, F.: Quantum hydrogen-bond +symmetrization in the superconducting hydrogen sulfide system. Nature +532(7597), 81–84 (2016) +[192] Vij, A., Pavlovich, J.G., Wilson, W.W., Vij, V., Christe, K.O.: Experi- +mental Detection of the Pentaazacyclopentadienide (Pentazolate) Anion, +cyclo-N− +5 . Angew. Chem. 114, 3177–3180 (2002) +[193] Wang, P., Xu, Y., Lin, Q., Lu, M.: Recent advances in the synthe- +ses and properties of polynitrogen pentazolate anion cyclo-N− +5 and its +derivatives. Chem. Soc. Rev. 47, 7522–7538 (2018) +[194] Shen, Y., Oganov, A.R., Qian, G., Zhang, J., Dong, H., Zhu, Q., Zhou, +Z.: Novel lithium-nitrogen compounds at ambient and high pressures. +Sci. Rep. 5, 14204 (2015) +[195] Peng, F., Yao, Y., Liu, H., Ma, Y.: Crystalline LiN5 Predicted from First- +Principles as a Possible High-Energy Material. J. Phys. Chem. Lett. 6, +2363–2366 (2015) +[196] Laniel, D., Weck, G., Loubeyre, P.: Direct Reaction of Nitrogen and +Lithium up to 75 GPA: Synthesis of the Li3N, LiN, LiN2, and LiN5 +Compounds. Inorg. Chem. 57, 10685–10693 (2018) +[197] Zhou, M., Sui, M., Shi, X., Zhao, Z., Guo, L., Liu, B., Liu, R., Wang, P., +Liu, B.: Lithium Pentazolate Synthesized by Laser Heating-Compressed +Lithium Azide and Nitrogen. J. Phys. Chem. C 124, 11825–11830 (2020) +[198] Laniel, D., Weck, G., Gaiffe, G., Garbarino, F., Loubeyre, P.: High- +Pressure Synthesized Lithium Pentazolate Compound Metastable under +Ambient Conditions. J. Phys. Chem. Lett. 9, 1600–1604 (2018) +[199] Steele, B.A., Oleynik, I.I.: Sodium pentazolate: A nitrogen rich high +energy density material. Chem. Phys. Lett. 643, 21–26 (2016) + +Springer Nature 2021 LATEX template +Article Title +49 +[200] Peng, F., Han, Y., Liu, H., Yao, Y.: Exotic stable cesium polynitrides at +high pressure. Sci. Rep. 5, 16902 (2015) +[201] Li, J., Sun, L., Wang, X., Zhu, H., Miao, M.: Simple Route to Metal +cyclo-N− +5 Salt: High-Pressure Synthesis of CuN5. J. Phys. Chem. C 122, +22339–22344 (2018) +[202] Xia, K., Zheng, X., Yuan, J., Liu, C., Gao, H., Wu, Q., Sun, J.: Pressure- +Stabilized High-Energy-Density Alkaline-Earth-Metal Pentazolate Salts. +J. Phys. Chem. C 123, 10205–10211 (2019) +[203] Liu, Z., Li, D., Tian, F., Duan, D., Li, H., Cui, T.: Moderate Pressure +Stabilized Pentazolate Cyclo-N− +5 Anion in Zn(N5)2 Salt. Inorg. Chem. +59, 8002–8012 (2020) +[204] Huang, B., Frapper, G.: Barium-Nitrogen Phases Under Pressure: Emer- +gence of Structural Diversity and Nitrogen-Rich Compounds. Chem. +Mater. 30, 7623–7636 (2018) +[205] Wang, B., Larklimi, R., Valencia, H., Gu´egan, F., Frapper, G.: Prediction +of Novel Tin Nitride SnxNy Phases Under Pressure. J. Phys. Chem. C +124, 8080–8093 (2020) +[206] Du, X., Yao, Y., J., W., Yang, Q., Wang, G.: IrN4 and IrN7 as potential +high-energy-density materials. J. Chem. Phys. 154, 054706 (2021) +[207] Wu, L., Tian, R., Wan, B., Liu, H., Gong, N., Chen, P., Shen, T., Yao, +Y., Gou, H., Gao, F.: Prediction of Stable Iron Nitrides at Ambient and +High Pressure with Progressive Formation of New Polynitrogen Species. +Chem. Mater. 30, 8476–8485 (2018) +[208] Niu, S., Li, Z., Li, H., Shi, X., Yao, Z., Liu, B.: New Cadmium-Nitrogen +Compounds at High Pressures. Inorg. Chem. 60, 6772–6781 (2021) +[209] Liu, L., Wang, D., Zhang, S., Zhang, H.: Pressure-stabilized GdN6 with +an armchair-antiarmchair structure as a high energy density material. J. +Mater. Chem. A 9, 16751–16758 (2021) +[210] Bykov, M., Bykova, E., Ponomareva, I.A. A. V.and Abrikosov, Chari- +ton, S., Prakapenka, V.B., Mahmood, M.F., Dubrovinsky, L., Goncharov, +A.F.: Stabilization of Polynitrogen Anions in Tantalum-Nitrogen Com- +pounds Under Pressure. Angew. Chem. 133, 9085–9090 (2021) +[211] Gao, G., Ashcroft, N.W., Miao, M., Hoffmann, R.: Novel Si Networks +in the Ca/Si Phase Diagram under Pressure. J. Phys. Chem. C 118, +25167–25175 (2014) + +Springer Nature 2021 LATEX template +50 +Article Title +[212] Li, W., Lu, M., Chen, L., Xin, X., Gao, L., Li, J., Zhou, X., Liu, W., +Zhang, X., Zhang, M.: Crystal structures of CsSi6 at high pressures. +Comput. Mater. Sci. 150, 144–148 (2018) +[213] H¨ubner, J.-M., Carrillo-Cabrera, W., Prots, Y., Bobnar, M., Schwarz, +U., Grin, Y.: Unconventional Metal-Framework Interaction in MgSi5. +Angew. Chem. Int. Ed. 58, 12914–12918 (2019) +[214] Zhang, S., Wang, Y., Yang, G., Ma, Y.: Silicon Framework-Based +Lithium Silicides at High Pressures. ACS Appl. Mater. Interfaces 8, +16761–167167 (2016) +[215] Zurek, E., Yao, Y.: Theoretical predictions of novel superconducting +phases of BaGe3 stable at atmospheric and high pressures. Inorg. Chem. +54(6), 2875–2884 (2015) +[216] Castillo, R., Baranov, A.I., Burkhardt, U., Cardoso-Gil, R., Schnelle, +W., Bobnar, M., Schwarz, U.: Germanium Dumbbells in a New Super- +conducting Modification of BaGe3. Inorg. Chem. 55, 4498–4503 (2016) +[217] Hooper, J., Zurek, E.: Lithium subhydrides under pressure and their +superatom-like building blocks. ChemPlusChem 77, 969–972 (2012) +[218] Schwarz, U., Wosylus, A., Rosner, H., Schnelle, W., Ormeci, A., Meier, +K., Baranov, A., Nicklas, M., Leipe, S., M¨uller, C.J., Grin, Y.: Dumb- +bells of five-connected silicon atoms and superconductivity in the binary +silicides MSi3 (M= Ca, Y, Lu). J. Am. Chem. Soc. 134(33), 13558–13561 +(2012) +[219] H¨ubner, J.-M., Akselrud, L., Schnelle, W., Burkhardt, U., Bobnar, M., +Prots, Y., Grin, Y., Schwarz, U.: High-Pressure Synthesis and Chemical +Bonding of Barium Trisilicide BaSi3. Materials 12, 145 (2019) +[220] Schnelle, W., Ormeci, A., Wosylus, A., Meier, K., Grin, Y., Schwarz, U.: +Dumbbells of five-connected Ge atoms and superconductivity in CaGe3. +Inorg. Chem. 51(10), 5509–5511 (2012) +[221] Nishikawa, T., Fukuoka, H., Inumaru, K.: High-Pressure Synthesis and +Electronic Structure of a New Superconducting Strontium Germanide +(SrGe3) Containing Ge2 Dumbbells. Inorg. Chem. 54, 7433–7437 (2015) +[222] Castillo, R., Schnelle, W., Baranov, A.I., Burkhardt, U., Bobnar, M., +Cardoso-Gil, R., Schwarz, U., Grin, Y.: Trigermanides AEGe3 (AE=Ca, +Sr, Ba): chemical bonding and superconductivity. Z. Naturforsch. 71, +585–592 (2016) + +Springer Nature 2021 LATEX template +Article Title +51 +[223] Lazicki, A., McGonegle, D., Rygg, J.R., Braun, D.G., Swift, D.C., Gor- +man, M.G., Smith, R.F., Heighway, P.G., Higginbotham, A., Suggit, +M.J., Fratanduono, D.E., Coppari, F., Wehrenberg, C.E., Kraus, R.G., +Erskine, D., Bernier, J.V., McNaney, J.M., Rudd, R.E., Collins, G.W., +Eggert, J.H., Wark, J.S.: Metastability of diamond ramp-compressed to +2 terapascals. Nature 589, 532–535 (2021) +[224] Simon, A.: Metal clusters inside out. Philos. Trans. R. Soc. A 368, 1285– +1299 (2010) +[225] Oka, T.: Interstellar H+ +3 . Chem. Rev. 113, 8738–8761 (2013) +[226] Wang, Z., Wang, H., Tse, J.S., Iitaka, T., Ma, Y.: Stabilization of H3+ +in the high pressure crystalline structure of HnCl (n = 2–7). Chem. Sci. +6(1), 522–526 (2015) +[227] Duan, D., Huang, X., Tian, F., Liu, Y., Li, D., Yu, H., Liu, B., Tian, +W., Cui, T.: Predicted Formation of H3+ in Solid Halogen Polyhydrides +at High Pressures. J. Phys. Chem. A 119, 11059–11065 (2015) +[228] Zeng, Q., Yu, S., Li, D., Oganov, A.R., Frapper, G.: Emergence of novel +hydrogen chlorides under high pressure. Phys. Chem. Chem. Phys. 19, +8236–8242 (2017) +[229] Bi, T., Shamp, A., Terpstra, T., Hemley, R.J., Zurek, E.: The Li–F–H +ternary system at high pressures. J. Chem. Phys. 154(12), 124709 (2021) +[230] Ayouz, M., Dulieu, O., Gu´erout, R., Robert, J., Kokoouline, V.: Potential +energy and dipole moment surfaces of H− +3 molecule. J. Chem. Phys. +132(19), 194309 (2010) +[231] Hooper, J., Zurek, E.: Rubidium Polyhydrides Under Pressure: Emer- +gence of the Linear H3− Anion. Chem. Eur. J. 18, 5013–5021 (2012) +[232] Shamp, A., Hooper, J., Zurek, E.: Compressed cesium polyhydrides: Cs+ +sublattices and H3−-three-connected nets. Inorg. Chem. 51(17), 9333– +9342 (2012) +[233] Hooper, J., Altintas, B., Shamp, A., Zurek, E.: Polyhydrides of the alka- +line earth metals: a look at the extremes under pressure. J. Phys. Chem. +C 117(6), 2982–2992 (2013) +[234] Struzhkin, V.V., Kim, D.Y., Stavrou, E., Muramatsu, T., Mao, H.-k., +Pickard, C.J., Needs, R.J., Prakapenka, V.B., Goncharov, A.F.: Synthe- +sis of sodium polyhydrides at high pressures. Nat. Commun. 7(1), 12267 +(2016) + +Springer Nature 2021 LATEX template +52 +Article Title +[235] Liu, Y., Duan, D., Tian, F., Liu, H., Wang, C., Huang, X., Li, D., Ma, Y., +Liu, B., Cui, T.: Pressure-Induced Structures and Properties in Indium +Hydrides. Inorg. Chem. 54, 9924–9928 (2015) +[236] Chen, Y., Geng, H.Y., Yan, X., Sun, Y., Wu, Q., Chen, X.: Prediction of +Stable Ground-State Lithium POlyhydrides under High Pressures. Inorg. +Chem. 56, 3867–3874 (2017) +[237] Ye, X., Zarifi, N., Zurek, E., Hoffmann, R., Ashcroft, N.W.: High +Hydrides of Scandium under Pressure: Potential Superconductors. J. +Phys. Chem. C 122, 6298–6309 (2018) +[238] Xie, H., Yao, Y., Duan, D., Song, H., Zhang, Z., Jiang, S., Redfern, +S.A.T., Kresin, V.Z., Pickard, C.J., Cui, T.: Hydrogen Pentagraphene- +like Structure Stabilized by Hafnium: A High-Temperature Conventional +Superconductor. Phys. Rev. Lett. 125, 217001 (2020) +[239] Feng, X., Zhang, J., Gao, G., Liu, H., Wang, H.: Compressed sodalite- +like MgH6 as a potential high-temperature superconductor. RSC Adv. +5, 59292–59296 (2015) +[240] Qian, S., Sheng, X., Yan, X., Chen, Y., Song, B.: Theoretical study of +stability and superconductivity of ScHn (n=4-8) at high pressure. Phys. +Rev. B 96, 094513 (2017) +[241] Abe, K.: Hydrogen-rich scandium compounds at high pressures. Phys. +Rev. B 96, 144108 (2017) +[242] Li, Y., Hao, J., Liu, H., Tse, J.S., Wang, Y., Ma, Y.: Pressure-stabilized +Superconductive Yttrium Hydrides. Sci. Rep. 5, 9948 (2015) +[243] Zhao, J., Ao, B., Li, S., Gao, T., Ye, X.: Phase Diagram and Bonding +States of Pu-H Binary Compounds at High Pressures. J. Phys. Chem. C +124, 7361–7369 (2020) +[244] Hai, Y.L., Lu, N., Tian, H.L., Jiang, M.J., Yang, W., Li, W.J., Yan, +X.W., Zhang, C., Chen, X.J., Zhong, G.H.: Cage Structure and Near +Room-Temperature Superconductivity in TbHn (n = 1-12). J. Phys. +Chem. C 125, 3640–3649 (2021) +[245] Semenok, D.V., Zhou, D., Kvashnin, A.G., Huang, X., Galasso, M., +Kruglov, I.A., Ivanova, A.G., Gavriliuk, A.G., Chen, W., Tkachenko, +N.V., Boldyrev, A.I., Troyan, I., Oganov, A.R., Cui, T.: Novel Strongly +Correlated Europium Superhydrides. J. Phys. Chem. Lett. 12(1), 32–40 +(2021) +[246] Sun, W., Kuang, X., Keen, H.D.J., Lu, C., Hermann, A.: Second group + +Springer Nature 2021 LATEX template +Article Title +53 +of high-pressure high-temperature lanthanide polyhydride superconduc- +tors. Phys. Rev. B 102, 144524 (2020) +[247] Abe, K.: High-pressure properties of dense metallic zirconium hydrides +studied by calculations. Phys. Rev. B 98, 134103 (2018) +[248] Hooper, J., Terpstra, T., Shamp, A., Zurek, E.: The Composition and +Constitution of Compressed Strontium Polyhydrides. J. Phys. Chem. C +118, 6433–6447 (2014) +[249] Wang, Y., Wang, H., Tse, J.S., Iitaka, T., Ma, Y.: Structural Morpholo- +gies of High-Pressure Polymorphs of Strontium Hydrides. Phys. Chem. +Chem. Phys. 17, 19379–19385 (2015) +[250] Bi, T., Zarifi, N., Terpstra, T., Zurek, E.: The search for supercon- +ductivity in high pressure hydrides. In: Reedijk, J. (ed.) Reference +Module in Chemistry, Molecular Sciences and Chemical Engineer- +ing, pp. 1–36. Elsevier, Waltham, MA (2019). https://doi.org/10.1016/ +b978-0-12-409547-2.11435-0 +[251] Hilleke, K.P., Zurek, E.: Rational Design of Superconducting Metal +Hydrides via Chemical Pressure Tuning. Angew. Chem. Int. Ed. 134, +202207589 (2022) +[252] Shao, Z., Duan, D., Ma, Y., Yu, H., Song, H., Xie, H., Li, D., Tian, +F., Liu, B., Cui, T.: Unique Phase Diagram and Superconductivity of +Calcium Hydrides at High Pressures. Inorg. Chem. 58, 2558–2564 (2019) +[253] Semenok, D.V., Kvashnin, A.G., Kruglov, I.A., Oganov, A.R.: Actinium +Hydrides AcH10, AcH12, and AcH16 as High-Temperature Conventional +Superconductors. J. Phys. Chem. Lett. 9(8), 1920–1926 (2018) +[254] Errea, I., Belli, F., Monacelli, L., Sanna, A., Koretsune, T., Tadano, +T., Bianco, R., Calandra, M., Arita, R., Mauri, F., Flores-Livas, J.A.: +Quantum crystal structure in the 250-kelvin superconducting lanthanum +hydride. Nature 578, 66–69 (2020) +[255] Sun, Y., Lv, J., Xie, Y., Liu, H., Ma, Y.: Route to a superconducting +phase above room temperature in electron-doped hydride compounds +under high pressure. Phys. Rev. Letters 123(9), 097001 (2019) +[256] Onnes, H.K.: Further Experiments with Liquid Helium. D. On the +Change of Electrical Resistance of Pure Metals at Very low Tempera- +tures, etc. V. The Disappearance of the Resistance of Mercury,. Proc. K. +Ned. Akad. Wet. B 14, 113–115 (1911) + +Springer Nature 2021 LATEX template +54 +Article Title +[257] Matthias, B.T., Geballe, T.H., Geller, S., Corenzwit, E.: Superconduc- +tivity of Nb3Sn. Phys. Rev. 95, 1435 (1954) +[258] Gavaler, J.R.: Superconductivity in Nb-Ge films above 22 K. Appl. Phys. +Lett. 23, 480 (1973) +[259] Anderson, P.W., Matthias, B.T.: Superconductivity. Science 144, 373– +381 (1964) +[260] Wu, M.K., Ashburn, J.R., Torng, C.J., Hor, P.H., Meng, R.L., Gao, L., +Huang, Z.J., Wang, Y.Q., Chu, C.W.: Superconductivity at 93 K in a +new mixed-phase Y-Ba-Cu-O compound system at ambient pressure. +Phys. Rev. Lett. 58, 908 (1987) +[261] Gao, L., Xue, Y.Y., Chen, F., Xiong, Q., Meng, R.L., Ramirez, D., +Chu, C.W., Eggert, J.R., Mao, H.K.: Superconductivity up to 164 K +in HgBa2Cam−1CumO2m+2+δ (m=1,2, and 3) under quasihydrostatic +pressures. Phys. Rev. B 50, 4260–4263 (1994) +[262] Finnemore, D.K., Stromberg, T.F., Swenson, C.A.: Superconducting +Properties of High-Purity Niobium. Phys. Rev. 149, 231–243 (1966) +[263] Nagamatsu, J., Nakagawa, N., Muranaka, T., Zenitani, Y., Akimitsu, J.: +Superconductivity at 39 K in magnesium diboride. Nature 410, 63–64 +(2001) +[264] Sakata, M., Nakamoto, Y., Shimizu, K.: Superconducting state of Ca- +VII below a critical temperature of 29 K at a pressure of 216 GPa. Phys. +Rev. B 83, 220512 (2011) +[265] Drozdov, A.P., Eremets, M.I., Troyan, I.A.: Superconductivity above +100 K in PH3 at high pressures (2015) arXiv:1508.06224 [cond-mat.supr- +con] +[266] Drozdov, A.P., Eremets, M.I., Troyan, I.A., Ksenofontov, V., Shylin, S.I.: +Conventional superconductivity at 203 Kelvin at high pressures in the +sulfur hydride system. Nature 525, 73–76 (2015) +[267] Salke, N.P., Esfahani, M.M.D., Zhang, Y., Kruglov, I.A., Zhou, J., Wang, +Y., Greenberg, E., Prakapenka, V.B., Liu, J., Oganov, A.R., Lin, J.-F.: +Synthesis of clathrate cerium superhydride CeH9 at 80-100 GPa with +atomic hydrogen sublattice. Nature Commun. 10, 4453 (2019) +[268] Snider, E., Dasenbrock-Gammon, N., McBride, R., Debessai, M., Vin- +dana, H., Vencatasamy, K., Lawler, K.V., Salamat, A., Dias, R.P.: +Room-temperature superconductivity in a carbonaceous sulfur hydride. +Nature 586(7829), 373–377 (2020) + +Springer Nature 2021 LATEX template +Article Title +55 +[269] Di Cataldo, S., Heil, C., von der Linden, W., Boeri, L.: LaBH8: Towards +high-Tc low-pressure superconductivity in ternary superhydrides. Phys. +Rev. B 104, 020511 (2021) +[270] Gao, M., Yan, X.-Y., Lu, Z.-Y., Xiang, T.: Phonon-mediated high- +temperature superconductivity in the ternary borohydride KB2H8 under +pressure near 12 GPa. Phys. Rev. B 104, 100504 (2021) +[271] Zhang, Z., Cui, T., Hutcheon, M.J., Shipley, A.M., Song, H., Du, +M., Kresin, V.Z., Duan, D., Pickard, C.J., Yao, Y.: Design Princi- +ples for High-Temperature Superconductors with Hydrogen-Based Alloy +Backbone at Moderate Pressure. Phys. Rev. Lett. 128, 047001 (2022) +[272] Matthias, B.T.: Progress in Low Temperature Physics. In: Supercon- +ductivity in the Periodic System, pp. 138–150. Elsevier, Amsterdam +(1957) +[273] Bardeen, J., Cooper, L.N., Schrieffer, J.R.: Microscopic Theory of +Superconductivity. Phys. Rev. 106, 162–164 (1957) +[274] Bardeen, J., Cooper, L.N., Schrieffer, J.R.: Theory of Superconductivity. +Phys. Rev. 108, 1175–1204 (1957) +[275] Pickett, W.E.: The next breakthrough in phonon-mediated superconduc- +tivity. Physica C 468, 126–135 (2008) +[276] Swift, R.M., White, D.: Low Temperature Heat Capacities of Magnesium +Diboride (MgB2) and Magnesium Tetraboride (MgB4). J. Am. Chem. +Soc. 79, 3641–3644 (1957) +[277] McMillan, W.L.: Transition Temperature of Strong-Coupled Supercon- +ductors. Phys. Rev. 167, 331–344 (1968) +[278] Dynes, R.C.: McMillan’s Equation and the Tc of Superconductors. Solid +State Commun. 10, 615–618 (1972) +[279] Allen, P.B., Dynes, R.C.: Transition Temperature of Strong-Coupled +Superconductors Reanalyzed. Phys. Rev. B 12, 905–922 (1975) +[280] Hopfield, J.J.: Angular Momentum and Transition-Metal Superconduc- +tivity. Phys. Rev. 186, 443–451 (1969) +[281] Quan, Y., Ghosh, S.S., Pickett, W.E.: Compressed hydrides as metallic +hydrogen superconductors. Phys. Rev. B 100, 184505 (2019) +[282] Buzea, C., Robbie, K.: Assembling the puzzle of superconducting +elements: a review. Supercond. Sci. Technol. 18(1), 1 (2004) + +Springer Nature 2021 LATEX template +56 +Article Title +[283] Schilling, J.S.: Superconductivity in the alkali metals. High Press. Res. +26, 145–163 (2006) +[284] Hamlin, J.J.: Superconductivity in the metallic elements at high pres- +sures. Physica C 514, 59–76 (2015) +[285] Shimizu, K.: Superconductivity from insulating elements under high +pressure. Physica C 514, 46–49 (2015) +[286] Tuoriniemi, J., Juntunen-Nurmilaukas, K., Uusvuori, J., Pentti, E., +Salmela, A., Sebedash, A.: Superconductivity in lithium below 0.4 +millikelvin at ambient pressure. Nature 447, 187–189 (2007) +[287] Falge Jr., R.L.: Superconductivity of hexagonal beryllium. Phys. Lett. +A 24, 579–580 (1967) +[288] Sizoo, G.J., Onnes, H.K.: Further experiments with liquid helium. Influ- +ence of elastic deformation on the supraconductivity of tin and indium. +Commun. Phys. Lab. Univ. Leiden. 180b, 13–26 (1925) +[289] Jennings, L.D., Swenson, C.A.: Effects of Pressure on the Superconduct- +ing Transition Temperatures of Sn, In, Ta, Tl, and Hg. Phys. Rev. 112, +31–43 (1958) +[290] Smith, T.F., Chu, C.W.: Will Pressure Destroy Superconductivity? +Phys. Rev. 159, 353–359 (1967) +[291] Struzhkin, V.V., Eremets, M.I., Gan, W., Mao, H.-k., Hemley, R.J.: +Superconductivity in Dense Lithium. Science 298, 1213–1215 (2002) +[292] Shimizu, K., Ishikawa, H., Takao, D., Yagi, T., Amaya, K.: Super- +conductivity in compressed lithium at 20 K. Nature 419, 597–599 +(2002) +[293] Deemyad, S., Schilling, J.S.: Superconducting Phase Diagram of Li Metal +in Nearly Hydrostatic Pressures up to 67 GPa. Phys. Rev. Lett. 91, +167001 (2003) +[294] Chen, X.-J.: Exploring high-temperature superconductivity in hard mat- +ter close to structural instability. Matter Radiat. Extremes 5, 068102 +(2020) +[295] Kasinathan, D., Kune˘s, J., Lazicki, A., Rosner, H., Yoo, C.S., Scalet- +tar, R.T., Pickett, W.E.: Superconductivity and Lattice Instability in +Compressed Lithium from Fermi Surface Hot Spots. Phys. Rev. Lett 96, +047004 (2006) + +Springer Nature 2021 LATEX template +Article Title +57 +[296] Profeta, G., Franchini, C., Lathiotakis, N.N., Floris, A., Sanna, A., Mar- +ques, M.A.L., L¨uders, M., Massidda, S., Gross, E.K.U., Continenza, +A.: Superconductivity in Lithium, Potassium, and Aluminum under +Extreme Pressure: A First-Principles Study. Phys. Rev. Lett. 96, 047003 +(2006) +[297] Rodriguez-Prieto, A., Bergara, A., Silkin, V.M., Echenique, P.M.: Com- +plexity and Fermi surface deformation in compressed lithium. Phys. Rev. +B 74, 172104 (2006) +[298] Xie, Y., Tse, J.S., Cui, T., Oganov, A.R., He, Z., Ma, Y., Zou, G.: Elec- +tronic and phonon instabilities in face-centered-cubic alkali metals under +pressure using ab initio calculations. Phys. Rev. B 75, 064102 (2007) +[299] Yao, Y., Tse, J.S., Tanaka, K., Marsiglio, F., Ma, Y.: Superconductivity +in lithium under high pressure investigated with density functional and +Eliashberg theory. Phys. Rev. B 79, 054524 (2009) +[300] Bazhirov, T., Noffsinger, J., Cohen, M.L.: Superconductivity and +electron-phonon coupling in lithium at high pressures. Phys. Rev. B 82, +184509 (2010) +[301] Rousseau, B., Xie, Y., Ma, Y., Bergara, A.: Exotic high pressure behavior +of light alkali metals, lithium and sodium. Eur. Phys. J. B 81, 1 (2011) +[302] Yue, S.-Y., Cheng, L., Liao, B., Hu, M.: Electron-phonon interaction and +superconductivity in the high-pressure cI 16 phase of lithium from first +principles. Phys. Chem. Chem. Phys. 20, 27125–27130 (2018) +[303] Matsuoka, T., Sakata, M., Nakamoto, Y., Takahama, K., Ichimaru, K., +Mukai, K., Ohta, K., Hirao, N., Ohishi, Y., Shimizu, K.: Pressure- +induced reentrant metallic phase in lithium. Phys. Rev. B 89, 144103 +(2014) +[304] Yan, Y., Zhang, Y., Wang, Y., Yang, G.: Pressure-induced reappearance +of superconductivity in the oC24 phase of lithium. Solid State Commun. +225, 7–11 (2016) +[305] Wittig, J.: Pressure-Induced Superconductivity in Cesium and Yttrium. +Phys. Rev. Lett. 24, 812–815 (1970) +[306] Schwarz, U., Takemura, K., Hanfland, M., Syassen, K.: Crystal Structure +of Cesium-V. Phys. Rev. Lett. 81, 2711–2714 (1998) +[307] Deng, Y., Schilling, J.S.: Evidence for superconductivity in Rb metal +above 55 GPa pressure. Phys. Rev. B 100, 041109 (2019) + +Springer Nature 2021 LATEX template +58 +Article Title +[308] Debessai, M., Hamlin, J.J., Schilling, J.S.: Comparison of the pressure +dependences of Tc in the trivalent d-electron superconductors Sc, Y, La, +and Lu up to megabar pressures. Phys. Rev. B 78, 064519 (2008) +[309] Slocombe, D.R., Kuznetsov, V.L., Grochala, W., Williams, R.J.P., +Edwards, P.P.: Superconductivity in transition metals. Phil. Trans. R. +Soc. A 373, 20140476 (2015) +[310] Ginzburg, V.L.: What problems of physics and astrophysics seem now +to be especially important and interesting (thirty years later, already on +the verge of XXI century)? Phys.-Usp. 42, 353–373 (1999) +[311] Ginzburg, V.L.: Nobel Lecture: On Superconductivity and Superfluid- +ity. https://www.nobelprize.org/uploads/2018/06/ginzburg-lecture.pdf. +Accessed: 2022-03-23 (2003) +[312] Wigner, E., Huntington, H.B.: On the Possibility of a Metallic Modifi- +cation of Hydrogen. J. Chem. Phys. 3, 764–770 (1935) +[313] Ashcroft, N.W.: Metallic Hydrogen: A High-Temperature Superconduc- +tor? Phys. Rev. Lett. 21, 1748–1749 (1968) +[314] Dias, R.P., Silvera, I.F.: Observation of the Wigner-Huntington transi- +tion to metallic hydrogen. Science 355, 715–718 (2017) +[315] Eremets, M.I., Drozdov, A.P., Kong, P.P., Wang, H.: Semimetallic molec- +ular hydrogen at pressure above 350 GPa. Nat. Phys. 15, 1246–1249 +(2019) +[316] Loubeyre, P., Occelli, F., Dumas, P.: Synchrotron infrared spectroscopic +evidence of the probable transition to metal hydrogen. Nature 577, 631– +635 (2020) +[317] Liu, X.-D., Dalladay-Simpson, P., Howie, R.T., Li, B., Gregoryanz, +E.: Comment on ”Observation of the Wigner-Huntington transition to +metallic hydrogen”. Science 357, 2286 (2017) +[318] Goncharov, A.F., Struzhkin, V.V.: Comment on ”Observation of the +Wigner-Huntington transition to metallic hydrogen”. Science 357, 9736 +(2017) +[319] Gregoryanz, E., Cheng, J., Dalladay-Simpson, P., Li, B., Howie, R.T., +Mao, H.-K.: Everything you always wanted to know about metallic +hydrogen but were afraid to ask. Matter Radiat. Extremes 5, 038101 +(2020) + +Springer Nature 2021 LATEX template +Article Title +59 +[320] Monacelli, L., Errea, I., Calendra, M., Mauri, F.: Black metal hydro- +gen above 360 GPa driven by proton fluctuations. Nature Phys. 17, 63 +(2021) +[321] McMahon, J.M., Ceperley, D.M.: High-temperature superconductivity +in atomic metallic hydrogen. Phys. Rev. B 84, 144515 (2011) +[322] Ashcroft, N.W.: Hydrogen Dominant Metallic Alloys: High Temperature +Superconductors? Phys. Rev. Lett. 92, 187002 (2004) +[323] Smithson, H., Marianetti, C.A., Morgan, D., Van der Ven, A., Pred- +ith, A., Ceder, G.: First-principles study of the stability and electronic +structure of metal hydrides. Phys. Rev. B 66, 144107 (2002) +[324] Bourgeois, N., Crivello, J.-C., Cenedese, P., Joubert, J.-M.: Systematic +First-Principles Study of Binary Metal Hydrides. ACS Comb. Sci. 19, +512–523 (2017) +[325] Troyan, +I.A., +Semenok, +D.V., +Kvashnin, +A.G., +Sadakov, +A.V., +Sobolevskiy, O.A., Pudalov, V.M., Ivanova, A.G., Prakapenka, V.B., +Greenberg, E., Gavriliuk, A.G., Lyubutin, I.S., Struzhkin, V.V., Bergara, +A., Errea, I., Bianco, R., Calandra, M., Mauri, F., Monacelli, L., Akashi, +R., Oganov, A.R.: Anomalous high-temperature superconductivity in +YH6. Adv. Mater. 33, 2006832–110 (2021) +[326] Kong, P., Minkov, V.S., Kuzovnikov, M.A., Drozdov, A.P., Besedin, +S.P., Mozaffari, S., Balicas, L., Balakirev, F.F., Prakapenka, V.B., Chari- +ton, S., Knyazev, D.A., Greenberg, E., Eremets, M.I.: Superconductivity +up to 243 K in the Yttrium-Hydrogen System under pressure. Nature +Commun. 12, 5075 (2021) +[327] Jeon, H., Wang, C., Yi, S., Cho, J.-H.: Origin of enhanced chemical +precompression in cerium hydride CeH9. Sci. Rep. 10, 16878 (2020) +[328] Liang, X., Bergara, A., Wei, X., Wang, L., Sun, R., Liu, H., Hemley, R.J., +Wang, L., Gao, G., Tian, Y.: Prediction of high-Tc superconductivity in +ternary lanthanum borohydrides. Phys. Rev. B 104, 134501 (2021) +[329] Durajski, A.P., Szcz¸e´sniak, R.: New superconducting superhydride +LaC2H8 at relatively low stabilization pressure. Phys. Chem. Chem. +Phys., 25070–25074 (2021) +[330] Shatruk, M.: ThCr2Si2 structure type: The “perovskite” of inter- +metallics. J. Solid State Chem. 272, 198–209 (2019) +[331] Bi, T., Zurek, E.: Electronic Structure and Superconductivity of Com- +pressed Metal Tetrahydrides. Chem. Eur. J. 27(60), 14858–14870 (2021) + +Springer Nature 2021 LATEX template +60 +Article Title +[332] Mishra, A.K., Muramatsu, T., Liu, H., Geballe, Z.M., Somayazulu, M., +Ahart, M., Baldini, M., Meng, Y., Zurek, E., Hemley, R.J.: New calcium +hydrides with mixed atomic and molecular hydrogen. J. Phys. Chem. C +122, 19370–19378 (2018) +[333] Li, B., Miao, Z., Ti, L., Liu, S., Chen, J., Shi, J., Gregoryanz, E.: Pre- +dicted high-temperature superconductivity in cerium hydrides at high +pressures. J. Appl. Phys. 126, 235901 (2019) +[334] Zhou, D., Semenok, D.V., Duan, D., Xie, H., Chen, W., Huang, X., +Li, X., Liu, B., Oganov, A.R., Cui, T.: Superconducting praseodymium +superhydrides. Sci. Adv. 6, 6849 (2020) +[335] Pe˜na-Alvarez, M., Binns, J., Hermann, A., Kelsall, L.C., Dalladay- +Simpson, P., Gregoryanz, E., Howie, R.T.: Praseodymium polyhydrides +synthesized at high temperatures and pressures. Phys. Rev. B 100, +184109 (2019) +[336] Zhou, D., Semenok, D.V., Xie, H., Huang, X., Duan, D., Aperis, A., +Oppeneer, P.M., Galasso, M., Kartsev, A.I., Kvashnin, A.G., Oganov, +A.R., Cui, T.: High-Pressure Synthesis of Magnetic Neodymium Poly- +hydrides. J. Am. Chem. Soc. 142(6), 2803–2811 (2020) +[337] Kvashnin, A.G., Semenok, D.V., Kruglov, I.A., Wrona, I.A., Oganov, +A.R.: High-Temperature Superconductivity in a Th-H System under +Pressure Conditions. ACS Appl. Mater. Interfaces 10, 43809–43816 +(2018) +[338] Semenok, D.V., Kvashnin, A.G., Ivanova, A.G., Svitlyk, V., Fominski, +V.Y., Sadakov, A.V., Sobolevskiy, O.A., Pudalov, V.M., Troyan, I.A., +Oganov, A.R.: Superconductivity at 161 K in thorium hydride ThH10: +Synthesis and properties. Mater. Today 33, 36–44 (2020) +[339] Hoffmann, R., Zheng, C.: Making and Breaking Bonds in the Solid State: +The ThCr2Si2 Structure. J. Phys. Chem. 89, 4175–4181 (1985) +[340] Li, Y., Hao, J., Liu, H., Li, Y., Ma, Y.: The metallization and super- +conductivity of dense hydrogen sulfide. J. Chem. Phys. 140(17), 174712 +(2014) +[341] Strobel, T.A., Ganesh, P., Somayazulu, M., Kent, P.R.C., Hemley, R.J.: +Novel Cooperative Interactions and Structural Ordering in H2S-H2. +Phys. Rev. Lett. 107, 255503 (2011) +[342] Duan, D., Liu, Y., Tian, F., Li, D., Huang, X., Zhao, Z., Yu, H., Liu, +B., Tian, W., Cui, T.: Pressure-induced metallization of dense (H2S)2H2 +with high-Tc superconductivity. Sci. Rep. 4, 6968 (2014) + +Springer Nature 2021 LATEX template +Article Title +61 +[343] Einaga, M., Sakata, M., Ishikawa, T., Shimizu, K., Eremets, M.I., Droz- +dov, A.P., Troyan, I.A., Hirao, N., Ohishi, Y.: Crystal structure of the +superconducting phase of sulfur hydride. Nat. Phys. 12(9), 835–838 +(2016) +[344] Li, Y., Wang, L., Liu, H., Zhang, Y., Hao, J., Pickard, C.J., Nelson, +J.R., Needs, R.J., Li, W., Huang, Y., Errea, I., Calandra, M., Mauri, +F., Ma, Y.: Dissociation products and structures of solid H2S at strong +compression. Phys. Rev. B. 93, 020103 (2016) +[345] Gordon, E.E., Xu, K., Xiang, H., Bussmann-Holder, A., Kremer, R.K., +Simon, A., K¨ohler, J., Whangbo, M.-H.: Structure and Composition of +the 200 K-Superconducting Phase of H2S at Ultrahigh Pressure: The +Perovskite (SH−)(H3S+). Angew. Chem. Int. Ed. 55, 3682–3684 (2016) +[346] Ishikawa, T., Nakanishi, A., Shimizu, K., Katayama-Yoshida, H., Oda, +T., Suzuki, N.: Superconducting H5S2 phase in sulfur-hydrogen system +under high-pressure. Sci. Rep. 6, 23160 (2016) +[347] Akashi, R., Sano, W., Arita, S. R.and Tsuneyuki: Possible “Magn´eli” +Phases and Self-Alloying in the Superconducting Sulfur Hydride. Phys. +Rev. Lett. 117, 075503 (2016) +[348] Yao, Y., Tse, J.S.: Superconducting Hydrogen Sulfide. Chem. Eur. J. 24, +1769–1778 (2018) +[349] Laniel, D., Winkler, B., Bykova, E., Fedotenko, T., Chariton, S., Mil- +man, V., Bykov, M., Prakapenka, V., Dubrovinsky, L., Dubrovinskaia, +N.: Novel sulfur hydrides synthesized at extreme conditions. Phys. Rev. +B. 102, 134109 (2020) +[350] Ortenzi, L., Cappelluti, E., Pietronero, L.: Band structure and electron- +phonon coupling in H3S: A tight-binding model. Phys. Rev. B 94(6), +064507 (2016) +[351] Durajski, A.P., Szcz¸e´sniak, R. R., Li, Y.: Non-BCS thermodynamic +properties of H2S superconductor. Physica C 515, 1–6 (2015) +[352] Durajski, A.P.: Quantitative analysis of nonadiabatic effects in dense +H3S and PH3 superconductors. Scientific reports 6(1), 1–8 (2016) +[353] Gor’kov, L.P., Kresin, V.Z.: Pressure and high-Tc superconductivity in +sulfur hydrides. Sci. Rep. 6(1), 1–7 (2016) +[354] Goncharov, A.F., Lobanov, S.S., Kruglov, I., Zhao, X.-M., Chen, X.-J., +Oganov, A.R., Konˆopkov´a, Z., Prakapenka, V.B.: Hydrogen sulfide at +high pressure: Change in stoichiometry. Phys. Rev. B 93(17), 174105 + +Springer Nature 2021 LATEX template +62 +Article Title +(2016) +[355] Jarlborg, T., Bianconi, A.: Breakdown of the Migdal approximation at +Lifshitz transitions with giant zero-point motion in the H3S supercon- +ductor. Sci. Rep. 6(1), 1–12 (2016) +[356] Bussmann-Holder, A., K¨ohler, J., Simon, A., Whangbo, M., Bianconi, +A.: Multigap superconductivity at extremely high temperature: a model +for the case of pressurized H2S. J. Supercond. Nov. Magn. 30(1), 151–156 +(2017) +[357] Szcz¸e´sniak, R., Durajski, A.P.: The isotope effect in H3S superconductor. +Solid State Commun. 249, 30–33 (2017) +[358] Azadi, S., K¨uhne, T.D.: High-pressure hydrogen sulfide by diffusion +quantum Monte Carlo. J. Chem. Phys. 146(8), 084503 (2017) +[359] Arita, R., Koretsune, T., Sakai, S., Akashi, R., Nomura, Y., Sano, W.: +Nonempirical Calculation of Superconducting Transition Temperatures +in Light-Element Superconductors. Advanced Materials 29(25), 1602421 +(2017) +[360] Quan, Y., Pickett, W.E.: Van Hove singularities and spectral smearing +in high-temperature superconducting H3S. Phys. Rev. B 93(10), 104526 +(2016) +[361] Sano, W., Koretsune, T., Tadano, T., Akashi, R., Arita, R.: Effect of +Van Hove singularities on high-Tc superconductivity in H3S. Phys. Rev. +B 93(9), 094525 (2016) +[362] Heil, C., Boeri, L.: Influence of bonding on superconductivity in high- +pressure hydrides. Phys. Rev. B 92, 060508 (2015) +[363] Ge, Y., Zhang, F., Yao, Y.: First-principles demonstration of supercon- +ductivity at 280 K in hydrogen sulfide with low phosphorus substitution. +Phys. Rev. B, 224513 (2016) +[364] Ge, Y., Zhang, F., Dias, R.P., Hemley, R.J., Yao, Y.: Hole-doped room- +temperature superconductivity in H3S(1−x)Zx (Z=C,Si). Mater. Today +Phys. 15, 100330 (2020) +[365] Liu, B., Cui, W., Shi, J., Zhu, J., Chen, J., Lin, S., Su, R., Ma, J., +Yang, K., Xu, M., Hao, J., Durajski, A.P., Qi, J., Li, Y., Li, Y.: Effect +of covalent bonding on the superconducting temperature of the H-S-Se +system. Phys. Rev. B, 174101 (2018) +[366] Amsler, M.: Thermodynamics and superconductivity in SxSe1−xH3. + +Springer Nature 2021 LATEX template +Article Title +63 +Phys. Rev. B 99, 060102 (2019) +[367] Guan, H., Sun, Y., Liu, H.: Superconductivity of H3S doped with light +elements. Phys. Rev. Res. 3, 043102 (2021) +[368] Wang, X., Bi, T., Hilleke, K.P., Lamichhane, A., Hemley, R.J., Zurek, E.: +Dilute Carbon in H3S Under Pressure. npj Comput. Mater. 8, 87 (2022) +[369] Cui, W., Bi, T., Shi, J., Li, Y., Liu, H., Zurek, E., Hemley, R.J.: Route to +high-Tc superconductivity via CH4-intercalated H3S hydride perovskites. +Phys. Rev. B 101(13), 134504 (2020) +[370] Lamichhane, A., Kumar, R., Ahart, M., Salke, N.P., Dasenbrock- +Gammon, N., Snider, E., Meng, Y., Lavina, B., Chariton, S., Prakapenka, +V.B., Somayazulu, M., Dias, R.P., Hemley, R.J.: X-ray diffraction and +equation of state of the C-S-H room-temperature superconductor. J. +Chem. Phys. 155, 114703 (2021) +[371] Goncharov, A.F., Bykova, E., Bykov, M., Zhang, X., Wang, Y., Chariton, +S., Prakapenka, V.B., Smith, J.S.: Synthesis and structure of carbon- +doped H3S compounds at high pressure. J. Appl. Phys. 131, 025902 +(2022) +[372] Sun, Y., Tian, Y., Jiang, B., Li, X., Li, H., Iitaka, T., Zhong, X., Xie, Y.: +Computational discovery of a dynamically stable cubic SH3-like high- +temperature superconductor at 100 GPa via CH4 intercalation. Phys. +Rev. B 101(17), 174102 (2020) +[373] Harshman, D.R., Fiory, A.T.: The superconducting transition temper- +atures of C-S-H based on inter-sublattice S-H4-tetrahedron electronic +interactions. J. Appl. Phys. 131, 015105 (2022) +[374] Hirsch, J.E., Marsiglio, F.: Unusual width of the superconducting +transition in a hydride. Nature 569, 9–10 (2021) +[375] Gubler, M., Flores-Livas, J.A., Kozhevnikov, A., Goedecker, S.: Missing +theoretical evidence for conventional room-temperature superconductiv- +ity in low-enthalpy structures of carbonaceous sulfur hydrides. Phys. +Rev. Mater. 6, 014801 (2022) +[376] Bykova, E., Bykov, M., Chariton, S., Prakapenka, V.B., Glazyrin, K., +Aslandukov, A., Aslandukova, A., Criniti, G., Kurnosov, A., Goncharov, +A.F.: Structure and composition of C-S-H compounds up to 143 GPa. +Phys. Rev. B 103, 140105 (2021) +[377] Somayazulu, M.S., Finger, L.W., Hemley, R.J., Mao, H.K.: High- +Pressure Compounds in Methane-Hydrogen Mixtures. Science 271, + +Springer Nature 2021 LATEX template +64 +Article Title +1400–1402 (1996) +[378] Smith, G.A., Collings, I.E., Snider, E., Smith, D., Petitgirard, S., Smith, +J., White, M., Jones, E., Ellison, P., Lawler, K.V., Dias, R.P., Salamat, +A.: Lower pressure phases and metastable states of superconduct- +ing photo-induced carbonaceous sulfur hydride (2021) arXiv:2111.15051 +[cond-mat.supr-con] +[379] Fu, Y., Du, X., Zhang, L., Peng, F., Zhang, M., Pickard, C.J., Needs, +R.J., Singh, D.J., Zheng, W., Ma, Y.: High-pressure phase stability +and superconductivity of pnictogen hydrides and chemical trends for +compressed hydrides. Chem. Mater. 28(6), 1746–1755 (2016) +[380] Liu, H., Li, Y., Gao, G., Tse, J.S., Naumov, I.I.: Crystal structure and +superconductivity of PH3 at high pressures. J. Phys. Chem. C 120(6), +3458–3461 (2016) +[381] Shamp, A., Terpstra, T., Bi, T., Falls, Z., Avery, P., Zurek, E.: +Decomposition products of phosphine under pressure: PH2 stable and +superconducting? J. Am. Chem. Soc. 138(6), 1884–1892 (2016) +[382] Flores-Livas, J.A., Amsler, M., Heil, C., Sanna, A., Boeri, L., Pro- +feta, G., Wolverton, C., Goedecker, S., Gross, E.K.U.: Superconductivity +in metastable phases of phosphorus-hydride compounds under high +pressure. Phys. Rev. B 93(2), 020508 (2016) +[383] Bi, T., Miller, D.P., Shamp, A., Zurek, E.: Superconducting phases of +phosphorus hydride under pressure. Stabilization by mobile molecular +hydrogen. Angew. Chem. Int. Ed. 56(34), 10192–10195 (2017) +[384] Liu, M., Huang, X., Wang, X., Huang, Y., Li, F., Wu, G., Li, X., Liang, +Y., Zhou, M. D. andLu, Liu, B., Cui, T.: Unravelling decomposition +products of phosphine under high pressure. J. Raman Spectosc. 49, 721– +727 (2018) +[385] Yuan, Y., Li, Y., Fang, G., Liu, G., Pei, C., Li, X., Zheng, H., Yang, +K., Wang, L.: Stoichiometric evolutions of PH3 under high pressure: +implication for high-Tc superconducting hydrides. Nat. Sci. Rev. 6, 524– +531 (2019) +[386] Li, X., Xie, Y., Sun, Y., Huang, P., Liu, H., C., C., Ma, Y.: Chemically +Tuning Stability and Superconductivity of P-H Compounds. J. Phys. +Chem. Lett. 11, 935–939 (2020) +[387] Shao, Z., Duan, D., Ma, Y., Yu, H., Song, H., Xie, H., Li, D., Tian, +F., Liu, B., Cui, T.: Ternary superconducting cophosphorus hydrides +stabilized via lithium. npj Comput. Mater. 5, 104 (2019) + +Springer Nature 2021 LATEX template +Article Title +65 +[388] Geng, N., Bi, T., Zurek, E.: Structural Diversity and Superconductivity +in S-P-H Ternary Hydrides Under Pressure. J. Phys. Chem. C 126, 7208– +7220 (2022) + diff --git a/h9AzT4oBgHgl3EQfMvvr/content/tmp_files/load_file.txt b/h9AzT4oBgHgl3EQfMvvr/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..97745bced1461e4ca7528579242d2d71fbdb5b03 --- /dev/null +++ b/h9AzT4oBgHgl3EQfMvvr/content/tmp_files/load_file.txt @@ -0,0 +1,4605 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf,len=4604 +page_content='Springer Nature 2021 LATEX template Crystal Chemistry at High Pressure Katerina P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Hilleke1 and Eva Zurek1* 1Department of Chemistry, State University of New York at Buffalo, 777 Natural Sciences Complex, Buffalo, 14260-3000, NY, USA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Corresponding author(s).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' E-mail(s): ezurek@buffalo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='edu;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Contributing authors: khilleke@gmail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='com;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 1 Introduction Jupiter, the largest planet in our solar system, is a gas giant comprised mostly of hydrogen and helium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' After penetrating its atmosphere, a mixture of hydro- gen and helium, methane and ammonia, one would find a massive sea of hydrogen – behaving very differently than one might expect from general chem- istry classes [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The hydrogen layer surrounding Jupiter’s core, under immense pressure and at high temperature, is metallic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This is just one example of how the pressure variable is so important in determining chemical and physical behavior, bearing consequences for the evolution and dynamics of planetary interiors [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Under pressure, the behavior of the elements as well as the compounds they form can change drastically from what we, as beings living and learn- ing at 1 atmosphere, are used to.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Hydrogen, as described above, can become metallic at high pressure, joining its Group I cousins the alkali metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' It may be perplexing then to learn that the alkali metals lithium [3] and sodium [4, 5] become semiconducting and insulating, respectively, when squeezed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Potas- sium starts to behave more like a transition metal [6–8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' When combined with one another, the unexpected behavior of the elements under pressure makes for correspondingly curious compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sodium chlorides with vastly divergent stoichiometries from the typical 1:1 (Figure 1) have been predicted and synthe- sized [9], stubbornly inert helium atoms form a compound with sodium [10], and metal superhydrides wherein the hydrogen atoms coalesce into clathrate- like networks have been reported [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Various “rules” for the behavior of materials at high pressure have been proposed by Prewitt and Downs [12], 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='01139v1 [cond-mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='mtrl-sci] 3 Jan 2023 Springer Nature 2021 LATEX template 2 Article Title Grochala and co-workers [13], and Zhang et.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [14], to name a few – and have undergone progressive revision as more is uncovered about the structures and materials far below us in the Earth’s core and far away in the planets and stars.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 1 The familiar crystal structure of NaCl, table salt, (a) is the only known stable compound of sodium and chlorine at ambient pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Under pressures exerted by a dia- mond anvil cell (b), a variety of additional stoichiometries and crystal structures decorate the sodium-chlorine phase diagram, including the P4/mmm Na3Cl, P4/m Na3Cl2, Imma NaCl2, and Pm¯3 NaCl7 [10] structures shown here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Over the past century, experimental methods have evolved to create pro- gressively higher pressures in a laboratory setting, allowing us to directly probe the behavior of materials under extreme conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pressures in the megabar range can now be routinely achieved, albeit still requiring delicate setups [15– 17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Diamond anvil cells (DACs) combine the superior hardness of this desired polymorph of carbon with its optical transparency, enabling the creation of the highest static pressures and interrogation of the sample via visual and spectroscopic means.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Engineering advances from multistage compression appa- ratuses [18, 19] to toroidal DACs [20, 21] have driven the experimental ceiling for static pressures ever higher.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Dynamic compression experiments under ramp or shock conditions, driven by gas guns, lases pulses, or magnetic fields can reach well into the terapascal regime [22–24], allowing us to study the behavior of diamond at 5 TPa [25] and iron at conditions thought to be in super-Earths [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Diagnostic techniques for characterizing the resulting substances can be difficult to implement in both dynamic and static compression, often requiring theoretical support for their interpretation [27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Far cheaper than high-pressure experiments are computations modeling high-pressure systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Band structures, phonons, Raman or infra-red spectra and numerous material properties can all be calculated without so much as stepping foot into a laboratory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Crystal structure prediction (CSP) techniques, not weighted down by preconceived, atmospheric pressure-based, notions of how atoms might appropriately arrange themselves in a unit cell, can identify which structures can exist under high-pressure conditions [28–30].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The com- putational exploration of potential energy landscapes will not be discussed Na3Cl2 b (a) NaCl NaCl2 NaCl7 Na3CSpringer Nature 2021 LATEX template Article Title 3 in our contribution;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' to learn about methods that can be employed to iden- tify the global, as well as important local, minima we point the reader to an excellent chapter in this book “Crystal Structure Prediction” by Andreas Hermann, Lewis J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Conway and Chris J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pickard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Exploratory calculations highlight promising phases for further experimental investigation, but compu- tation can just as well follow experimental results, elucidating behavior and filling in the gaps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The resulting feedback loop of experiment and theory has driven the discovery and characterization of a plethora of phases ranging from the superhard [31–33] to the superconducting [11, 34–38].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the following sections we build a framework for understanding the behavior of materials at high pressure, starting from the effects of pressure on the atoms themselves, driving electronic transitions and altering periodic trends.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' From there, the various manifestations of high pressure in the solid state are sorted into categories (which are not mutually exclusive, but rather illustrative), starting with exotic electronic structures and electrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' We dis- cuss compounds of the noble gases and those containing elements that are immiscible at ambient pressure, as well as crystal lattices that contain bizarre geometrical motifs and bonding configurations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Finally, we survey the effects of high pressure on superconductivity, a field that has recently undergone a veritable explosion as high pressure phases toe the line of room-temperature superconductivity [39].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 2 The atom under pressure Chemistry describes the interactions between the 118 distinct elements that are organized in the periodic table, proposed by Mendeleev while classifying elements according to their chemical properties observed at atmospheric condi- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The trends found within the periodic table can be used to compare atoms according to their size, the number of electrons surrounding their nuclei, and to make predictions as to whether those electrons are held tightly or loosely.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Moreover, the periodic table allows students and researchers to predict how different elements will interact with one another: will they form compounds, emulsions or alloys, or will they be unreactive?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' If reactivity is suspected, the periodic table can be used to guess if the bonds are covalent, ionic, or (usu- ally) somewhere in between.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Across the periodic table, trends in properties such as atomic radii, electronegativity, and oxidation state can be mapped, leading one to conclude that fluorine, the most electronegative element, will gain electrons in a binary phase thereby achieving an F− configuration with a filled valence shell.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' On the other hand, cesium, as the least electronega- tive element (neglecting francium, whose miniscule half-life renders it basically experimentally irrelevant), typically assumes an oxidation state of 0 or +1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1 Yet at high pressure, several LinCs phases have been predicted [44] where Cs attains unusual formal oxidation states thought to be in excess of -2 due to 1When dissolved in amines or ethers, Na, K, Rb, and Cs can assume the unusual oxidation state of -1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [40] These so-called alkalides are thought to form ion-pairs with solvated metal cations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [41– 43] Springer Nature 2021 LATEX template 4 Article Title substantial electron transfer from lithium to cesium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' How can the stability of these unintuitive stoichiometries and their resulting electronic structures be rationalized?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Let us consider electronegativity a little more.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Although Pauling’s [45] is the most widely adopted, a number of metrics have been used to produce scales of electronegativity for the elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In Pauling’s formulation, electronega- tivity differences between pairs of atoms A and B are calculated from the homo- and heteronuclear bond dissociation energies, then referenced against the electronegativity of H being set to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1 (later adjusted to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In this regime, the electronegativities of fluorine, lithium, and cesium are 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='98, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='98, and 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='79, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Several other scales have been proposed, including Mul- liken’s “absolute electronegativity” [46] being the average of the first ionization energy and electron affinity of an atom [47].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Dong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' modify the Mulliken definition for elements under high pressure taking as a reference the homoge- neous electron gas rather than the vacuum [48].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Allen’s electronegativities are derived from the average energies of the valence electrons in the atom [49], and a closely related scheme has recently been proposed by Rahm et al [50– 52], where electronegativity is calculated as the average of the electron binding energies of ground state valence electrons – approximately translated to the Allen scale if averaging over valence electrons alone.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Broadly speaking, the common factor of importance in all of these is the collection of atomic orbital energies and the differences between them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Under pressure, those change.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The prototypical model for understanding the quantized behavior of energy levels in a confined system is the particle in a box.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The resulting energy levels for a particle of mass m in a box of width L are given by En = ℏ2n2π2 2mL2 , where n is the principal quantum number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' To consider the effects of pressure on this model, we could simply make the box smaller by reducing the width L, which has the effect of increasing the En.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Thus, energy levels (orbital energies) will increase under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The complicating factor is that the rate of this increase is not the same for each orbital because the number of radial nodes plays a role.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The peak density of a 4s orbital is further from the nucleus than a 4p orbital, since the electrons occupying the 4s orbital must maintain orthogonality to those in the 1s, 2s, and 3s orbitals, while those in the 4p orbital must only contend with the 3p and 2p orbitals (this analogy may be extended to the 4d and 4f shells).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With more electron density being further from the nucleus, the electrons in the 4s orbital will feel the effects of pressure more strongly than those in the 4p orbital.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This is illustrated schematically in Figure 2a and 2b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At certain levels of confinement, electronic ns → np, ns −→ (n−1)d, and (n−1)d −→ (n−2)f transitions can become favorable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This is the reason why pressure drives rearrangements of the orbital energies of an atom, with ensuing electronic transitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' (In section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2, we will explore the opportunities presented by another sort of destination orbital, this time one not centered on an atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=') In compressed lithium, the ground state electronic configuration transitions from 1s22s1 → 1s22p1, while cesium undergoes a 6s1 → 5d1 pressure induced Springer Nature 2021 LATEX template Article Title 5 transformation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The exact pressures at which these electronic changes occur depend, of course, on the chemical environment of the lithium or cesium atoms – and in calculations, on the theoretical methodology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In Cs, the s → d tran- sition is thought to drive the transformation to the complex Cs-III structure with 84 atoms in the unit cell that is stable near 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2 GPa [53–58], while in lithium the pressure at which the s-p mixing occurs is believed to be somewhat higher, in the megabar to multimegabar range, dependent on the chemical environment and level of theory [57, 59–61].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another example includes potas- sium, whose complex phase diagram under pressure (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Figure 2c,d), has been in part attributed to this s → d electronic transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [62] The energies of core orbitals can also become relevant;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' for example, in Cs-VI the 5p bands hybridize with the 6s [63, 64], making them accessible for chemical interac- tions, and the core orbitals of K have also been proposed to be key to its structural diversity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [62] Overall, one effect of pressure on the alkali metals is that their electronegativities undergo quite a rearrangement, ending up with cesium being more electronegative than lithium [48, 57].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' From this perspective, the formation of cesium anions in the LinCs phases begins to make sense.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 2 External pressure on an atom raises the energy of atomic orbitals as they are con- strained to a smaller space, but the rate of this increase differs between s, p, and d orbitals (a), (b), favoring s → d transitions for many of the alkali metals, including potassium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Potassium adopts the bcc structure at 0 GPa (c) but at higher pressures takes on a series of complex structures including the incommensurately modulated host-guest tI19 phase [65] (d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Among the d-block similar reorderings are predicted to take place, with the group 10 metals Ni, Pd, and Pt preferring d10 closed-shell configurations, as compared to the s2d8 favored at ambient conditions, while the group 11 and 12 metals become electron donors [48].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The former transition is associated with a spike in the estimated chemical hardness – obtained as half of the HOMO-LUMO gap – of Ni, Pd, and Pt, reaching values comparable to some of the noble gases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This is contrary to the general trend where the hardness of a) b) c) bcc K (O GPa) d) tI19 K (20 GPa) Incommensurately modulated host-guest framework Ambient pressure orbital energies External pressure alters orderingSpringer Nature 2021 LATEX template 6 Article Title most of the elements in the periodic table decreases with pressure as energy levels become closer to one another [48].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The resulting changes in the relative hardness of pairs of elements can lead to changes in compound stability arising from HSAB (hard-soft acid-base) arguments, and the appearance of strange multicenter bonding manifolds in certain high pressure phases have been linked to general increases in softness [66].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' From transition-metal-like behavior in the s-block to relative inertness in the formerly-d8-transition metals, atoms under compression can behave very differently from their ambient-pressure selves, and the consequences for mate- rials under pressure are far-reaching.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' We will now explore some of the wild and wonderful structures and phenomena that result.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3 The crystal under pressure At ambient pressure a majority of the solid, metallic elements of the periodic table adopt very simple, symmetric, structures that are close-packed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The most stable geometries are those that minimize the free energy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' However, at low tem- peratures the entropic contributions between different solid phases are typically negligible, so computational studies often employ the enthalpy to determine the structures that are preferred.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With increasing pressure the enthalpy, consist- ing of the internal energy and pressure-volume terms (H = U +PV ), becomes dominated by the PV term.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' It would be natural, therefore, to imagine that close-packed structures with increased coordination numbers become preferred at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The reality, however, does not coincide with our expecta- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For example, within cesium the nearest neighbor coordination number first increases from 8 (bcc) to 12 (fcc), then decreases to about 10 (Cs-III), 8 (Cs-IV) and finally increases again to 10/11 (Cs-V) and 12 (Cs-VI).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These structural transitions are believed to be driven by the previously discussed pressure-induced s → d valence electronic transition within the constituent atoms, which causes the interatomic distances to become smaller compared to the ranges of the wavefunction [53].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Moving beyond elemental crystal structures, several compounds with seem- ingly bizarre stoichiometries (at least from the perspective of minds that experience a 1 atmosphere reality) have been predicted and/or synthesized under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The familiar combination of sodium and chlorine, table salt and prototypical ionic compound with a 1:1 ratio, is not the only stable crystalline structure in the Na-Cl phase diagram.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At least two unique stoichiometries, Na3Cl and NaCl3, were synthesized, and several others were predicted to become stable when squeezed [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Noble gases xenon, argon, and helium are active components of solid compounds that have been synthesized [10, 67, 68], and very hydrogen-rich compounds such as YH9, LaH10, and CaH6 that are high-temperature superconductors with superconducting critical temper- atures (Tcs) approaching room temperature have been made [11, 37, 69–72].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These metal superhydrides are materials-by-design success stories inspired by theoretical predictions [34, 35, 73].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Springer Nature 2021 LATEX template Article Title 7 In the following sections, we explore the plethora of exciting materials that can be created using high pressure, with all their intriguing structural and behavioral phenomena.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' While their stability and existence can be traced to the pressure driven electronic rearrangements of the constituent atoms described in Section 2, the manifestations of these rearrangements can take many forms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Below, we describe a variety of illustrative phases sorted into a series of cate- gories, but by necessity these categories will, at times, overlap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nevertheless, all exemplify the ramifications of high pressure on solid-state chemistry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1 Electronic structure At atmospheric pressure, the stoichiometries of many inorganic solid-state compounds can be predicted from the most common oxidation states of their constituent elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Usually, alkali metals and alkaline earth metals possess oxidation states of +1 and +2 respectively, so when combined with the O2− ion one would expect Na2O and MgO, as well as K2O and SrO to form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The noble gases are mostly inert, the p-block is amenable to forming covalently bonded networks, and while a wide variety of oxidation states, which correspond to various filled or half-filled orbitals, are available to several transition metals, Zn, Cd, and Hg steadily persist in maintaining their d10 configurations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The consequences of the orbital energy shifts discussed in Section 2 mean that several of these “rules” no longer apply at high pressures, and elements can adopt unusual oxidation states in compounds with unexpected stoichiometries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The LinCs phases [44] used to illustrate the effects of pressure on electroneg- ativity provide one such example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Above 70 GPa, the very lithium rich Li5Cs phase is predicted to become stable, joined at higher pressures by Li3Cs, Li4Cs, and LiCs [44].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Remarkably, the calculated Bader charges on the Cs atoms in these stoichiometries are all more negative than -1, and since calculated Bader charges frequently underestimate formal oxidation states, in these compounds Cs may attain a formal oxidation state that is potentially lower than -2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' While alkali metal anions (alkalides) have previously been captured at ambient pres- sures with cryptands [74], they achieve charges only up to -1 with the additional electron going into the ns orbital.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the case of the LinCs phases, however, pressure induces significant Li 2s → 2p and Cs 6s → 5d electronic transitions, with the latter increasing progressively with higher Li content, thereby facil- itating the acceptance of electron density by the typically unoccupied Cs 5d orbitals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For K, Cs, and Rb the pressure-driven ns → (n − 1)d transitions led to predictions of transition-metal like behavior [6, 75] in the formation of inter- metallic compounds with actual transition metals [8, 76–78].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Within some of these compounds, the transition metal elements assume exotic electronic con- figurations, as in the case of the predicted potassium iridide K3Ir (Figure 3a) containing the Ir3− anion [79].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Ir 5d orbital becomes fully occupied as a result of electron transfer from K, echoed in the later predicted Rb3Ir and Cs3Ir phases [80].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' K3Ir and Rb3Ir share the Pmnm Cu3Ti structure type, while Springer Nature 2021 LATEX template 8 Article Title Cs3Ir adopts the P21/m Ni3Ta type, which consist of Ir@M8 and M@M8 dis- torted cubes, and Ir@M12 distorted cuboctahedra respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In combination with Li under high pressure, Au displays a similar ability to adopt a signif- icantly negative formal charge of less than -3 in the predicted phases Li4Au and Li5Au – where electrons donated from Li are placed into the empty Au 6p orbitals [81], which are less destabilized than the Li 2s or 2p under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3 High pressure compounds where the transition metal atoms adopt curious electronic configurations in (a) K3Ir [79] (Cu3Ti type), with Ir@K8 and K@K8 distorted cubes and iridide I3− anions, and (b) HgF4 [82], in which d8 configurations on the Hg lead to square planar geometries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pressure can also promote chemical interactions with core or semi-core orbitals, as in the case of HgF3 and HgF4 [82].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These stoichiometries are pre- dicted to become stable above 73 and 38 GPa, respectively, and above 200 GPa HgF4 is computed to decompose into HgF3 and F2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The I4m symmetry HgF4 crystal possesses square planar HgF4 units typical of a d8 organometallic com- plex, with the Electron Localization Function (ELF) [83] confirming covalent Hg-F interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' To form the four Hg-F bonds, not only the Hg 6s but also two of the semicore 5d electrons are required.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In HgF3, the Fm¯3m structure (which distorts below 100 GPa to C2/m symmetry) involves a fluorite-type HgF+ 2 lattice stuffed with F− ions, leaving Hg with a d9 configuration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A series of predicted CsFn phases, in which the Cs 5p electrons participate in Cs-F covalent bonds, are another example of the pressure-induced activation of core electrons [84].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Their crystal structures display motifs resembling the isoelectronic [XeFn]− molecules – for example, Fdd2 CsF5 contains planar pentagonal CsF5 units, similar to the [XeF5]− anion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With increasing fluorine content, the formal oxidation state on cesium reaches values greater than +1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2 High Pressure Electrides Electrides are solids where the electrons, localized on non nuclear-centric sites, behave as anions [85].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' They are conceptually related to solvated electrons, in a) K3Ir (Cu3Ti type) b) HgF4 Iridide Ir3- d8 Hg(IV)Springer Nature 2021 LATEX template Article Title 9 which the excess electrons can be thought to occupy cavities in the fluid [86], as well as alkalide liquids [41, 42] or alkalide solids where alkali metal anions fill the interstitial voids [74].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Although many types of electride families are known at atmospheric conditions, including those that are organic, inorganic, intermetallic and those where the electron localization is restricted to various dimensions or possesses topological properties, herein, we restrict the discus- sion to high pressure electrides: systems where the electron localization occurs as a response to compression [87].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Though the formation of high pressure electrides has been rationalized in many ways, including pressure induced orbital rehybridizations [4, 88, 89], and multicenter bond formation [90, 91], herein we focus on a simple model proposed by Miao and Hoffmann [87, 92].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' As atoms in a solid compound are compressed, raising their orbital energies, the electrons in the highest-energy orbitals can vacate the atom entirely and occupy the interstices of the crys- tal lattice instead.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' To understand why this might occur, Miao and Hoffmann pointed out that orbitals can be ascribed to these voids, which can be thought of as interstitial quasiatoms (ISQs).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At ambient pressure, the ISQ energies are higher than the atom-centered ones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' However, unlike the atom-centered orbitals, those of the ISQ do not experience the repulsive effect caused by the core electrons, and their increase in energy with pressure will be less than the atom-centered orbitals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' When the ISQ orbital energies fall below the atom- centered ones they will be occupied, thereby localizing the valence electrons in the interstitial regions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These electrons, detached from the nuclei, serve as anions and the corresponding compounds are called electrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For several simple metals, high-pressure phases identified as electrides via calculations have been subsequently studied experimentally.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In sodium, Neaton and Ashcroft posited that under pressures high enough to induce overlap of the 2p orbitals, a combination of Pauli repulsion and core orthogonality constraints would drive the valence electrons away from the ionic cores, to localize in the crystalline interstices instead.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This electronic redistribution would result in a metal-to-insulator transition [88].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Later CSP calculations – with experimental confirmation of an optically transparent, wide-bandgap insulating phase in the same publication – proposed an insulating hP4 phase with P63/mmc sym- metry to become stable above 260 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [4] The new hP4 phase was in fact experimentally observed at pressures as low as 200 GPa, but the discrepancy was ascribed to a combination of thermal effects as well as the preferential stabilization of metallic states by the computational method employed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mov- ing to higher pressures and temperatures, evidence for the hP4 phase has been obtained in shock-ramp experiments [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' However, in situ X-ray diffrac- tion (XRD) revealed peaks that could not be attributed to hP4 between 240-325 GPa at temperatures in the thousands-of-degrees Kelvin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Consistent with these dynamic compression experiments, calculations showed that the free energy of a P63/m symmetry phase that is a topological electride was lower than that of hP4 at these pressures and temperatures [93].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At higher Springer Nature 2021 LATEX template 10 Article Title pressures yet, ca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5 TPa, sodium is predicted to adopt a curious, metal- lic cI 24 electride phase consisting of Na12 icosahedra [94].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the insulating hP4 structure that dominates much of the high-pressure landscape in sodium, highlighted in Figure 4a, the atoms occupy the Ni sites of the Ni2In structure type with the ISQs on the In sites, in line with the treatment of this phase as (Na+)2E2− (where E2− denotes a doubly-occupied ISQ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, several A2X alkali metal chalcogenides adopt the antifluorite Fm¯3m crystal structure at ambient conditions, but eventually transition to the Ni2In structure type under pressure [95, 96].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Potassium also adopts the hP4 phase when squeezed [62, 97].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lithium presents another example of complex structural and electronic behavior under pressure, as first postulated by Neaton and Ashcroft who suggested it may adopt an insulating, paired ground state [59].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Subsequent experiments revealed that Li assumes the same semimetallic cI16 [60] struc- ture found in Na [98].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At higher pressures, Li transitions to a number of unique phases, such as those with orthorhombic C-centered lattices and 88, 40, and 24 atoms in their unit cells which have been observed [99].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One of these, oC40 with the Aba2 space group, is an electride displaying especially interesting behavior [100].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In this phase, ISQs occupy three separate symmetry-distinct sites, two that are doubly occupied (EII) and the third singly occupied (EI), so its primitive cell can be considered as Li20EII 8 EI 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The EI-EI distance remains roughly constant at a short 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='3 ˚A from 50-80 GPa, with an elevated elec- tron density found between the ISQs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Crystal Orbital Hamilton Population (COHP) [101], ELF, and projected density of states (PDOS) analyses all indi- cate bonding character between the EI sites, and examination of the Γ-point band-decomposed charge densities revealed bonding and antibonding states analogous to the σg and σ∗ u orbitals in H2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, computations have shown that ISQs can form covalent, ionic and metallic bonds with atoms as well as with other ISQs [87, 89, 100, 102, 103].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For this reason, Miao has espoused the idea that ISQs may be thought of as a chemical element and placed above helium in the periodic table under pressure [104].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The proclivity of the elements to ISQ formation has been investigated by comparing the energies of their orbitals calculated at different pressures using a He confinement model with those of an ISQ 1s orbital [92].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Unsurprisingly, Li and Na were found to favor ISQ formation at relatively low pressures, with Mg, Al, In, and Tl predicted to follow suit at higher pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Among the heavier alkali metals, the energies of the valence s orbitals were found to rapidly increase in energy relative to that of the ISQ 1s – but as previously discussed, these elements are also susceptible to a pressure induced electronic ns → (n − 1)d transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Within cesium, for example, the increased d occu- pation, already noted by Sternheimer in 1950 [53], was invoked to explain the curious structure of Cs-IV adopted at 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='3 GPa, where the Cs atoms possess a coordination number of 8 [55].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This decrease in coordination number with increasing pressure appeared so counterintuitive to Linus Pauling that he pre- sented an alternative structure solution assuming cubic symmetry and invoking icosahedral clusters [105].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pauling’s hypothesis turned out to be incorrect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Springer Nature 2021 LATEX template Article Title 11 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4 High pressure electride formation is observed in a number of complex high-pressure phases of the alkali metals, including both the insulating (a) hP4 phase of Na [4] (Ni2In struc- ture type) where ISQs containing two electrons occupy the In sites, and (b) semiconducting oC40 Li, containing three inequivalent ISQ sites of which two (black) are doubly occupied, and the third (pink) is singly occupied so a bond is formed between nearest neighbors, as highlighted by the lines that join them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Importantly, von Schnering and Nesper [106] recognized that the Cs atoms in the Cs-IV structure occupied the Th sites of the ThSi2 lattice – and that the valence electron density exhibited maxima not near the Cs atoms but at the Si positions of ThSi2 and between the Si-Si bonds, dubbing this phase an electride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This I41/amd structure is also assumed by K and Rb under pressure [107–109].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' If the high-pressure phase behavior of the alkali metals were not complex enough, Na, K, and Rb all adopt different versions of an incommensurately modulated host-guest lattice (similar to the W5Si3 type), often referred to as the tI19 structure, a model of which is illustrated for K in Figure 2d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [4, 65, 107, 110–114] All three share the same host lattice, but display different periodicity in the guest lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In a study using commensurate approximants to model the electronic structure, electron localization in the interstitial spaces was found, with some highly localized basins as well as a 1D channel of electron density lying in channels of the host structure [114].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another study proposed that electrides in the heavy alkali metals could be stabilized via ferromagnetic ordering [75].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Numerous predicted binary and ternary systems under pressure also behave as electrides, including two Li3Fe phases (with P6/mmm and P4/mbm sym- metries) [115], P4/mbm Na3Fe [116], and a superconducting Y3Si phase [117].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In a range of LinI stoichiometry phases predicted above 50 GPa, ISQs form within the interstices between I-centered Li polyhedra – but higher pressures drive electron density back from the ISQs to the iodine atoms, filling the 5p and eventually the 5d orbitals, skipping the 6s, in line with disfavoring s orbital occupation under pressure [103].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Finally, several electride phases have been two-electron ISO a) hP4 Na b) oC40 Li one-electron ISQ ISOSpringer Nature 2021 LATEX template 12 Article Title identified involving the famously unreactive noble gases - including a particu- larly surprising case [10] where the noble gas is crucial to the stability of the synthesized structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='3 Compounds of noble gases Where did all the xenon go?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Relative to the abundance of Ar and Kr in the Earth’s atmosphere, the amount of Xe is strikingly lower than it should be, a problem known as the “missing xenon paradox”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Geoscientists have explained this discrepancy in many different ways [118–120], but a growing body of evidence suggests that the Xe did not escape, and instead it has been incorporated into the miner- als found within the Earth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At Earth’s core pressures, both the atomic orbital energies and the relative electronegativies of the elements, including Xe and the Fe and Ni that comprise the majority of the core, are significantly per- turbed [48, 50].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Therefore, it should not be a surprise that their reactivity differs from our 1 atmosphere expectations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Prior to the advent of widespread CSP, computational investigations con- cluded that Xe incorporation into Fe and Ni would not occur, at least not to a large extent [121, 122].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These studies, which relied on the assumption that the Xe-metal alloys adopted crystal lattices similar to those of the ele- mental metals, turned out to be incorrect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Later CSP-based studies predicted the emergence of stable Xe-Fe and Xe-Ni compounds above 250 and 200 GPa, respectively [123], with Pm¯3m XeFe3 (AuCu3-type) and Pmmn XeNi3 (based on Xe@Ni12 cuboctahedra) having the lowest enthalpies of formation, although P¯62m XeFe5 and XeNi5, as well as P21/m XeNi6 also appeared on the con- vex hull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Confirming these predictions, experimental studies have synthesized XeNi3 and Xe(Fe/Ni)3 phases at high pressure, although with slightly differ- ent structures than those predicted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This includes Pm¯3m for XeNi3, either as an ordered AuCu3 [67] or disordered CrNi3 alloy [68], and a mixture of fcc and Pmmn-symmetry phases for XeFe3 [68].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In these systems, the Fe and Ni atoms behave as oxidants, accepting 5p electron density from Xe [68, 123], in agreement with the predicted increase in electronegativity differences between Xe and the transition metals at high pressure [50].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Xe2FeO2 and XeFe3O6, both involving substantial Fe-O and Xe-O bonding, have also been computed to be stable at pressures relevant to the Earth’s core [124].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The high-pressure ArNi phase, in which some Ni 3d electron density is transferred to the Ar 4s, inducing a magnetic moment on the Ni, has been synthesized [125].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' CSP calculations have also predicted stable compounds containing Xe, or other noble gas (NG) elements, and Mg above 125-250 GPa [102].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This includes Mg-Xe and Mg-Kr phases, which adopt structures based on stacked square lattices of Mg and the NGs in different patterns, ranging from the CsCl type (Pm¯3m) to more complex P4/nmm or I4/mmm arrangements for MgNG and Mg2NG stoichiometries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Compounds of Mg with Ar, on the other hand, were found to favor hexagonal arrangements such as anti-NiAs type MgAr (P63/mmc).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In these compounds the energies of the metal 3s orbitals Springer Nature 2021 LATEX template Article Title 13 increase precipitously in comparison to the outer shell d orbitals of the noble gases inducing Mg 3s → NG d orbital transfer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The ELF of Mg2NG (NG=Xe, Kr, and Ar) phases shows an additional interesting feature: ISQ formation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This occurs far below the pressures at which elemental Mg is predicted to form an electride [89, 92].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Two reasons have been used to explain this phe- nomenon [102].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' First, far fewer ISQ sites – concomitantly occupying less space – relative to the elemental Mg electride are necessary to accept the displaced valence electrons of Mg, as many of them are transferred to the NG atoms instead.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In addition, the NG atoms promote the formation of larger interstitial spaces in the structure, stabilizing the ISQ at lower pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Under moder- ate pressures of ca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 50-300 GPa, the energies of the outer shell Xe d orbitals are similar to those of the ISQ 1s, although with higher pressure they become lower in energy, congruent with the gradual ISQ 1s → NG d electron transfer with increasing pressure up to 600 GPa [102].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Several other stable compounds of Xe have been predicted at high pres- sures, including XeO, XeO2, and XeO3 [126], as well as Xe3O2, Xe2O, and Xe7O2 [127], while Xe2O5 and Xe3O2 have both been experimentally observed at pressures lower than 100 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' [128] Krypton oxide, KrO, has been predicted as well [129], as have xenon nitrides [130, 131] and carbides [132].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Fluorides of argon [133], krypton [134], and xenon [135] – with Xe-Xe dimers cropping up in Xe2F and XeF – have all been predicted.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Several xenon chlorides including XeCl, XeCl2, and XeCl4, with the former two being metastable by 10 GPa and reaching the convex hull by 60 GPa have been computationally studied [136].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' When they are combined with Li, the noble gases Ar [137] and Xe [138] are predicted to behave as anions, with the Li 2s orbital rising above the Xe and Kr outer shell d orbitals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Several cesium xenides are predicted to be stable at high pressures, many adopting alternate colorings of a distorted bcc lattice [139].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' There is experimental evidence for the formation of a phase mixing Xe with water ice at conditions expected for planets such as Uranus and Neptune [140].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Helium is famously the most inert element at ambient pressure by virtue of its closed-shell electronic configuration, zero electron affinity and large ioniza- tion potential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nonetheless, a number of stable helium-containing compounds have recently been predicted at high pressure, including those with iron [141], ammonia [142], water [143, 144], nitrogen [145, 146], and even with other noble gases [147, 148] (the van der Waals compound NeHe2, a Laves phase in the MgZn2 structure has been experimentally observed [149]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A particularly noteworthy example is provided by Na2He [10], an electride phase with a fluorite-like lattice (Figure 5a) in which every Na8 cube that does not contain a He atom is instead occupied by an electron pair (Figure 5b), so that the phase can be expressed as (Na+)2(E2−)He.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Although He does not participate in any bond formation, its presence is nevertheless a crucial stabilizing force in this phase, which has been successfully synthesized above 113 GPa [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A subsequent computational study showed that the He atoms act as inert “spacers” to reduce the Madelung repulsion resulting from the unequal amounts of cations and anions in the parent (Na+)2(E2−) phase [150].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Reaction Springer Nature 2021 LATEX template 14 Article Title Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5 Conventional unit cell of helium-containing Na2He [10] in the Fm¯3m space group (a), an electride with localized electron pairs occupying octahedral vacancies (b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The mech- anism by which He incorporation serves to stabilize the structure is illustrated in (c), which shows schematically how He insertion reduces electrostatic repulsions involved in the A2B (Na2E) stoichiometry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' enthalpies for helium in combination with ionic AB, A2B and AB2 compounds revealed that He incorporation was generally favored when the cation:anion ratios were unequal such as in MgF2 and Li2O, but not for AB-type phases such as LiF or MgO, in line with the prediction of successful He incorporation into certain alkali oxides and sulfides [151].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Helium placement in the ionic lattices tends to separate ions of similar charge as shown schematically in Figure 5c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' An FeO2He phase in which the Fe and O atoms form a fluorite lattice and the He atoms occupy the remaining Fe8 cubes (isopointal to Na2EHe) was predicted to be stable above 120 GPa [152], with He appearing to play the same role of spacing agent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This mechanism allows even the most inert noble gases to play an active role in stabilizing compounds at high pressure, all without forming a single chemical bond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='4 Miscibility under pressure The noble gases obtained their names due to their general lack of reactiv- ity, but under ambient conditions, numerous combinations of elements resist mixing to form alloys or stoichiometric compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The proclivity or reluc- tance of a pair of elements towards mixing has been explained in a variety of ways, resulting in predictive rules including those of Hume-Rothery and co- workers [153], Miedema’s model [154, 155], Darken and Gurry’s maps [156], and more [157–159].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' As we will soon see, pressure turns out to be a useful variable that can alter the (im)miscibility of two or more elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Consider, for example, magnesium and iron, whose large size and small electronegativity differential at ambient pressure makes compound forma- tion intractable [160].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' According to Miedema’s rules, compound formation is favored by large electronegativity differences and similar charge densities [154].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The electronegativity difference between the two elements greatly increases under pressure [48, 57] – and because Mg is more compressible, its radius a) Na2He b) Na2EHe AB stoichiometry electrostatic attraction no spacer needed A2B stoichiometry attraction + repulsion spacer needed Na>EHel 111 directionSpringer Nature 2021 LATEX template Article Title 15 approaches that of Fe when squeezed, thereby increasing the miscibility of the two elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' As a result, stable Mg-Fe compounds have been computation- ally [161, 162] and experimentally [163] studied under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The higher compressibility of K has also been found to favor compound formation with transition metals such as Ag, even at pressures below which K is anticipated to undergo an ns → (n − 1)d transition [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another case-in-point of pressure induced reactivity are the Li-Be alloys predicted to be stable in the megabar regime [164].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' By 20 GPa LiBe2 reaches the Li-Be convex hull, where it is joined by LiBe4 (shown in Figure 6a) and LiBe by 80 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At 100 GPa the latter trades its place on the convex hull with Li3Be.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Alignment of strong diffraction peaks with 2kF (twice the free-electron Fermi wavevector) is suggestive of stabilization through a Fermi surface-Brillouin zone interaction mechanism, which has been used to explain the particular stabilities (and electron counts which make them so) of Hume- Rothery electron phases [165, 166].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Furthermore, at around 82 GPa, an odd feature emerges in the DOS curve of P21/m LiBe: the base of the valence band appears as a step-like function, remains flat for ∼4 eV, and sharply increases once more before more complex features take over.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This is linked to a distinct separation – made possible by the pressure-induced increase in the electroneg- ativity difference between Li and Be – between high- and low-electron density planes associated with the Be and Li atoms, respectively, leading to quasi- 2D-like behavior in a geometrically 3D structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Stabilization through Fermi sphere interaction with higher zones (referred to as Jones zones from the Mott- Jones formulation of this mechanism) was invoked to propose a high-pressure NaAl phase in the NaTl structure type [167] just above 12 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another element that does not undergo compound formation with Li at ambient conditions is Fe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Just above 40 GPa, however, Li3Fe (P6/mmm) and LiFe (Fd¯3m, NaTl-type) phases are computed to lie on the convex hull [115].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Some interstitial electron localization appears in Li3Fe, both in P6/mmm symmetry as well as the P4/mbm symmetry computed to prevail just above 60 GPa which is shown in Figure 6b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Both Li3Fe phases are host-guest lat- tices, with the Fe atoms lying in larger hexagonal (P6/mmm) and heptagonal (P4/mbm) channels, and the electron localization is found within the smaller triangular and square channels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Na3Fe is predicted to adopt this same P4/mbm phase between 120 and 300 GPa [116].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' An additional Li3Fe2 phase with C2/m symmetry appears on the convex hull by 80 GPa, involving Fe zigzag chains in combination with alternating Li linear or armchair chains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Recently, a host of bismuth-containing phases have been found to be sta- bilized at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Bismuth is a component in numerous topologically nontrivial materials [170, 171] and superconducting systems at ambient pres- sure [172–174], yet under these conditions does not form stable compounds with many of the transition metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In combination with the exotic bulk prop- erties of bismuth, the various magnetic or otherwise electronically nontrivial properties of the transition metals make for a tantalizing combination in high pressure experiments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Under pressure, the estimated electronegativity of Bi Springer Nature 2021 LATEX template 16 Article Title Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6 High pressure enables the mixing of elements which do not form compounds or alloys otherwise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These include (a) LiBe4, a layered compound of Li and Be [164], (b) Li3Fe, a host-guest compound with Fe atoms as the guests to a Li-based host lattice [115], and (c) CuBi [168, 169], one representative of the many bismuth-containing compounds which have been found at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' CuBi has been experimentally observed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' drops precipitously in comparison to most of the d-block, rendering it more reactive [48, 57].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Indeed, above 32 GPa, FeBi2 [175, 176] was observed to form in DAC experiments in the I4/mcm Al2Cu structure type shared by NiBi2 [177] and MnBi2 [178] (both are stabilized by high pressure although other Ni-Bi and Mn-Bi phases are accessible at atmospheric pressure), as well as certain high-pressure transition metal pnictides [179–181].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A second phase, FeBi3 with Cmcm symmetry, has also been predicted to lie on the convex hull between 36 and 39 GPa, but this narrow stability range likely hinders synthetic accessibility [176].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, high-pressure high-temperature methods have been used to syn- thesize a wide variety of bismuth-containing compounds including CoBi3 [182, 183], which adopts the Pnma NiBi3 structure type, becomes stable by 5 GPa, and is a superconductor with a Tc just below 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Synthesized binaries of Cu and Bi include Cu11Bi7 at 3 GPa [184] and CuBi (Figure 6c) at 6 GPa [168, 169], as well as a I4/mmm Cu2Bi phase above 50 GPa [185], which a) LiBe4 (P6/mmm) c) CuBi (Pmma) b) Li3Fe (P4/mbm)Springer Nature 2021 LATEX template Article Title 17 is possibly overtaken by a Cu7Bi2 phase [177].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Of the second-row transition metals, MoBi2, also in the Al2Cu structure type, has been synthesized above 35 GPa, while evidence for a Mo-Bi bcc-type alloy appeared above 5 GPa [186].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Linear Approximation to Enthalpy (LAE), a tool for rapid and com- putationally cheap evaluation of formation enthalpies, was used to explore the high-pressure stability of structures in binary ambient-immiscible sys- tems [177].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In concert with the minima hopping CSP method, several new phases were found to be stabilized by 50 GPa: PbAs, Si3Al, SiAl, SiAl3, BiSn3 – yet another bismuthide – In3Fe, Hg3In, HgIn, HgIn3, Hg3Sn, ReSn3, ReBr3, ReGa, and ReGa3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Only a limited range of stoichiometries (A3B, AB, and AB3) was sampled, encouraging further investigation into each of these systems – but now there is preliminary data to suggest fertile ground.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5 Geometries and bonding In the previous sections, we have explored curious electronic interactions made possible by external pressure and compound formation between unexpected species.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Here, we shift our focus to the particular geometrical arrangements that emerge in materials under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With higher density, atoms are forced into closer proximity – the possibility of electride formation notwithstanding – which can promote multicenter bonding in both electron-poor and electron-rich contexts as coordination numbers increase [13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' When electron-precise species are closely bunched under compression, electron-deficient multicenter bonding can emerge as the constituent electrons are needed to span more bonding interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Bond symmetrization is a frequent secondary consequence of compression: an asymmetric fragment forced to occupy a progressively smaller space has less room for asymmetry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Often, this leads to a collapse into a symmetric and multicentered bonding regime, as was predicted for water ice under pres- sure by Pauling [187].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' He suggested that the intermolecular hydrogen bonds between adjacent water molecules would shorten with pressure [187], eventu- ally becoming equivalent with the intramolecular O-H bonds, as illustrated in Figure 7a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This prediction was verified experimentally upon the discovery of ice X, where the lone pairs of the oxygen atoms are used to form addi- tional covalent bonds, rendering the oxygen atoms tetrahedrally coordinated by hydrogens in a diamond-like network [188, 189].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pressure-induced hydro- gen bond symmetrization has also been observed in computations on hydrogen halide systems such as HF, HCl and HBr [190], as well as in the record-breaking superconductor Im¯3m H3S [191].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Small homoatomic clusters alien to the 1 atmosphere pressure-trained mind are found or predicted for other elements in high pressure crystal structures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The wide structural variety has been explained by the increased stabiliza- tion of homonuclear bonds as compared to more polar or ionic bonds under pressure [104], favoring single-element clustering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' An example of a compound containing novel homonuclear motifs is Pnma NaCl3 (Figure 7b), computed to be stable from 20 to 48 GPa, featuring a linear Cl− 3 anion reminiscent of Springer Nature 2021 LATEX template 18 Article Title the more familiar triiodide I− 3 [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another such motif is the pentazolate N− 5 ring, which can store more energy than the related azide anion N− 3 , but is challenging to synthesize at ambient pressure [192, 193].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This species, ubiqui- tous in high-pressure phases, is predicted to be a constituent of LiN5 [194–197] – a phase that has been successfully quenched to ambient conditions after synthesis at 45 GPa [198] – to sodium pentazolates NaN5 and Na2N5 [199], CsN5 [200], CuN5 [201], MgN10 and BeN10 [202], ZnN10 [203], BaN5 and BaN10 [204], SnN20 [205], and IrN7 [206].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Polynitrogen chains feature in many proposed high-pressure compounds of Cs [200], Fe [207], Zn [203], Ba [204], Sn [205], Cd [208], Gd [209], and Ta [210], the last of which has been experi- mentally realized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' High pressure also facilitates the formation of silicon clusters in predicted phases including Si4 squares in CaSi [211], extended networks and clathrate-like cages in silicides such as CsSi6 [212], MgSi5 [213], and several lithium silicide compounds [214].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 7 Geometrical and bonding adaptations at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With pressure (a), cova- lent O-H and intermolecular hydrogen bonds between separate water molecules in the ice VIII phase equilibrate to yield the symmetric ice X phase [188, 189].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pressure coincides with the appearance of strange motifs including (b) linear trichloride anions in Pnma NaCl3 [9], (c) clusters of Ge dumbbells (inset) in I4/mmm BaGe3 [215, 216], and (d) Li8H “superatom-like” building blocks (inset) in Abm2 Li5H [217].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Except for Li5H, all have been experimentally realized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' b) NaCl3 (Pnma) a) Bond symmetrization in ice differentiated symmetric O-H bonds ice VIII (I41/amd ice X (Pn3m) )BaGe3 (I4/mmm) d) Li5H (Abm2) Ge2 dumbbell Li8H clusters "superatom"Springer Nature 2021 LATEX template Article Title 19 Tetrel clusters comprise a family of polar intermetallic I4/mmm symme- try compounds formed from alkaline earth or rare earth metals and group 14 tetrels in a 1:3 ratio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' An example of these isotypic compounds, BaGe3, is shown in Figure 7c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The clusters within it may be described as tetrel dumb- bells condensed into cubes, which are capped on four equatorial faces by additional dumbbells shared with a neighboring cube, forming a loose three- dimensional network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This structure has been experimentally observed in Ca, Y, and Lu silicides [218] (and later identified in a CSP investigation of the Y-Si system [117]), and a related distorted I¯42m BaSi3 phase has also been synthesized [219].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Alkaline earth trigermanides CaGe3 [220], SrGe3 [221], and BaGe3 [215, 216] have also been found to adopt this structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The distribution of electrons within the clusters aligns with two-center two-electron bonds along the tetrel dumbbells, with multicenter interactions between the tetrel and rare earth/alkaline earth [215, 218, 222].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Superconductivity has been measured in some of these compounds, albeit at low temperatures [216, 218], augmented by predictions from first principles calculations [104, 117, 222].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Additionally, the stability conferred by the strong covalently bonded networks permits some of these phases, synthesized at high-temperatures and high-pressures, to be recovered at ambient conditions [219, 222].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A similar example is presented by elemental carbon – diamond is its ground state at high-pressures, but due to the immense strength of its sp3 covalently bonded network, the energetic bar- rier for its transition to the lower-enthalpy allotrope graphite is too high and it persists “forever” under ambient conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Furthermore, laser-driven ramp compression studies of carbon to 2 TPa have found that carbon stubbornly maintains the diamond structure well beyond its predicted high-pressure sta- bility limits [223], as the barriers to breaking the sp3 bonds remain large under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another example of unique clusters predicted to form only at high pressures are found within a family of lithium subhydrides [217].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Computations uncov- ered two nearly isoenthalpic Li5H phases that had the most negative enthalpies of formation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Both were built of Li8H units that behaved as superatoms anal- ogous to similar units in synthesized Rb9O2 and Cs11O3 suboxides [224].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One of these, with Abm2 symmetry, is shown in Figure 7d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Li8H cluster, a dis- torted bicapped trigonal antiprism of Li encapsulating a single H atom, has one electron in excess of the closed-shell octet and thus behaves as a superalkali atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In addition to the well-known H2 molecular units and H− hydridic species, hydrogen atoms can form other distinct clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One of these, the trihydrogen cation, H+ 3 , is in fact one of the most abundant species in the universe – but it is also largely relegated to interstellar space and the atmospheres of gas giant planets [225].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' High pressure crystal chemistry offers another opportunity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The halogen polyhydride Cc H5Cl [226–228], predicted to become stable above 100 GPa, contains slightly distorted H+ 3 clusters with H-H distances of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='74, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='97 and 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='01 ˚A [226].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' By 300 GPa the three distances converge to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='87-0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='88 ˚A, yielding a nearly-perfect equilateral triangle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With sufficient pressure, this H+ 3 Springer Nature 2021 LATEX template 20 Article Title unit interacts with a neighboring H2 molecule forming a twisted bowtie-like loosely interacting H+ 5 motif [226, 228].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Metastable predicted H2F, H3F, and H5F species, as well as H5Br also contain this triangular H+ 3 cation [227], as does the metastable P1 LiF4H4 [229].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With two extra electrons, the trihydride anion, H− 3 , prefers a linear arrangement involving a three-center four-electron bond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Quantum chemical calculations have shown that the ground state geometry of the isolated tri- hydride anion possesses one H-H bond that is substantially longer than the other (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='84 vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='75 ˚A), with the transition state between the H-H· · ·H and H· · ·H-H configurations corresponding to the symmetric case [230].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Neverthe- less, certain predicted high-pressure hydrides of the heavy alkali metals K [66], Rb [231], and Cs [232] (as well as the alkaline earth metal Ba [233]) feature an H− 3 anion symmetrized via pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Synthesized NaH7 is thought to contain an asymmetric linear H− 3 motif [234].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Linear H− 3 units are also predicted to appear in various indium [235] and lithium polyhydrides [236].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Scandium polyhydrides, meanwhile, are predicted to feature five-membered rings of hydrogen atoms in various arrangements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In I41/md ScH9, which lies on the convex hull around 300 GPa [237], strips of edge-sharing H5 pentagons are stacked perpendicular to one another along the c axis, linked by ver- tices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The strips are separated by additional H atoms in molecular H2 units.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Around 250 GPa, ScH10 adopts a Cmcm structure in which H5 pentagons are grouped into sets of three, sharing edges and a single common vertex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This phase is nearly isoenthalpic with another ScH10 structure with the same (H5)3 “pentagraphenelike” clusters but arranged in P63/mmc symmetry [238].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ELF demonstrates bonding within the H5 units [237, 238].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This same penta- graphenelike structure is expected to lie on the Lu-H convex hull at 300 GPa and to be very near the Hf-H and Zr-H hulls [238].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For higher hydrogen con- tents yet, ScH12 is predicted to be built of stacked strips of edge-sharing H5 pentagons spaced by Sc [237].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Indeed, the first, most simple, element does not like to be outdone!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One more class of high-pressure hydrogen rich materials whose prediction sparked tremendous experimental synthetic efforts are the so-called “metal super- hydrides”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The reason why scientists have pursued them in earnest is the prediction, verified by recent experiments, of conventional superconductivity at temperatures approaching those experienced in a cold room (the Tc of LaH10 is about 10 ◦F – January in Siberia), or a crisp fall day (the Tc of C-S-H is about 60 ◦F), albeit still at very high pressures!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' CSP investigations into the high-pressure Ca-H system revealed a curious Im¯3m CaH6 phase (Figure 8a) stable above 150 GPa in which the Ca atoms were arranged in a bcc lattice and the H atoms condensed into a sodalite-like H24 framework [73].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' All H-H distances were equivalent, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='24 ˚A at 150 GPa, and ELF analysis confirmed weak covalent bonding between the H atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This phase, a good metal, was predicted to exhibit large electron-phonon coupling, and indeed first principles Springer Nature 2021 LATEX template Article Title 21 calculations estimated a superconducting transition temperature Tc of 220- 235 K at 150 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Subsequent synthetic exploration led to measurement of Tc over 200 K at 160-170 GPa for CaH6 [71, 72].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 8 Clathrate-like metal hydrides predicted then synthesized at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Their remarkable superconducting properties are tied to strong electron-phonon coupling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sev- eral metal-hydrogen stoichiometries adopt these so-called “superhydride” motifs, including Im¯3m MH6, exemplified by CaH6 [71–73], (b) P63/mmc MH9, exemplified by YH9 [35, 69], and (c) Fm¯3m MH10, exemplified by LaH10 [34, 35, 37, 70].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The computational discovery of CaH6 was shortly followed up by fruit- ful theoretical investigations into related metal-hydrogen systems, turning up isostructural phases for Mg [239], Sc [237, 240, 241], Y [34, 35, 242], Pu [243], Tb [244], Eu [245], and Pm-Lu [246].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Structures that are distortions of this high symmetry phase have been predicted as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This includes a tetragonally- distorted I4/mmm ZrH6 variant [247], along with an R¯3m phase in which opposite hexagonal faces of the H24 cubic sodalite framework are opened for SrH6 [248–250] and LaH6 [34, 35], although the latter may not lie on the convex hull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Imm2 BaH6 [233], which contains some of the H− 3 trihydride anions explored above, can be thought of as a highly fragmented version of the sodalite framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The role of distortions of the high-symmetry Im¯3m structure adopted by CaH6 tends to reduce the density of states at the Fermi level, EF , thereby also lowering Tc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The origin of such distortions has recently been investigated using the lens of DFT-Chemical Pressure [251].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Higher hydrogen content allows for other clathrate-like arrangements of hydrogen, from P63/mmc YH9 [35] (Figure 8b) to Fm¯3m LaH10 [34, 35] (Figure 8c).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Distorted versions of these two structures have also been predicted, including C2/m CaH9 [252] and P1 Eu4H36 [245], as well as R¯3m CaH10 [252] and AcH10 [253].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the case of LaH10, quantum anharmonic effects were found to be key in stabilizing the Fm¯3m structure over less symmetric variants [254].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Other more complex clathrate-like hydrogenic frameworks have been predicted as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One example is a diamond-like lattice of Mg@H28 clusters interca- lated with Li@H18 units, which comprise the Fd¯3m Li2MgH16 phase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This compound is an example of a “hot” superconductor whose estimated Tc, 351 K at 300 GPa, is well above room temperature [255].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Such materials are a) CaH6 (Im3m) b)YH9 (P63/mmc) C) LaH10 (Fm3m)Springer Nature 2021 LATEX template 22 Article Title under intense speculation and investigation for their promise towards achieving room-temperature superconductivity, as described in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='3 below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4 Superconductivity The 1911 discovery of a phenomenon in which a substance’s resistivity can plummet to zero [256] sparked countless investigations and resulted in a Nobel prize for Heike Kamerlingh Onnes, as well as a number of future Nobel prizes (directly or indirectly).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The mechanism of superconductivity, and the search for new superconducting materials, has fascinated scientists for over a cen- tury.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A key parameter for superconductors is the critical temperature, Tc, below which a material becomes superconducting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For a number of illustra- tive superconducting materials, Tc is plotted against the pressure at which they are stable in Figure 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Of course, practical applications of superconduc- tivity are limited if temperatures very near 0 K are required, and for some decades the highest known Tc values lingered in the low twenties [257, 258].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This sparked debate regarding a possible natural “cap” on superconductivity around these temperatures [259].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' However, a family of cuprates whose super- conducting mechanism has yet to be explained were the first to break the liquid nitrogen barrier [260], achieving Tcs over 160 K in the case of pressurized HgBa2Cam−1CumO2m+2+δ [261].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The 2001 discovery of superconductivity up to 39 K in MgB2 [263] pro- vided experimental evidence that non-cuprate phases could be promising superconductors, despite the fact that they deviated from the collection of empirical rules enumerated by B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Matthias in the 1950s and 1960s [272].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These ranged from favorable valence electron concentrations to a general distrust for theorists.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' MgB2 belongs, with early and long-term record hold- ers Nb3Sn and Nb3Ge, as well as the clathrate-like hydrides discussed in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5, to the family of “conventional” superconductors whose mecha- nism is thought to follow the theory propounded by Bardeen, Cooper, and Schrieffer in 1957 [273, 274].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This discovery revitalized interest in conventional superconductors [275], leading researchers to wonder what the trajectory of superconductivity research would have looked like had the 1957 measurements of the heat capacity of MgB2 [276] captured the discontinuity that appears upon the superconducting transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Within BCS theory, two electrons of opposite momentum and spin that are within ±ℏωcut (an energy in line with the phonon energies) of the Fermi surface may, at long distances, overcome Coulombic repulsion and experience a net attractive potential when the lattice is polarized through phonon vibrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This forms a Cooper pair, a composite boson of weakly interacting species.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Thermal energy can break the Cooper pairs, with Tc describing the temper- ature at which this occurs and the superconducting state is destroyed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' From Springer Nature 2021 LATEX template Article Title 23 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9 Critical superconducting temperature, Tc, plotted against pressure for selected mate- rials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Data points based on experimental measurements of Tc are plotted with filled circles, while those estimated via theoretical calculations are plotted with empty circles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Elemen- tal Tcs are given in brown, intermetallics in green, cuprates (belonging to the non-BCS unconventional family) in orange, and hydrides in blue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The boiling point of liquid nitro- gen and room temperature are provided to guide the eye, with a star marking the “holy grail” of room-temperature superconductivity at ambient pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Data was collected from references [34, 35, 37, 70, 255, 256, 258, 260–271].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' this construction, the Tc for a material can be estimated by kBTc = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='14ℏω exp � −1 NF V � (1) where ω is the average phonon energy, NF is the single spin electronic den- sity of states at EF , and V is the pairing potential between two electrons resulting from the electron-phonon interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This suggests that high Tc is correlated with a large NF (a feature potentially tunable via judicious doping), strong coupling between electrons and phonons, and high phonon frequencies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A frequently used semiempirical formula to estimate the Tc of a conventional superconductor is the Allen-Dynes modified McMillan equation [277–279]: T c = ωln 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2 exp � − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='04(1 + λ) λ − µ∗(1 + 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='62λ) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' (2) 500 Li2MgH16 o Computed only 450 Experimentally observed 350 Elements Cuprates Intermetallics Hydrides oYH10 300 C-S-H (K) Room temperature YH9 250 LaHi0 200 H3S O LaBeH: HgBaCaCuO 150 OKB2Hs OLaBHs 100 YBaCuO PHn Liquid nitrogen CeH9 50 MgB2 Nb3Ge Ca 1Hg Nb 0 0 50 100 150 200 250 300 Pressure (GPa)Springer Nature 2021 LATEX template 24 Article Title where λ is the electron-phonon coupling constant,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ωln is the logarithmic aver- age phonon frequency and µ∗ is the Coulomb repulsion parameter,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' which is typically treated semiempirically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' An approximate – and illustrative – formula to estimate λ was proposed by Hopfield as [280] λ = NF ⟨I2⟩ M⟨ω2⟩ (3) where ⟨I2⟩ are the electron-phonon matrix elements averaged over the Fermi surface, M is the atomic mass, and ⟨ω2⟩ the mean phonon frequency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' To increase λ, then, NF and the electron-phonon matrix elements should be increased.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Converse to expectations from Equation 1 and Equation 2, where the average phonon energy was directly proportional to Tc, here an increase in ⟨ω2⟩ serves to decrease λ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the denominator of Equation 3, ⟨ω2⟩ and M will naturally tend to counteract one another, as increases in atomic mass lead to softer phonon frequencies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, evidence suggests that Tc frequently increases at the edge of dynamic instability, where soft phonons promote strong coupling [281].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the following sections, we describe the effect of pressure on the propensity for superconductivity of the elements, hydrogen in particular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' We end by dis- cussing families of hydrogen-rich phases that are extremely promising towards achieving the once-distant goal of room-temperature superconductivity (albeit potentially only at very high pressures).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1 The elements Most of the elements in the periodic table can be superconducting given the right conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This includes, so far without exception, rather low temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' However, external pressure can affect the Tc, or even induce superconductivity in some elements [282–285].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, calcium at ambient pressure is not a superconductor and achieves Tc = 29 K at 216 GPa [264].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In comparison, the highest elemental Tc at ambient pressure is 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2 K for niobium [262], while only lithium among the alkali metals [286], and only beryllium [287] among the alkaline earth metals are known to superconduct at ambient pressure, both well under 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Of the fifty-four known supercon- ducting elements of the periodic table, only thirty-one are superconductors at ambient pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Early indications regarding the ability of pressure to either enhance or suppress superconductivity [288–290] were less than encouraging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For simple metals whose electronic structure aligns with a mostly free-electron model, such as Zn, Cd, Hg, and the group 13 metals, applied pressure serves to sup- press what superconductivity is present at ambient pressure [289, 290].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In such free-electron-like metals, the effect of pressure is to broaden electronic bands and increase phonon frequencies due to the stiffer lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Band broadening reduces the electronic density of states at EF , while a stiffer lattice is correlated Springer Nature 2021 LATEX template Article Title 25 with weaker coupling between electrons and phonons, both effects being detri- mental to the superconductivity of a system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At ambient pressure, the alkali metals behave as free-electron metals, but as we have seen in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2, with pressure their electronic structure rapidly diverges from these expectations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' For Li, Tc is highly pressure-dependent and reliant on complex crystal chemistry [291–293].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Like all of the alkali metals, it adopts a bcc crystal structure at ambient pressure, which in short order transitions to the fcc struc- ture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Up to 8 GPa, the increase in pressure is reflected in an increase in phonon frequencies, typical for a stiffening lattice, but with further pressure the phonons become softer and just above 30-40 GPa, imaginary modes related to a structural instability appear.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This pressure corresponds to another phase transition, this time to the hR1 (R¯3m) structure, and shortly thereafter to the cI16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Maxima in Tc are associated with the onset of dynamical instability, as the very soft phonon motions strongly bolster the electron-phonon cou- pling [281, 294].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' From its mostly spherical character at 0 GPa, where lithium adopts a bcc crystal structure, the pressure-induced 2s → 2p electronic tran- sition drives an increasingly anisotropic Fermi surface featuring hot spots of especially strong coupling [295], losing free-electron-like behavior and lead- ing to Fermi surface nesting (FSN) [296–298].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phonon softening, in particular along the Γ → K path, is accompanied by enhancement of the electron-phonon coupling [299, 300], with the result that Tc grows from practically zero to a maximum of ∼20 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This value is among the higher elemental Tcs, as a result of the pressure-induced electronic transitions in lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Following the struc- tural transition to the cI16 phase, lithium remains superconducting but Tc decreases due to a reduction of the FSN and concomitant smaller electron- phonon coupling [299, 301, 302].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At high enough pressures, lithium undergoes a metal-semiconductor transition – and eventually goes back to being a (poor) metal upon transitioning to the oC24 (Cmca) phase [303], but one nonetheless predicted to be superconducting with an estimated Tc of 14 K at 200 GPa [304].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In cesium [305, 306] and rubidium [110, 307], the onset of superconductiv- ity with pressure is associated with the oC16 (Cs-V and Rb-IV) structures, alongside the ns → (n − 1)d transition [307].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' An increase in d-character in the electronic states at EF is correlated with higher Tc in the transition met- als [308, 309] and applies here as well to the heavier alkali metals – which, as we have seen, behave under pressure as transition metals themselves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2 Hydrogen Vitaly Ginzburg, awarded the Nobel Prize in Physics in 2003 for his work in superconductivity and superfluidity, formulated a list of, in his view, the thirty most pressing problems for physics in the 21st century [310, 311].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Following controlled nuclear fusion, the second and third items on this list were high- temperature (room-temperature) superconductivity and metallic hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' These problems are not unrelated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In 1926, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Bernal proposed that under sufficient pressure hydrogen would transition to a metallic state, but it took nearly a decade for pen to Springer Nature 2021 LATEX template 26 Article Title be put to paper by Wigner and Huntington [312].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Their 1935 suggestion that hydrogen could be metallized by 25 GPa, estimated using a series of assump- tions regarding crystal structure and compressibility, proved to be an immense underestimate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In 1968 Neil Ashcroft explicitly linked the quest for metal- lic hydrogen with the quest for superconductors with higher Tcs, with the suggestion that metallic hydrogen itself would be quite a fantastic supercon- ductor [313].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Hydrogen, being the lightest element, can possess the highest frequencies (as a diatomic molecule) and experience a large electron-phonon coupling due to the lack of screening by core electrons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Moreover, in the metal- lic state its DOS at EF is thought to be quite high, making for a very attractive material.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The pressure required to metallize hydrogen, however, is in the multi- megabar range.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Claims of metallic or semimetallic hydrogen have been made for DAC experiments at very high pressures and low temperatures, toeing the line of the practical limits of these techniques [314–316].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Complicating the picture, different experiments used different measures to characterize hydro- gen’s transition to metallicity, from vibrational spectroscopic techniques such as IR [316], optical measurements such as reflectance and opacity [314], and resistivity measurements [315], as well as different scales to calibrate pres- sure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At times, this led to seemingly contradictory results, prompting questions regarding experimental accuracy and reproducibility [317–319].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ab initio cal- culations taking into account the quantum fluctuations of the hydrogen nuclei, however, can reconcile some of these differences, finding in the C2/c-24 high- pressure phase closing of the electronic gap (and transition to metallicity) before the closing of the optical gap (and transition from transparency to reflectivity) [320].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Additionally, the impractically low Tcs of ambient pressure materials are then traded for a much higher Tc in pure metallic hydrogen, but at imprac- tically high pressures!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, ab initio modeling suggests progressive jumps in Tc with a transition from the molecular (estimated Tc = 356 K near 500 GPa) to the atomic phase (Tc increasing to 481 K ca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 700 GPa), and with an atomic-atomic phase transition at ∼1-1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5 TPa driving up λ and resulting in an immense estimated Tc= 764 K [321].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' To address the hydrogen metallization problem, Ashcroft proposed another strategy – instead of pure hydrogen, hydrogen-rich metallic alloys could be targeted as putative super- conductors [322].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The presence of additional atoms in the hydrogen matrix would confer a chemical precompression, thereby lowering the external pres- sure required to reach the metallic state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Under ambient conditions, the crystal structures adopted by metal hydrides are largely subject to the dictates of balanced oxidation states, hence alkali metal hydrides assume the rock salt structure, and hydrides of +2 metals favor fluorite or Co2Si structures, while trivalent metal hydrides tend towards the BiF3 structure, and so on [323, 324].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Much higher hydrogen content would be needed for such hydrogen-rich alloys, as suggested by Ashcroft, and furthermore many of the resulting structures might differ greatly from anything observed at ambient conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Defying the Springer Nature 2021 LATEX template Article Title 27 recommendations of Matthias, the simultaneous and serendipitous advances in CSP methods meant that theoreticians were well poised to answer this call!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='3 Clathrate-like hydrides They were successful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' As described above, calculations on the high-pressure Ca-H system located the CaH6 phase described in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5 [73], in which Ca atoms are embedded into a hydrogenic sodalite-like clathrate framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The strong electron-phonon coupling predicted for Im¯3m CaH6 can be traced to breathing and rocking modes of the square H4 units of the sodalite frame- work [73].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The molecular orbital diagram for such an H4 square has, above a filled bonding state, a half-occupied degenerate non-bonding state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Assum- ing full ionization (integrated charges within atomic basins according to the Quantum Theory of Atoms in Molecules indicate roughly 1 electron per Ca is transferred to the hydrogen network), these orbitals accept the roughly 1/3 electron per H transferred from the Ca atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This favors symmetry-breaking Jahn-Teller distortions – key contributions to the electron-phonon coupling – that lift the degeneracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Similar clathrate-like hydrides, as outlined in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5, were rapidly predicted in several other systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Im¯3m YH6 has been synthesized with a measured Tc of 224 K at 160 GPa [325].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Hydrides with even higher hydrogen content have been synthesized as well, including a P63/mmc YH9 phase with a Tc of 262 K at ca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 180 GPa [69] (a subsequent study reported a slightly lower Tc of 243 K at 200 GPa [326]), and an isotypic CeH9 phase whose Tc has only been predicted (57 K at 88 GPa and up to 100 K at 130 GPa [35], or 105- 117 K by 200 K [267]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The relatively low pressure required to stabilize CeH9 has been ascribed to strong chemical precompression from the delocalized Ce 4f electrons [327].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The reported Tcs for Fm¯3m LaH10 (250 and 260 K at 170 and 185 GPa, respectively [37, 70]), are in line with theoretical predictions of 257-274 K at 250 GPa [34].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Isotypic Fm¯3m YH10 is computed to be a room temperature superconductor with a Tc of 305-327 K at 250 GPa [34].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' However, YH10 has thus far eluded synthetic efforts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Partial doping with lanthanum appears to be one strategy to stabilize YH10: a series of ternary (La/Y)H10 phases have been experimentally observed, with measured Tc = 253 K [38].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Although the Tcs of many clathrate-like hydrides are stunning, these phases will surely decompose well above atmospheric pressures!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' CeH9 is remarkable for the comparatively low, sub-megabar, pressures at which it maintains dynamic stability [267].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In an attempt to preserve the loosely- bound hydrogenic clathrate frameworks that are associated with such strong electron-phonon coupling to lower pressures, one promising strategy involves the addition of a third element in an attempt to further chemically precom- press the hydrogenic lattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Fm¯3m LaBH8 phase, which is predicted to maintain dynamic stability down to 40 GPa [269], can be derived from LaH10 by removing two hydrogen atoms per formula unit and placing boron atoms into the center of H8 cubes that are empty in LaH10 [271, 328].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' LaBH8 has an estimated Tc of 126 K at 50 GPa [269] – and the isostructural LaBeH8 phase Springer Nature 2021 LATEX template 28 Article Title is predicted to achieve a Tc of 183 K at 20 GPa [271].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Other XYH8 phases have been proposed, with a variety of possible elemental combinations ripe for tuning stability and properties [271].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A second possibility is afforded by the XY2H8 phases that can be constructed by leaving the H8 cubes empty, but stuffing the center of H4 tetrahedra instead.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' KB2H8 (dynamically stable to 12 GPa [270]) and LaC2H8 (dynamically stable down to 70 GPa [329]) are two representatives of this structural arrangement, with estimated Tcs of 134-146 and 69 K, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Key to the success of the clathrate-like hydrides is the maintenance of loosely-coordinated networks of hydrogen, rather than condensation into H2 molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The effect of H-H interatomic distances on superconductivity can be seen in the MH4 hydrides, which adopt the I4/mmm structure shared with ThCr2Si2 [330, 331].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Hydrogen occupies two inequivalent sites in the ThCr2Si2 structure – the apical Ha (Wyckoff position 4e, Si) and basal Hb (4d, Cr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A plethora of metal hydrides have been predicted or synthesized in this structure type under pressure, including Ca [73, 252, 332] and Sr [248, 249], Sc [237, 240, 241], Y [34, 242], and Zr [247] and rare earths La [34], Ce [35, 267, 333], Pr [35, 334, 335], Pu [243], Tb [244], Eu [245], Nd [35, 336], and Th [337, 338], making systematic study enticing and useful [331].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With metal oxidation states ranging from +2 to +4, the formulas of these compounds can be written as Mx+(H− b )2(Ha)(x−2)− 2 , with hydridic Hb and a range of charges possible on the Ha atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Tcs of these phases are correlated with the length of the Ha- Ha contacts, which can behave anywhere from covalently bound H2 units to fully dissociated hydridic anions depending on the metal atom – similar to the behavior of the X-X bond in ThCr2Si2-type AB2X2 phases [339].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The size of the metal atom can be relevant, as larger atoms will stretch the Ha-Ha contact through purely steric interactions, but more important is the valency of the metal atom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Electron transfer from the electropositive metal into the Ha-Ha motif directly populates the H2 σ∗ u antibonding orbitals, but is also driven by a Kubas-like two-pronged mechanism of H2 σg → M d donation, and M d → H2 σ∗ u back-donation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' With enough H2 σ∗ u population, the Ha atoms behave in a hydridic fashion lowering the Tc, as seen in ZrH4 [247] and ThH4 [337].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Donation of sufficient electron density to weaken, but not fully break, the Ha- Ha bonding interaction results in a much higher DOS at EF and enhanced Tc, as in YH4 [242].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='4 Covalent hydrides The first hydride to top the charts, as it were, was not of the metal clathrate- like family, but instead came from attempts to metallize H2S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Theory identified an H2S compound that was computed to possess a Tc of 80 K at 160 GPa [340].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Experimental confirmation followed shortly thereafter, finding a phase with Tc < 100 K, but in the process a higher-temperature preparation method yielded a sample with a Tc of 203 K at 150 GPa [266].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A few years before, synthetic exploration into the (H2S)2H2 stoichiometry found a phase stabi- lized by pressure-induced hydrogen bonding above 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='5 GPa [341].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This inspired Springer Nature 2021 LATEX template Article Title 29 CSP investigations of the H3S stoichiometry, which found an R3m phase with Tc = 155-166 K at 130 GPa [342].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' By 180 GPa this structure transitioned to one with Im¯3m symmetry (Figure 10a) for which the estimated Tc was 191- 204 K at 200 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Serendipitously the experimental [266] and theoretical [342] manuscripts appeared at nearly the same time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Subsequent XRD studies sup- ported the identification of the 203 K superconductor as Im¯3m H3S [343], though other structures have been proposed [344–349].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A host of studies on the H3S superconductor have followed, exploring the isotope effect, role of anharmonicity, and possible quantum effects [191, 350– 359].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The inclusion of quantum nuclear motions lowered the pressure where the less symmetric R3m phase was predicted to transition to the Im¯3m structure with symmetric H-S bonds into the range of pressures where high Tcs had been measured [191].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One of the most striking features of the electronic structure of H3S are a pair of van Hove singularities bracketing EF [360, 361].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Shifting the position of EF , potentially by doping, could increase the number of states that can participate in the electron-phonon coupling mechanism, and therefore increase the Tc of the system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Doping is a common strategy used for precise tuning of EF , and com- putations using the virtual crystal approximation (VCA) suggested that the addition of a little bit of phosphorus, carbon, or silicon could raise the Tc into the room-temperature regime > 280 K [362–364].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the VCA, alchem- ical pseudoatoms are constructed from weighted averages of the component atom potentials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The resulting chemical chimeras, however, cannot accurately model the local structural and electronic effects that arise when one atom is replaced with an entirely different element.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This throws off, in particular, the very dynamical response properties that one must calculate carefully to obtain reasonable estimates of Tc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Additional studies based on actual doped H3S models constructed as supercells have sought to explore the local effects of doping [365–368], although the calculation of dynamical properties of the requisite large unit cells can be prohibitively expensive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' One particularly promising system involved the addition of carbon to the H3S lattice by way of methane intercalation [369, 372, 373].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Stoichiometries that are a linear combination of CH4 and H3S (the most simple of which is CSH7) proved especially interesting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' They yielded a variety of dynamically stable (although energetically metastable) structures, which differed in the ori- entation of the methane molecules encapsulated in the H3S lattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Some of the highest Tcs predicted for these phases were 181 K at 100 GPa for I¯43m [372], and 181-194 K for the R3m symmetry structure [369] shown in Figure 10b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Independently, photochemical synthesis in the C-S-H system yielded the first report of room-temperature superconductivity, achieving a Tc of 288 K at 267 GPa [268].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This report has inspired a slew of follow-up work and much debate [371, 374, 375].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' XRD analysis performed at pressures below the pur- ported room-temperature superconducting transition [370, 376] is consistent with the Al2Cu geometry (as well as an orthorhombic Pnma variant) associ- ated with CH4-H2 [377] and H2S-H2 [341] (Figure 10c).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This may suggest an Springer Nature 2021 LATEX template 30 Article Title Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 10 The covalent hydride Im¯3m H3S [266] (a) represents a breakthrough in high- Tc conventional superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Peaks in the electronic DOS near EF have prompted numerous investigations on doped versions of H3S, discovering phases such as the methane- intercalated (CH4)H3S = CSH7 [369] (b), among many others.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A recent synthesis of a room-temperature superconductor consisting of carbon, sulfur, and hydrogen generated even more momentum, with diffraction analysis performed at pressures below those where the room-temperature superconducting transition were observed, revealing the presence of a phase based on the I4/mcm Al2Cu-like structure adopted by the van der Waals (H2S)H2 phase [341, 370, 371] (c).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' overall stoichiometry of [(CH4)2H2]x[(H2S)2H2]y for the room-temperature C- S-H superconductor, although subsequent phase transitions at higher pressure to the high-Tc superconducting phase cannot be ruled out.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In fact, additional studies indicate just such a structural transition occurs to form the room- temperature superconducting phase, with indications of methane signatures in the Raman spectra [371].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' As was the case for the binary H3S system, it appears that a panoply of metastable phases may be accessible by slight vari- ations on synthetic procedure, in particular on carbon content [378], offering plenty of space for further experimental and theoretical discoveries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In addition to sulfur-based covalent hydrides, phosphorus hydrides have sparked interest after compression of a phosphine (PH3) sample yielded a material that became superconducting at 30 K at 83 GPa, increasing to 103 K at 207 GPa [265].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The structure and composition of the responsible phase or phases was unclear, prompting an array of follow-up studies levying CSP tech- niques to identify plausible compounds [379–383].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pressure was found to drive the decomposition of phosphine into a variety of products with stoichiome- tries including PH, PH2, PH3, and more.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A predicted C2/m PH3 phase [380] featuring P-P bonds (in contrast to the H3S superconductor, which has no S-S bonding) was estimated to be superconducting below 83 K at 200 GPa, in line with the experimental values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Another study suggested that multiple metastable decomposition products of phosphine, including those with PH and PH2 stoichiometries, might in combination be responsible for the observed superconductivity [382].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' PH2 phases with C2/m and I4/mmm symmetries, differing by a tilt in the component H-P-H moieties, had estimated Tcs of 76 and 70 K, respectively [381].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Later, another set of PH2 phases were pro- posed consisting of simple cubic layers of phosphorus capped with hydrogen a) H3S (Im3m) b)(CH4)H3S(R3m C)(H2S)2H2(I4/mcm)Springer Nature 2021 LATEX template Article Title 31 atoms and further intercalated with H2 molecules acting as Coulombic spac- ing agents [383].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' At 80 GPa, these structures had estimated Tcs ca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 30 K, similar to the values that were measured.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Raman spectroscopic measurements provided evidence for phosphine dimerization coupled with dehydrogenation under pressure, yielding compositions such as P2H4 and P3H6 [384, 385].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In these phases, low temperatures were required to maintain stability at multi- megabar pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Tc of a predicted C2/m P4H6 phase was estimated to be 67 K at 200 GPa [385].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The plethora of metastable P-H compounds under pressure has prompted computational investigations into ternary systems containing phosphorus and hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Above 250 GPa, an R¯3 LiP2H14 phase, consisting of P@H9 clusters spaced by Li atoms as well as isolated H atoms, achieves an estimated Tc of 169 K at 230 GPa (where it is metastable) [386].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pm¯3 LiPH6, a colored variant of the A15 crystal structure adopted by intermetallic superconductors Nb3Ge [258] and Nb3Sn [257], has an estimated Tc of 150-167 K at 200 GPa (where it is metastable) [387].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In the S-P-H system, obviously tantalizing for its connection to the H3S superconductor as well as to phosphine derivatives, relatively low Tcs were predicted for phases on the high-pressure convex hulls, but low-lying metastable structures based on phosphorus substitution into the Im¯3m H3S lattice were promising, including Im¯3m S7PH24, which had an estimated Tc of 183 K at 200 GPa [388].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5 Conclusion Although the entirety of the lived human experience resides within a vastly narrow pressure range, the universe is not so simple.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The chemistry we know at 1 atmosphere is not the chemistry of Jupiter, Saturn, or even the center of our own planet Earth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Starting from the periodic table itself, the ramifications of pressure are rapidly found to alter elemental behavior and, consequently, how the elements interact with one another to form new and bizarre phases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Potassium, in its guise as a “transition metal”, enjoys all manner of new chem- ical interactions – in compound formation and in the wildly complex electride elemental structures it adopts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Cesium can become anionic, and helium takes an active role in stabilizing a network of sodium and interstitial quasiatoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Strange geometrical and bonding motifs from clusters to networks abound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Yet not only are the structures of phases – electronic and crystalline – molded by high pressure, but high-pressure studies have revolutionized the search for high-temperature superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Pressure induces superconduc- tivity in a plethora of elements, and drives the formation of phases containing structural motifs whose atomic vibrations can be strongly coupled to the underlying electronic structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' From the clathrate-like LaH10 to the covalent H3S – and the intensely-discussed CSH room-temperature superconductor – the playing field of high-pressure materials is a promising one for the future.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Springer Nature 2021 LATEX template 32 Article Title Acknowledgments We acknowledge the NSF (DMR-1827815) for financial support.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' This mate- rial is based upon work supported by the U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Department of Energy, Office of Science, Fusion Energy Sciences under Award Number DE-SC0020340 to E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' thanks the US Department of Energy, National Nuclear Security Administration, through the Chicago-DOE Alliance Center under Coopera- tive Agreement Grant No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' DE-NA0003975 for financial support.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' We thank Giacomo Scilla for his help in editing and preparing the manuscript.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' References [1] Guillot, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The Interiors of Giant Planets: Models and Outstanding Questions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Annu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Earth Planet Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 33, 493–530 (2005) [2] Zaghoo, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Silvera, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Conductivity and dissociation in liquid metallic hydrogen and implications for planetary interiors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' USA 114(45), 11873–11877 (2017) [3] Matsuoka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Direct observation of a pressure-induced metal-to-semiconductor transition in lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 458(7235), 186– 189 (2009) [4] Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Trojan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Medvedev, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lyakhov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Valle, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Transparent dense sodium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 458(7235), 182–185 (2009) [5] Polsin, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lazicki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gong, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burns, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Coppari, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hansen, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Henderson, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huff, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Millot, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Paul, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eggert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Collins, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rygg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structural com- plexity in ramp-compressed sodium to 480 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 13, 2354 (2022) [6] Bukowinski, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : The Effect of Pressure on the Physics and Chem- istry of Potassium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Geophys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3, 491–494 (1976) [7] Atou, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hasegawa, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Parker, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Badding, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Unusual Chemical Behavior for Potassium under Pressure: Potassium-Silver Compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 118, 12104–12108 (1996) [8] Parker, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Atou, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Badding, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Transition Element-Like Chem- istry for Potassium Under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 273, 95–97 (1996) [9] Zhang, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boulfelfel, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lyakhov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stavrou, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Konˆopkov´a, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unexpected stable stoichiometries of sodium chlorides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 342(6165), 1502–1505 (2013) Springer Nature 2021 LATEX template Article Title 33 [10] Dong, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stavrou, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lobanov, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Saleh, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Qian, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gatti, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Deringer, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dron- skowski, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Konˆopkov´a, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Popov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boldyrev, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': A stable compound of helium and sodium at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 440–445 (2017) [11] Geballe, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mishra, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ahart, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Baldini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Synthesis and stability of lanthanum superhydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 57(3), 688 (2018) [12] Prewitt, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Downs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure crystal chemistry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mineral 37, 283–318 (1998) [13] Grochala, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Feng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : The chemical imagination at work in very tight places.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 46(20), 3620–3642 (2007) [14] Zhang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lv, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Materials discovery at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 2, 17005 (2017) [15] Jayaraman, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Diamond anvil cell and high-pressure physical investiga- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 55, 65–108 (1983) [16] Bassett, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Diamond anvil cell, 50th birthday.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' High Press.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 29, 163–186 (2009) [17] Li, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ji, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xu, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cai, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Diamond anvil cell behavior up to 4 Mbar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' USA 115, 1713–1717 (2018) [18] Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinskaia, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Abakumov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Implementation of micro-ball nanodiamond anvils for high- pressure studies above 6 Mbar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Comm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3, 1163 (2012) [19] Dubrovinskaia, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Solopova, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Abakumov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Turner, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykova, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prescher, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Petitgirard, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chuvashova, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gasharova, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mathis, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ershov, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Snigireva, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Snigirev, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Terapascal static pressure generation with ultrahigh yield strength nanodiamond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 2, 1600341 (2016) [20] Dewaele, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Loubeyre, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Occelli, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marie, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mezouar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Toroidal diamond anvil cell for detailed measurements under extreme static pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Comm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 2913 (2018) [21] Jenei, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', O’Bannon, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Weir, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cynn, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lipp, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Evans, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Single crystal toroidal diamond anvils for high pressure experiments Springer Nature 2021 LATEX template 34 Article Title beyond 5 megabar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Comm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 3563 (2018) [22] Duffy, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Ultra-High Pressure Dynamic Compression of Geological Materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Front.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Earth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 7, 1–20 (2019) [23] Hansen, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fratanduono, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hicks, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Suer, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sprowal, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huff, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gong, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Henderson, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Polsin, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zaghoo, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Collins, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rygg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Melting of magnesium oxide up to two terapascals using double-shock compression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 104, 014106 (2021) [24] Fratanduono, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Millot, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Braun, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ali, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fernandez- Pa˜nella, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Seagle, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Davis, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Brown, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akahama, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kraus, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marshall, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', O’Bannon, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McNaney, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eggert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Establishing gold and platinum standards to 1 terapascal using shockless compression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 372, 1063–1068 (2021) [25] Smith, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eggert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jeanloz, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duffy, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Braun, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pat- terson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rudd, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Biener, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lazicki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hamza, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Braun, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Benedict, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Celliers, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Collins, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Ramp compression of diamond to five terapascals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 511, 330–333 (2014) [26] Kraus, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ali, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Belof, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Benedict, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bernier, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Braun, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cohen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Collins, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Coppari, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Desjarlais, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fratanduono, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hamel, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Krygier, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lazicki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mcnaney, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Millot, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Myint, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Newman, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rygg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sterbentz, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stewart, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stixrude, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Swift, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wehrenberg, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eggert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Measur- ing the melting curve of iron at super-Earth core conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 375(6577), 202–205 (2022) [27] Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grochala, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Predicting crystal structures and properties of matter under extreme conditions via quantum mechanics: the pressure is on.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 17, 2917–2934 (2015) [28] Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structures at high pressure from random searching.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Status Solidi B 246(3), 536–540 (2009) [29] Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lyakhov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Valle, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': How Evolutionary Crystal Structure Prediction Works – and Why.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 44(3), 227–237 (2011) [30] Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Perspective: Crystal structure prediction at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 140(4), 040901 (2014) [31] Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gatti, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Glass, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kurakevych, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Solozhenko, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Ionic high-pressure form of elemental boron.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 457, 863–867 (2009) Springer Nature 2021 LATEX template Article Title 35 [32] Zhang, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superhard BC3 in Cubic Diamond Structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 114, 015502 (2015) [33] Avery, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oses, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gossett, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Proserpio, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Toher, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Curtarolo, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Predicting superhard materials via a machine learning informed evolutionary structure search.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' npj Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 89 (2019) [34] Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Naumov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Potential high-Tc superconducting lanthanum and yttrium hydrides at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 114, 6990–6995 (2017) [35] Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Hydro- gen clathrate structures in rare earth hydrides at high pressures: possible route to room-temperature superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 119, 107001 (2017) [36] Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lobanov, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Holtgrewe, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jiang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yanilkin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Uranium polyhydrides at moderate pressures: Pre- diction, synthesis, and expected superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4, 9776 (2009) [37] Drozdov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kong, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Minkov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Besedin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kuzovnikov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mozaffari, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Balicas, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Balakirev, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Graf, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Knyazev, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tkacz, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Super- conductivity at 250 K in lanthanum hydride under high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 569(7757), 528–531 (2019) [38] Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ivanova, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sadakov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sobolevskiy, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pervakov, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lyubutin, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Glazyrin, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Giordano, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Karimov, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Vasiliev, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akashi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pudalov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity at 253 K in lanthanum-yttrium ternary hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Today 48, 18–28 (2021) [39] Pickett, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Heil, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pierleoni, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kolmogorov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gross, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Errea, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Arita, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Margine, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gonnelli, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Valenti, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boeri, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hirschfeld, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hennig, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Profeta, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sanna, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The 2021 room-temperature superconductivity roadmap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Condens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 34, 183002 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1088/1361-648x/ ac2864 [40] Dye, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Electrides: Early Examples of Quantum Confinement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 42, 1564–1572 (2009) Springer Nature 2021 LATEX template 36 Article Title [41] Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Alkali Metals in Ethylenediamine: A Computational Study of the Optical Absorption Spectra and NMR Parameters of [M(en)+ 3 ·M −] Ion-Pairs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 133, 4829–4839 (2011) [42] Riedel, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Seel, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Malko, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miller, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sperling, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Choi, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Headen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Porch, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kucernak, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pyper, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Edwards, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Barrett, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superalkali-Alkalide Interactions and Ion Pairing in Low-Polarity Solvents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 143, 3934–3943 (2021) [43] Abella, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Philips, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Autschbach, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Ab initio molecular dynam- ics study of sodium NMR chemical shifts in the methylamine solution of [Na+[2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2]cryptand Na−].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 23, 339–346 (2021) [44] Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-stabilized lithium caesides with cae- sium anions beyond the −1 state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5(1), 1–8 (2014) [45] Pauling, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The Nature of the Chemical Bond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Energy of Single Bonds and the Relative Electronegativity of Atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 54, 3570–3582 (1932) [46] Pearson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Absolute electronegativity and absolute hardness of Lewis acids and bases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 107, 6801–6806 (1985) [47] Mulliken, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : A New Electroaffinity Scale;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Together with Data on Valence States and on Valence Ionization Potentials and Electron Affinities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 2, 782–793 (1934) [48] Dong, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Electroneg- ativity and chemical hardness of elements under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 119, 2117416119 (2022) [49] Allen, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Electronegativity Is the Average One-Electron Energy of the Valence-Shell Electrons in Ground-State Free Atoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 111, 9003–9014 (1989) [50] Rahm, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zeng, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Electronegativity Seen as the Ground-State Average Valence Electron Binding Energy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 141, 342–351 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1021/jacs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='8b10246 [51] Rahm, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Erhart, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cammi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Relating atomic energy, radius, and electronegativity through compression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 12, 2397–2403 (2021) [52] Racioppi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rahm, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': In-Situ Electronegativity and the Bridging of Chemical Bonding Concepts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 27, 18156–18167 (2021) [53] Sternheimer, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': On the Compressibility of Metallic Cesium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Springer Nature 2021 LATEX template Article Title 37 78, 235–243 (1950) [54] Louie, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cohen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Electronic structure of cesium under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 10, 3237–3245 (1974) [55] Takemura, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Minomura, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shimomura, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': X-Ray Diffraction Study of Electronic Transitions in Cesium under High Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 49, 1772–1775 (1982) [56] Connerade, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Semaoune, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Atomic compressibility and reversible insertion of atoms into solids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B: At.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Opt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 33, 3467– 3484 (2000) [57] Rahm, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cammi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Squeezing All Elements in the Periodic Table: Electron Configuration and Electroneg- ativity of the Atoms under Compression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 141, 10253–10271 (2019) [58] Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jepsen, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Andersen, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Muffin-Tin Orbital Wannier- Like Functions for Insulators and Metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ChemPhysChem 6, 1934–1942 (2005) [59] Neaton, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Pairing in dense lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 400, 141–144 (1999) [60] Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Syassen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Christensen, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Novikov, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : New high- pressure phases of lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 408, 174–178 (2000) [61] Naumov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Chemical bonding in hydrogen and lithium under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 143, 064702 (2015) [62] Degtyareva, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Potassium under pressure: Electronic origin of com- plex structures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 36, 62–72 (2014) [63] Takemura, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Christensen, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Novikov, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Syassen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Phase stability of highly compressed cesium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 61, 14399–14404 (2000) [64] McMahan, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Alkali-metal structures above the s-d transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 29, 5982–5985 (1984) [65] McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Syassen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Composite incommensurate K-III and a commensurate form: Study of a high- pressure phase of potassium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 74, 140102 (2006) [66] Hooper, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High pressure potassium polyhydrides: a chemical perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 116(24), 13322–13328 (2012) Springer Nature 2021 LATEX template 38 Article Title [67] Dewaele, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pepin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geneste, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Garbarino, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Reaction between nickel or iron and xenon under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' High Pressure Research 37, 137–146 (2017) [68] Stavrou, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lobanov, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zaug, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Synthesis of Xenon and Iron-Nickel Intermetallic Compounds at Earth’s Core Thermodynamic Conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 120, 096001 (2018) [69] Snider, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dasenbrock-Gammon, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McBride, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mey- ers, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lawler, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Salamat, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Synthesis of Yttrium Superhydride Superconductor with a Transition Temperature up to 262 K by Catalytic Hydrogenation at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 126, 117003 (2021) [70] Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ahart, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mishra, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geballe, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Baldini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Struzhkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Evidence for Superconductivity above 260 K in Lanthanum Superhydride at Megabar Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 122, 027001 (2019) [71] Ma, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-temperature superconducting phase in clathrate calcium hydride CaH6 up to 215 K at a pressure of 172 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 128, 167001 (2022) [72] Li, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jia, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Feng, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Min, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Long, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ye, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chariton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ginsberg, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bass, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yuan, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity above 200 K discovered in superhydrides of calcium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 13, 2863 (2022) [73] Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', John, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tanaka, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Iitaka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconduc- tive sodalite-like clathrate calcium hydride at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' USA 109(17), 6463–6466 (2012) [74] Dye, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Compounds of Alkali Metal Anions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 18, 587–598 (1979) [75] Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Predicted Pressure-Induced s-Band Ferro- magnetism in Alkali Metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 107, 087201 (2011) [76] Hasegawa, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Atou, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Badding, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : High-Pressure Synthesis of an Alkali Metal-Transition metal Laves Phase: KAg2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 130, 311–315 (1997) [77] Lee, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jeanloz, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure alloying of potassium and iron: Radioactivity in the Earth’s core?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Geophys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 30 (2003) Springer Nature 2021 LATEX template Article Title 39 [78] Adeleke, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stavrou, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Adeniyi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wan, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gou, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Two good metals make a semiconductor: A potassium-nickel compound under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 102, 134120 (2020) [79] Brgoch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermus, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-Stabilized Ir3− in a Superconducting Potasium Iridide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 120, 20033–20039 (2016) [80] Lotfi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Brgoch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Predicting pressure-stabilized alkali metal iridides: A-Ir (A=Rb, Cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 158, 124–129 (2019) [81] Yang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bergara, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Gold as a 6p-Element in Dense Lithium Aurides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 138, 4046–4052 (2016) [82] Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hou, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Mercury under Pressure acts as a Transition Metal: Calculated from First Principles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 54, 9280–9283 (2015) [83] Savin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The Electron Localization Function (ELF) and Its Relatives: Interpretations and Difficulties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Struc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Theochem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 727(1-3), 127– 131 (2005) [84] Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Caesium in high oxidation states and as a p-block element.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5(10), 846–852 (2013) [85] Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nikolaev, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burton, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Electrides: a review.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 8, 10551–10567 (2020) [86] Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Edwards, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': A Molecular Perspective on Lithium-Ammonia Solutions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 48, 8198–8232 (2009) [87] Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-Pressure Electrides: The Chemical Nature of Interstitial Quasiatoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 137, 3631–3637 (2015) [88] Neaton, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : On the constitution of sodium at higher densities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 86, 2830–2833 (2001) [89] Li, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystal Structures and Exotic Behav- ior of Magnesium under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 114, 21745–21749 (2010) [90] Dong, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Electrides and Their High-Pressure Chemistry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In: Correlations in Condensed Matter Under Extreme Conditions, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 69–84.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Springer, Cham (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6 [91] Modak, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Verma, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Pressure induced multi-centre bonding and metal-insulator transition in PtAl2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 21, 13337–13346 (2019) Springer Nature 2021 LATEX template 40 Article Title [92] Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High pressure electrides: A predictive chemical and physical theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 47(4), 1311–1317 (2014) [93] Wang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hilleke, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Polsin, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Topolog- ical Electride Phase of Sodium at High Pressures and Temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' arXiv (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='48550/ARXIV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='06251.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' https:// arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='org/abs/2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='06251 [94] Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Metallic Icosahedron Phase of Sodium at Terapascal Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 114, 125501 (2015) [95] Vegas, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grzechnik, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', M¨uhle, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jansen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Antiflu- orite to Ni2In-type phase transition in K2S at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4, 1077–1081 (2002) [96] Santamaria-Perez, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Vegas, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Muehle, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jansen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure experimental study on Rb2S: antifluorite to Ni2In-type phase transitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acta Crystallogr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 67, 109–115 (2011) [97] Marques, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ackland, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lundegaard, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stinton, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Contreras-Garcia, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Potassium under pressure: A Pseudobinary Ionic Compound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 103, 115501 (2009) [98] McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lundegaard, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Loa, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Guillaume, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kleppe, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Amboage, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wilhelm, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jeph- coat, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Structure of sodium above 100 GPa by single-crystal x-ray diffraction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 104, 17297–17299 (2007) [99] Guillaume, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Degtyareva, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Han- fland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Evans, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Guthrie, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sinogeikin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Cold melting and solid structures of dense lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 7, 211–214 (2011) [100] Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Naumov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Quasi- molecules in Compressed Lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 56, 972–975 (2017) [101] Dronskowski, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bl¨ochl, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Crystal Orbital Hamilton Populations (COHP).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Energy-Resolved Visualization of Chemical Bonding in Solids Based on Density-Functional Calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 97, 8617– 8624 (1993) [102] Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Brgoch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Spera, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jackson, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kresse, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Anionic chemistry of noble gases: formation of Mg–NG (NG= Xe, Kr, Ar) compounds under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 137(44), 14122–14128 (2015) Springer Nature 2021 LATEX template Article Title 41 [103] Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Brgoch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hou, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Iodine Anions beyond -1: Formation of LinI (n=2-5) and Its Interaction with Quasiatoms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 55, 9377–9382 (2016) [104] Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Chemistry under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4, 508–527 (2020) [105] Pauling, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Principles determining the structure of high-pressure metals: The structures of cesium(IV) and cesium(V).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' USA 86, 1431–1433 (1989) [106] von Schnering, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nesper, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': How Nature Adapts Chemical Struc- tures to Curved Surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 26, 1059–1080 (1987) [107] Olijnyk, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Holzapfel, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Phase transitions in K and Rb under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A 99, 381–383 (1983) [108] Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure structures of lithium, potassium, and rubidium predicted by an ab initio evolutionary algo- rithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 78, 014102 (2008) [109] Lundegaard, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marques, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stinton, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ackland, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Observation of the oP8 crystal structure in potassium at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 80, 020101 (2009) [110] Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grzechnik, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Syassen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Loa, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Rubidium-IV: A High Pressure Phase with Complex Crystal Structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 83, 4085–4088 (1999) [111] McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rekhi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure Dependent Incom- mensuration in Rb-IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 87, 055501 (2001) [112] McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure structures and phase transformations in elemental metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 35, 943–963 (2006) [113] Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lundegaard, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McMahon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Guillaume, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelmes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mezouar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structural Diversity of Sodium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 320, 1054–1057 (2008) [114] Woolman, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Robinson, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marques, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Loa, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ackland, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structural and electronic properties of the alkali metal incommensurate phases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 2, 053604 (2018) [115] Zhou, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Predicted lithium-iron compounds under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' RSC Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 66721–66728 Springer Nature 2021 LATEX template 42 Article Title (2016) [116] Zhou, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of Host-Guest Na-Fe Intermetallics at High Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 55, 7026–7032 (2016) [117] Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Stable Structures and Superconductivity in a Y-Si System under High Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 12, 10388– 10393 (2021) [118] Pepin, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Porcelli, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Origin of Noble Gases in the Terrestrial Planets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mineral.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Geochem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 47, 191–246 (2002) [119] Shcheka, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Keppler, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The origin of the terrestrial noble-gas signature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 490, 531–534 (2012) [120] Marty, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The origins and concentrations of water, carbon, nitrogen, and noble gases on Earth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Earth & Planet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 313-314, 56–66 (2012) [121] Caldwell, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nguyen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pfrommer, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mauri, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Louie, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jeanloz, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structure, Bonding, and Geochemistry of Xenon at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 277, 930–933 (2012) [122] Lee, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Steinle-Neumann, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure alloying of iron and xenon: “Missing” Xe in the Earth’s core?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Geophys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 111, 02202 (2006) [123] Zhu, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zou, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Reactions of xenon with iron and nickel are predicted in the Earth’s inner core.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6(7), 644–648 (2014) [124] Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Song, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Xenon iron oxides predicted as potential Xe hosts in Earth’s lower mantle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 11(1), 1–7 (2020) [125] Adeleke, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kunz, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stavrou, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': A High-Pressure Compound of Argon and Nickel: Noble Gas in the Earth’s Core?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ACS Earth Space Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3, 2517–2524 (2019) [126] Zhu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jung, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Glass, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gatti, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lyakhov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Stability of xenon oxides at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5(1), 61–65 (2013) [127] Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwerdtfeger, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Xenon suboxides stable under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5(24), 4336–4342 (2014) [128] Dewaele, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Worth, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pascarelli, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Springer Nature 2021 LATEX template Article Title 43 Mathon, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mezouar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Irifune, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Synthesis and stability of xenon oxides Xe2O5 and Xe3O2 under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 8, 784–790 (2016) [129] Zaleski-Ejgierd, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lata, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Krypton oxides under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 18938 (2016) [130] Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Stable xenon nitride at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 92(9), 094104 (2015) [131] Howie, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Turnbull, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Binns, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Frost, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dalladay-Simpson, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Formation of xenon-nitrogen compounds at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 34896 (2016) [132] Bovornratanaraks, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tsuppayakorn-aek, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Luo, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ahuja, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Ground-state structure of semiconducting and superconducting phases in xenon carbides at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 2459 (2019) [133] Kurzyd�lowski, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zaleski-Ejgierd, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Ground-state structure of semi- conducting and superconducting phases in xenon carbides at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 18, 2309–2313 (2016) [134] Kurzyd�lowski, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', So�ltysiak, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', D˙zoleva, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zaleski-Ejgierd, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High- Pressure Reactivity of Kr and F2-Stabilization of Krypton in the +4 Oxidation State.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Crystals 7, 329 (2017) [135] Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unexpected Trend in Stability of Xe-F Compounds under Pressure Driven by Xe-Xe Covalent Bonds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 7, 4562–4567 (2016) [136] Zarifi, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystal Structures and Electronic Properties of Xe-Cl Compounds at High Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 122, 2941–2950 (2018) [137] Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lv, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Stable Lithium Argon compounds under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 16675 (2015) [138] Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unexpected Xe anions in XeLin intermetallic compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' EPL 117, 26002 (2017) [139] Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bi, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wei, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystal Structures and Electronic Properties of Cesium Xenides at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 119, 24996–25002 (2015) [140] Sanloup, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bonev, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hochlaf, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Maynard-Casely, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Reactivity of xenon with ice at planetary conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 110(26), Springer Nature 2021 LATEX template 44 Article Title 265501 (2013) [141] Monserrat, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Martinez-Canales, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Helium- Iron Compounds at Terapascal Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 121, 015301 (2018) [142] Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xing, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Plastic and Superionic Helium Ammonia Compounds under High Pressure and High Temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' X 10, 021007 (2020) [143] Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Klug, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Stable structures of He and H2O at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 91, 014102 (2015) [144] Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xing, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Multiple superionic states in helium-water compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 15, 1065–1070 (2019) [145] Vos, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Finger, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schouten, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : A high-pressure van der Waals compound in solid nitrogen-helium mixtures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 358, 46–48 (1992) [146] Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Feng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Redfern, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lei, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Route to high-energy density polymeric nitrogen t-N via He-N compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 722 (2018) [147] Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of the Xe-He binary phase diagram at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 640, 115–118 (2015) [148] Cazorla, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Errandonea, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sola, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure phases, vibrational properties, and electronic structure of Ne(He)2 and Ar(He)2: A first- principles study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 80, 064105 (2009) [149] Loubeyre, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jean-Louis, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', LeToullec, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Charon-G´erard, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High pressure measurements of the He-Ne binary phase diagram at 296 K: Evidence for the stability of a stoichiometric Ne(He)2 solid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 70, 178–181 (1993) [150] Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Botana, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Valdez, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Reactivity of He with ionic compounds under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Comm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 951 (2018) [151] Gao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of pressure- induced stabilization of noble-gas-atom compounds with alkali oxides and alkali sulfides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3, 015002 (2019) [152] Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lv, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Feng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Redfern, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Springer Nature 2021 LATEX template Article Title 45 Chen, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Rare Helium-Bearing Compound FeO2He Stabilized at Deep-Earth Conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 121, 255703 (2018) [153] Hume-Rothery, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Raynor, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : The structure of metals and alloys, 4th edn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Institute of Metals, London (1962) [154] Miedema, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : A simple model for alloys, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rules for the alloying behavious of transition metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Philips Tech.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 33, 149–160 (1973) [155] Miedema, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : A simple model for alloys, II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The influence of ionicity on the stability and other physical properties of alloys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Philips Tech.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 33, 196–202 (1973) [156] Darken, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gurry, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Physical chemistry of metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' McGraw-Hill, New York (1953) [157] Chelikowsky, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Solid solubilities in divalent alloys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 19, 686–701 (1979) [158] Alonso, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Simozar, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of solid solubility in alloys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 22, 5583–5589 (1980) [159] Wang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Atomic-size effect and solid solubility of multicomponent alloys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Scr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 94, 28–31 (2015) [160] Dubrovinskaia, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McCammon, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Iron-magnesium alloying at high pressures and temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Condens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Matter 16, 1143 (2004) [161] Gao, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Su, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shao, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lv, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Iron- magnesium compounds under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' New J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 43, 17403– 17407 (2019) [162] Fang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zheng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wentz- covitch, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ho, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unconventional iron-magnesium compounds at terapascal pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 104, 144109 (2021) [163] Dubrovinskaia, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kantor, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Crichton, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dmitriev, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shen, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Vitos, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ahuja, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Johansson, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Abrikosov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Beating the Miscibility Barrier between Iron Group Elements and Magnesium by High-Pressure Alloying.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 95, 245502 (2005) [164] Feng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hennig, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Emergent reduc- tion of electronic state dimension in dense ordered Li-Be alloys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 451, 445–448 (2008) Springer Nature 2021 LATEX template 46 Article Title [165] Degtyareva, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Simple metals at high pressures: the Fermi sphere- Brillouin zone interaction model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Usp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 49, 369–388 (2006) [166] Berger, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Walters, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lee, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Connecting the Chem- ical and Physical Viewpoints of What Determines Structure: From 1-D Chains to γ-Brasses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 111, 4522–4545 (2011) [167] Feng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Double-diamond NaAl via pres- sure: Understanding structure through Jones zone activation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 132, 114106 (2010) [168] Guo, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akselrud, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bobnar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burkhardt, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schmidt, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Weak Interactions under Pressure: hp-CuBi and Its Analogues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 56, 5620–5624 (2017) [169] Clarke, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Walsh, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jacobsen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wolverton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Freedman, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Creating binary Cu-Bi Compounds via High-Pressure Synthesis: A Combined Experimental and Theoretical Study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 29, 5276–5285 (2017) [170] Chen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Analytis, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mo, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Qi, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dai, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fisher, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hus- sain, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shen, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Experimental Realization of a Three-Dimensional Topological Insulator, Bi2Te.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 325, 178–181 (2009) [171] Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Weng, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prabhakaran, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mo, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shen, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dai, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hussain, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Discovery of a Three-Dimensional Topological Dirac Semimetal, Na3Bi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 343, 864–867 (2014) [172] Reynolds, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lane, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconducting Bismuth Alloys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 79, 405 (1950) [173] Matthias, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geballe, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Andres, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Corenzwit, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Many More Superconducting Bismuth Phases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 17, 640–643 (1966) [174] Hor, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Williams, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Checkelsky, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Roushan, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Seo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zandbergen, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yazdani, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ong, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cava, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Supercon- ductivity in CuxBi2Se3 and its Implications for Pairing in the Undoped Topological Insulator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 104, 057001 (2010) [175] Walsh, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Clarke, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jacobsen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Freedman, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Discovery of FeBi2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ACS Cent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 867–871 (2016) [176] Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Naghavi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wolverton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of superconducting iron-bismuth intermetallic compounds at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 8, Springer Nature 2021 LATEX template Article Title 47 2226 (2017) [177] Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hegde, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jacobsen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wolverton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Exploring the High-Pressure Materials Genome.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' X 8, 041021 (2018) [178] Walsh, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Clarke, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Puggioni, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tamerius, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rondinelli, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jacobsen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Freedman, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : MnBi2: A Metastable High-Pressure Phase in the Mn-Bi System.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 31, 3083–3088 (2019) [179] Takizawa, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Uheda, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Endo, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': A new ferromagnetic polymorph of CrSb2 synthesized under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Alloys Compd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 287, 145–149 (1999) [180] Wu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Steinle-Neumann, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Qin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kanzaki, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-induced phase transitions of AX2-type iron pnictides: an ab initio study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Condens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Matter 21, 185403 (2009) [181] Poffo, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Souza, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Trichˆes, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', de Lima, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grandi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Polian, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gauthier, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structural and optical studies of FeSb2 under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Physica B 407, 4686–4694 (2012) [182] Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tence, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Janson, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Koz, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Krellner, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burkhardt, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rosner, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Steglich, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': CoBi3: A Binary Cobalt-Bismuth Compound and Superconductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 52, 9853–9857 (2013) [183] Tenc´e, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Janson, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Krellner, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rosner, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Steglich, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': CoBi3- the first binary compound of cobalt with bis- muth: high-pressure synthesis and superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Condens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Matter 26, 395701 (2014) [184] Clarke, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Walsh, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Malliakas, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goedecker, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wolverton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Freedman, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Discovery of a Superconducting Cu-Bi Intermetallic Compound by High-Pressure Synthesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 128, 13644–13647 (2016) [185] Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wolverton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Dense superconducting phases of copper- bismuth at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' M 1, 031801 (2017) [186] Altman, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tamerius, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Koocher, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Walsh, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rondinelli, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jacobsen, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Freedman, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Compu- tationally Directed Discovery of MoBi2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 143, 214–222 (2021) [187] Pauling, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The Nature of the Chemical Bond vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 260.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Cornell University Press, Ithaca, NY (1960) Springer Nature 2021 LATEX template 48 Article Title [188] Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Struzhkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Compression of ice to 210 gigapascals: Infrared evidence for a symmetric hydrogen-bonded phase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 273(5272), 218–220 (1996) [189] Bernasconi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Silvestrelli, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Parrinello, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Ab initio infrared absorp- tion study of the hydrogen-bond symmetrization in ice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 81(6), 1235 (1998) [190] Zhang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure phase transitions of solid HF, HCl, and HBr: An ab initio evolutionary study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 82(1), 014108 (2010) [191] Errea, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Calandra, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mauri, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Quantum hydrogen-bond symmetrization in the superconducting hydrogen sulfide system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 532(7597), 81–84 (2016) [192] Vij, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pavlovich, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wilson, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Vij, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Christe, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Experi- mental Detection of the Pentaazacyclopentadienide (Pentazolate) Anion, cyclo-N− 5 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 114, 3177–3180 (2002) [193] Wang, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Recent advances in the synthe- ses and properties of polynitrogen pentazolate anion cyclo-N− 5 and its derivatives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 47, 7522–7538 (2018) [194] Shen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Qian, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dong, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Novel lithium-nitrogen compounds at ambient and high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 14204 (2015) [195] Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystalline LiN5 Predicted from First- Principles as a Possible High-Energy Material.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 2363–2366 (2015) [196] Laniel, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Weck, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Loubeyre, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Direct Reaction of Nitrogen and Lithium up to 75 GPA: Synthesis of the Li3N, LiN, LiN2, and LiN5 Compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 57, 10685–10693 (2018) [197] Zhou, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sui, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shi, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhao, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Guo, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Lithium Pentazolate Synthesized by Laser Heating-Compressed Lithium Azide and Nitrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 124, 11825–11830 (2020) [198] Laniel, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Weck, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gaiffe, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Garbarino, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Loubeyre, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High- Pressure Synthesized Lithium Pentazolate Compound Metastable under Ambient Conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9, 1600–1604 (2018) [199] Steele, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oleynik, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Sodium pentazolate: A nitrogen rich high energy density material.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 643, 21–26 (2016) Springer Nature 2021 LATEX template Article Title 49 [200] Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Han, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Exotic stable cesium polynitrides at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 16902 (2015) [201] Li, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Simple Route to Metal cyclo-N− 5 Salt: High-Pressure Synthesis of CuN5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 122, 22339–22344 (2018) [202] Xia, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zheng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yuan, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure- Stabilized High-Energy-Density Alkaline-Earth-Metal Pentazolate Salts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 123, 10205–10211 (2019) [203] Liu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Moderate Pressure Stabilized Pentazolate Cyclo-N− 5 Anion in Zn(N5)2 Salt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 59, 8002–8012 (2020) [204] Huang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Frapper, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Barium-Nitrogen Phases Under Pressure: Emer- gence of Structural Diversity and Nitrogen-Rich Compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 30, 7623–7636 (2018) [205] Wang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Larklimi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Valencia, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gu´egan, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Frapper, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of Novel Tin Nitride SnxNy Phases Under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 124, 8080–8093 (2020) [206] Du, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': IrN4 and IrN7 as potential high-energy-density materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 154, 054706 (2021) [207] Wu, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wan, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gong, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shen, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gou, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of Stable Iron Nitrides at Ambient and High Pressure with Progressive Formation of New Polynitrogen Species.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 30, 8476–8485 (2018) [208] Niu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shi, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': New Cadmium-Nitrogen Compounds at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 60, 6772–6781 (2021) [209] Liu, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-stabilized GdN6 with an armchair-antiarmchair structure as a high energy density material.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A 9, 16751–16758 (2021) [210] Bykov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykova, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ponomareva, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='and Abrikosov, Chari- ton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mahmood, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Stabilization of Polynitrogen Anions in Tantalum-Nitrogen Com- pounds Under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 133, 9085–9090 (2021) [211] Gao, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Novel Si Networks in the Ca/Si Phase Diagram under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 118, 25167–25175 (2014) Springer Nature 2021 LATEX template 50 Article Title [212] Li, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xin, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystal structures of CsSi6 at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 150, 144–148 (2018) [213] H¨ubner, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Carrillo-Cabrera, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prots, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bobnar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unconventional Metal-Framework Interaction in MgSi5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 58, 12914–12918 (2019) [214] Zhang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Silicon Framework-Based Lithium Silicides at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ACS Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Interfaces 8, 16761–167167 (2016) [215] Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Theoretical predictions of novel superconducting phases of BaGe3 stable at atmospheric and high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 54(6), 2875–2884 (2015) [216] Castillo, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Baranov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burkhardt, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cardoso-Gil, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schnelle, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bobnar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Germanium Dumbbells in a New Super- conducting Modification of BaGe3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 55, 4498–4503 (2016) [217] Hooper, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Lithium subhydrides under pressure and their superatom-like building blocks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ChemPlusChem 77, 969–972 (2012) [218] Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wosylus, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rosner, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schnelle, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ormeci, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meier, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Baranov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nicklas, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Leipe, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', M¨uller, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Dumb- bells of five-connected silicon atoms and superconductivity in the binary silicides MSi3 (M= Ca, Y, Lu).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 134(33), 13558–13561 (2012) [219] H¨ubner, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akselrud, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schnelle, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burkhardt, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bobnar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prots, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-Pressure Synthesis and Chemical Bonding of Barium Trisilicide BaSi3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Materials 12, 145 (2019) [220] Schnelle, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ormeci, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wosylus, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meier, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Dumbbells of five-connected Ge atoms and superconductivity in CaGe3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 51(10), 5509–5511 (2012) [221] Nishikawa, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fukuoka, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Inumaru, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-Pressure Synthesis and Electronic Structure of a New Superconducting Strontium Germanide (SrGe3) Containing Ge2 Dumbbells.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 54, 7433–7437 (2015) [222] Castillo, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schnelle, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Baranov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Burkhardt, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bobnar, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cardoso-Gil, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grin, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Trigermanides AEGe3 (AE=Ca, Sr, Ba): chemical bonding and superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Naturforsch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 71, 585–592 (2016) Springer Nature 2021 LATEX template Article Title 51 [223] Lazicki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McGonegle, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rygg, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Braun, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Swift, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gor- man, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Heighway, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Higginbotham, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Suggit, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fratanduono, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Coppari, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wehrenberg, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kraus, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Erskine, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bernier, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McNaney, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rudd, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Collins, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eggert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wark, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Metastability of diamond ramp-compressed to 2 terapascals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 589, 532–535 (2021) [224] Simon, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Metal clusters inside out.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Philos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Trans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A 368, 1285– 1299 (2010) [225] Oka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Interstellar H+ 3 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 113, 8738–8761 (2013) [226] Wang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Iitaka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Stabilization of H3+ in the high pressure crystalline structure of HnCl (n = 2–7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6(1), 522–526 (2015) [227] Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Predicted Formation of H3+ in Solid Halogen Polyhydrides at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A 119, 11059–11065 (2015) [228] Zeng, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Frapper, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Emergence of novel hydrogen chlorides under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 19, 8236–8242 (2017) [229] Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shamp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Terpstra, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The Li–F–H ternary system at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 154(12), 124709 (2021) [230] Ayouz, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dulieu, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gu´erout, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Robert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kokoouline, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Potential energy and dipole moment surfaces of H− 3 molecule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 132(19), 194309 (2010) [231] Hooper, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Rubidium Polyhydrides Under Pressure: Emer- gence of the Linear H3− Anion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 18, 5013–5021 (2012) [232] Shamp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hooper, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Compressed cesium polyhydrides: Cs+ sublattices and H3−-three-connected nets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 51(17), 9333– 9342 (2012) [233] Hooper, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Altintas, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shamp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Polyhydrides of the alka- line earth metals: a look at the extremes under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 117(6), 2982–2992 (2013) [234] Struzhkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kim, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stavrou, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Muramatsu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Synthe- sis of sodium polyhydrides at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 7(1), 12267 (2016) Springer Nature 2021 LATEX template 52 Article Title [235] Liu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-Induced Structures and Properties in Indium Hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 54, 9924–9928 (2015) [236] Chen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geng, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of Stable Ground-State Lithium POlyhydrides under High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 56, 3867–3874 (2017) [237] Ye, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zarifi, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : High Hydrides of Scandium under Pressure: Potential Superconductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 122, 6298–6309 (2018) [238] Xie, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Song, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jiang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Redfern, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kresin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Hydrogen Pentagraphene- like Structure Stabilized by Hafnium: A High-Temperature Conventional Superconductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 125, 217001 (2020) [239] Feng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Compressed sodalite- like MgH6 as a potential high-temperature superconductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' RSC Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 59292–59296 (2015) [240] Qian, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sheng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Song, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Theoretical study of stability and superconductivity of ScHn (n=4-8) at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 96, 094513 (2017) [241] Abe, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Hydrogen-rich scandium compounds at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 96, 144108 (2017) [242] Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-stabilized Superconductive Yttrium Hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 9948 (2015) [243] Zhao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ao, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ye, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Phase Diagram and Bonding States of Pu-H Binary Compounds at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 124, 7361–7369 (2020) [244] Hai, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jiang, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhong, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Cage Structure and Near Room-Temperature Superconductivity in TbHn (n = 1-12).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 125, 3640–3649 (2021) [245] Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Galasso, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ivanova, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gavriliuk, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tkachenko, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boldyrev, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Novel Strongly Correlated Europium Superhydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 12(1), 32–40 (2021) [246] Sun, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kuang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Keen, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Second group Springer Nature 2021 LATEX template Article Title 53 of high-pressure high-temperature lanthanide polyhydride superconduc- tors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 102, 144524 (2020) [247] Abe, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure properties of dense metallic zirconium hydrides studied by calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 98, 134103 (2018) [248] Hooper, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Terpstra, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shamp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The Composition and Constitution of Compressed Strontium Polyhydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 118, 6433–6447 (2014) [249] Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Iitaka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structural Morpholo- gies of High-Pressure Polymorphs of Strontium Hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 17, 19379–19385 (2015) [250] Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zarifi, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Terpstra, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The search for supercon- ductivity in high pressure hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In: Reedijk, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' (ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=') Reference Module in Chemistry, Molecular Sciences and Chemical Engineer- ing, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 1–36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Elsevier, Waltham, MA (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='1016/ b978-0-12-409547-2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='11435-0 [251] Hilleke, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Rational Design of Superconducting Metal Hydrides via Chemical Pressure Tuning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 134, 202207589 (2022) [252] Shao, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Song, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unique Phase Diagram and Superconductivity of Calcium Hydrides at High Pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Inorg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 58, 2558–2564 (2019) [253] Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Actinium Hydrides AcH10, AcH12, and AcH16 as High-Temperature Conventional Superconductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 9(8), 1920–1926 (2018) [254] Errea, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Belli, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Monacelli, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sanna, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Koretsune, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tadano, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bianco, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Calandra, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Arita, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mauri, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Flores-Livas, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Quantum crystal structure in the 250-kelvin superconducting lanthanum hydride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 578, 66–69 (2020) [255] Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lv, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Route to a superconducting phase above room temperature in electron-doped hydride compounds under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Letters 123(9), 097001 (2019) [256] Onnes, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Further Experiments with Liquid Helium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' On the Change of Electrical Resistance of Pure Metals at Very low Tempera- tures, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' The Disappearance of the Resistance of Mercury,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ned.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Akad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Wet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 14, 113–115 (1911) Springer Nature 2021 LATEX template 54 Article Title [257] Matthias, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geballe, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geller, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Corenzwit, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconduc- tivity of Nb3Sn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 95, 1435 (1954) [258] Gavaler, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity in Nb-Ge films above 22 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 23, 480 (1973) [259] Anderson, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Matthias, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 144, 373– 381 (1964) [260] Wu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ashburn, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Torng, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hor, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity at 93 K in a new mixed-phase Y-Ba-Cu-O compound system at ambient pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 58, 908 (1987) [261] Gao, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xue, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xiong, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ramirez, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eggert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity up to 164 K in HgBa2Cam−1CumO2m+2+δ (m=1,2, and 3) under quasihydrostatic pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 50, 4260–4263 (1994) [262] Finnemore, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Stromberg, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Swenson, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconducting Properties of High-Purity Niobium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 149, 231–243 (1966) [263] Nagamatsu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nakagawa, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Muranaka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zenitani, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akimitsu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity at 39 K in magnesium diboride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 410, 63–64 (2001) [264] Sakata, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nakamoto, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconducting state of Ca- VII below a critical temperature of 29 K at a pressure of 216 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 83, 220512 (2011) [265] Drozdov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity above 100 K in PH3 at high pressures (2015) arXiv:1508.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='06224 [cond-mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='supr- con] [266] Drozdov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ksenofontov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shylin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Conventional superconductivity at 203 Kelvin at high pressures in the sulfur hydride system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 525, 73–76 (2015) [267] Salke, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Esfahani, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Synthesis of clathrate cerium superhydride CeH9 at 80-100 GPa with atomic hydrogen sublattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 10, 4453 (2019) [268] Snider, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dasenbrock-Gammon, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', McBride, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Debessai, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Vin- dana, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Vencatasamy, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lawler, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Salamat, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Room-temperature superconductivity in a carbonaceous sulfur hydride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 586(7829), 373–377 (2020) Springer Nature 2021 LATEX template Article Title 55 [269] Di Cataldo, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Heil, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', von der Linden, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boeri, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': LaBH8: Towards high-Tc low-pressure superconductivity in ternary superhydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 104, 020511 (2021) [270] Gao, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yan, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xiang, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Phonon-mediated high- temperature superconductivity in the ternary borohydride KB2H8 under pressure near 12 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 104, 100504 (2021) [271] Zhang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hutcheon, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shipley, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Song, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Du, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kresin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Design Princi- ples for High-Temperature Superconductors with Hydrogen-Based Alloy Backbone at Moderate Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 128, 047001 (2022) [272] Matthias, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Progress in Low Temperature Physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' In: Supercon- ductivity in the Periodic System, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 138–150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Elsevier, Amsterdam (1957) [273] Bardeen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cooper, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schrieffer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Microscopic Theory of Superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 106, 162–164 (1957) [274] Bardeen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cooper, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schrieffer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Theory of Superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 108, 1175–1204 (1957) [275] Pickett, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : The next breakthrough in phonon-mediated superconduc- tivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Physica C 468, 126–135 (2008) [276] Swift, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', White, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Low Temperature Heat Capacities of Magnesium Diboride (MgB2) and Magnesium Tetraboride (MgB4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 79, 3641–3644 (1957) [277] McMillan, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Transition Temperature of Strong-Coupled Supercon- ductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 167, 331–344 (1968) [278] Dynes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : McMillan’s Equation and the Tc of Superconductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 10, 615–618 (1972) [279] Allen, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dynes, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Transition Temperature of Strong-Coupled Superconductors Reanalyzed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 12, 905–922 (1975) [280] Hopfield, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Angular Momentum and Transition-Metal Superconduc- tivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 186, 443–451 (1969) [281] Quan, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ghosh, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickett, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Compressed hydrides as metallic hydrogen superconductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 100, 184505 (2019) [282] Buzea, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Robbie, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Assembling the puzzle of superconducting elements: a review.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Supercond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Technol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 18(1), 1 (2004) Springer Nature 2021 LATEX template 56 Article Title [283] Schilling, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity in the alkali metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' High Press.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 26, 145–163 (2006) [284] Hamlin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity in the metallic elements at high pres- sures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Physica C 514, 59–76 (2015) [285] Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity from insulating elements under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Physica C 514, 46–49 (2015) [286] Tuoriniemi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Juntunen-Nurmilaukas, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Uusvuori, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pentti, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Salmela, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sebedash, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity in lithium below 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='4 millikelvin at ambient pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 447, 187–189 (2007) [287] Falge Jr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity of hexagonal beryllium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A 24, 579–580 (1967) [288] Sizoo, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Onnes, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Further experiments with liquid helium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Influ- ence of elastic deformation on the supraconductivity of tin and indium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lab.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Leiden.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 180b, 13–26 (1925) [289] Jennings, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Swenson, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Effects of Pressure on the Superconduct- ing Transition Temperatures of Sn, In, Ta, Tl, and Hg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 112, 31–43 (1958) [290] Smith, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Will Pressure Destroy Superconductivity?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 159, 353–359 (1967) [291] Struzhkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gan, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity in Dense Lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 298, 1213–1215 (2002) [292] Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ishikawa, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Takao, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yagi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Amaya, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Super- conductivity in compressed lithium at 20 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 419, 597–599 (2002) [293] Deemyad, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schilling, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconducting Phase Diagram of Li Metal in Nearly Hydrostatic Pressures up to 67 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 91, 167001 (2003) [294] Chen, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Exploring high-temperature superconductivity in hard mat- ter close to structural instability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Matter Radiat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Extremes 5, 068102 (2020) [295] Kasinathan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kune˘s, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lazicki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Rosner, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yoo, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Scalet- tar, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickett, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity and Lattice Instability in Compressed Lithium from Fermi Surface Hot Spots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett 96, 047004 (2006) Springer Nature 2021 LATEX template Article Title 57 [296] Profeta, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Franchini, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lathiotakis, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Floris, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sanna, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mar- ques, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', L¨uders, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Massidda, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gross, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Continenza, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity in Lithium, Potassium, and Aluminum under Extreme Pressure: A First-Principles Study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 96, 047003 (2006) [297] Rodriguez-Prieto, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bergara, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Silkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Echenique, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Com- plexity and Fermi surface deformation in compressed lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 74, 172104 (2006) [298] Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', He, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zou, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Elec- tronic and phonon instabilities in face-centered-cubic alkali metals under pressure using ab initio calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 75, 064102 (2007) [299] Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tanaka, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marsiglio, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity in lithium under high pressure investigated with density functional and Eliashberg theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 79, 054524 (2009) [300] Bazhirov, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Noffsinger, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cohen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity and electron-phonon coupling in lithium at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 82, 184509 (2010) [301] Rousseau, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bergara, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Exotic high pressure behavior of light alkali metals, lithium and sodium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 81, 1 (2011) [302] Yue, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cheng, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liao, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Electron-phonon interaction and superconductivity in the high-pressure cI 16 phase of lithium from first principles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 20, 27125–27130 (2018) [303] Matsuoka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sakata, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nakamoto, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Takahama, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ichimaru, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mukai, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ohta, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hirao, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ohishi, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure- induced reentrant metallic phase in lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 89, 144103 (2014) [304] Yan, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-induced reappearance of superconductivity in the oC24 phase of lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 225, 7–11 (2016) [305] Wittig, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-Induced Superconductivity in Cesium and Yttrium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 24, 812–815 (1970) [306] Schwarz, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Takemura, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hanfland, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Syassen, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystal Structure of Cesium-V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 81, 2711–2714 (1998) [307] Deng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schilling, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Evidence for superconductivity in Rb metal above 55 GPa pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 100, 041109 (2019) Springer Nature 2021 LATEX template 58 Article Title [308] Debessai, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hamlin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Schilling, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Comparison of the pressure dependences of Tc in the trivalent d-electron superconductors Sc, Y, La, and Lu up to megabar pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 78, 064519 (2008) [309] Slocombe, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kuznetsov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Grochala, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Williams, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Edwards, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity in transition metals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phil.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Trans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' A 373, 20140476 (2015) [310] Ginzburg, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : What problems of physics and astrophysics seem now to be especially important and interesting (thirty years later, already on the verge of XXI century)?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-Usp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 42, 353–373 (1999) [311] Ginzburg, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Nobel Lecture: On Superconductivity and Superfluid- ity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='nobelprize.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='org/uploads/2018/06/ginzburg-lecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='pdf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Accessed: 2022-03-23 (2003) [312] Wigner, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huntington, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : On the Possibility of a Metallic Modifi- cation of Hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3, 764–770 (1935) [313] Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Metallic Hydrogen: A High-Temperature Superconduc- tor?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 21, 1748–1749 (1968) [314] Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Silvera, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Observation of the Wigner-Huntington transi- tion to metallic hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 355, 715–718 (2017) [315] Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Drozdov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kong, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Semimetallic molec- ular hydrogen at pressure above 350 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 15, 1246–1249 (2019) [316] Loubeyre, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Occelli, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dumas, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Synchrotron infrared spectroscopic evidence of the probable transition to metal hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 577, 631– 635 (2020) [317] Liu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dalladay-Simpson, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Howie, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Comment on ”Observation of the Wigner-Huntington transition to metallic hydrogen”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 357, 2286 (2017) [318] Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Struzhkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Comment on ”Observation of the Wigner-Huntington transition to metallic hydrogen”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 357, 9736 (2017) [319] Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cheng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dalladay-Simpson, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Howie, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Everything you always wanted to know about metallic hydrogen but were afraid to ask.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Matter Radiat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Extremes 5, 038101 (2020) Springer Nature 2021 LATEX template Article Title 59 [320] Monacelli, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Errea, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Calendra, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mauri, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Black metal hydro- gen above 360 GPa driven by proton fluctuations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 17, 63 (2021) [321] McMahon, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ceperley, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : High-temperature superconductivity in atomic metallic hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 84, 144515 (2011) [322] Ashcroft, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Hydrogen Dominant Metallic Alloys: High Temperature Superconductors?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 92, 187002 (2004) [323] Smithson, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marianetti, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Morgan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Van der Ven, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pred- ith, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ceder, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': First-principles study of the stability and electronic structure of metal hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 66, 144107 (2002) [324] Bourgeois, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Crivello, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cenedese, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Joubert, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Systematic First-Principles Study of Binary Metal Hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ACS Comb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 19, 512–523 (2017) [325] Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sadakov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sobolevskiy, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pudalov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ivanova, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gavriliuk, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lyubutin, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Struzhkin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bergara, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Errea, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bianco, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Calandra, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mauri, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Monacelli, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akashi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Anomalous high-temperature superconductivity in YH6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 33, 2006832–110 (2021) [326] Kong, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Minkov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kuzovnikov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Drozdov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Besedin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mozaffari, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Balicas, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Balakirev, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chari- ton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Knyazev, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Greenberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity up to 243 K in the Yttrium-Hydrogen System under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 12, 5075 (2021) [327] Jeon, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cho, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Origin of enhanced chemical precompression in cerium hydride CeH9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 10, 16878 (2020) [328] Liang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bergara, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wei, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Prediction of high-Tc superconductivity in ternary lanthanum borohydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 104, 134501 (2021) [329] Durajski, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Szcz¸e´sniak, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': New superconducting superhydride LaC2H8 at relatively low stabilization pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', 25070–25074 (2021) [330] Shatruk, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': ThCr2Si2 structure type: The “perovskite” of inter- metallics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 272, 198–209 (2019) [331] Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Electronic Structure and Superconductivity of Com- pressed Metal Tetrahydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 27(60), 14858–14870 (2021) Springer Nature 2021 LATEX template 60 Article Title [332] Mishra, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Muramatsu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Geballe, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ahart, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Baldini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': New calcium hydrides with mixed atomic and molecular hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 122, 19370–19378 (2018) [333] Li, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miao, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ti, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pre- dicted high-temperature superconductivity in cerium hydrides at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 126, 235901 (2019) [334] Zhou, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconducting praseodymium superhydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 6849 (2020) [335] Pe˜na-Alvarez, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Binns, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hermann, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kelsall, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dalladay- Simpson, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gregoryanz, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Howie, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Praseodymium polyhydrides synthesized at high temperatures and pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 100, 184109 (2019) [336] Zhou, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Aperis, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oppeneer, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Galasso, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kartsev, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-Pressure Synthesis of Magnetic Neodymium Poly- hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 142(6), 2803–2811 (2020) [337] Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wrona, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : High-Temperature Superconductivity in a Th-H System under Pressure Conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' ACS Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Interfaces 10, 43809–43816 (2018) [338] Semenok, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kvashnin, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ivanova, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Svitlyk, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fominski, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sadakov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sobolevskiy, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pudalov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity at 161 K in thorium hydride ThH10: Synthesis and properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Today 33, 36–44 (2020) [339] Hoffmann, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zheng, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Making and Breaking Bonds in the Solid State: The ThCr2Si2 Structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 89, 4175–4181 (1985) [340] Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The metallization and super- conductivity of dense hydrogen sulfide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 140(17), 174712 (2014) [341] Strobel, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ganesh, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kent, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Novel Cooperative Interactions and Structural Ordering in H2S-H2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 107, 255503 (2011) [342] Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhao, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Pressure-induced metallization of dense (H2S)2H2 with high-Tc superconductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 4, 6968 (2014) Springer Nature 2021 LATEX template Article Title 61 [343] Einaga, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sakata, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ishikawa, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Eremets, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Droz- dov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Troyan, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hirao, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ohishi, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Crystal structure of the superconducting phase of sulfur hydride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 12(9), 835–838 (2016) [344] Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nelson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Errea, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Calandra, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mauri, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Dissociation products and structures of solid H2S at strong compression.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 93, 020103 (2016) [345] Gordon, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xiang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bussmann-Holder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kremer, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Simon, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', K¨ohler, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Whangbo, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structure and Composition of the 200 K-Superconducting Phase of H2S at Ultrahigh Pressure: The Perovskite (SH−)(H3S+).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 55, 3682–3684 (2016) [346] Ishikawa, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nakanishi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shimizu, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Katayama-Yoshida, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oda, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Suzuki, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconducting H5S2 phase in sulfur-hydrogen system under high-pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 23160 (2016) [347] Akashi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sano, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Arita, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='and Tsuneyuki: Possible “Magn´eli” Phases and Self-Alloying in the Superconducting Sulfur Hydride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 117, 075503 (2016) [348] Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconducting Hydrogen Sulfide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 24, 1769–1778 (2018) [349] Laniel, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Winkler, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykova, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fedotenko, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chariton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mil- man, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinsky, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dubrovinskaia, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Novel sulfur hydrides synthesized at extreme conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 102, 134109 (2020) [350] Ortenzi, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cappelluti, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pietronero, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Band structure and electron- phonon coupling in H3S: A tight-binding model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 94(6), 064507 (2016) [351] Durajski, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Szcz¸e´sniak, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Non-BCS thermodynamic properties of H2S superconductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Physica C 515, 1–6 (2015) [352] Durajski, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Quantitative analysis of nonadiabatic effects in dense H3S and PH3 superconductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Scientific reports 6(1), 1–8 (2016) [353] Gor’kov, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kresin, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Pressure and high-Tc superconductivity in sulfur hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6(1), 1–7 (2016) [354] Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lobanov, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kruglov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhao, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Oganov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Konˆopkov´a, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Hydrogen sulfide at high pressure: Change in stoichiometry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 93(17), 174105 Springer Nature 2021 LATEX template 62 Article Title (2016) [355] Jarlborg, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bianconi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Breakdown of the Migdal approximation at Lifshitz transitions with giant zero-point motion in the H3S supercon- ductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6(1), 1–12 (2016) [356] Bussmann-Holder, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', K¨ohler, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Simon, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Whangbo, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bianconi, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Multigap superconductivity at extremely high temperature: a model for the case of pressurized H2S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Supercond.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Magn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 30(1), 151–156 (2017) [357] Szcz¸e´sniak, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Durajski, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : The isotope effect in H3S superconductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Solid State Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 249, 30–33 (2017) [358] Azadi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', K¨uhne, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : High-pressure hydrogen sulfide by diffusion quantum Monte Carlo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 146(8), 084503 (2017) [359] Arita, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Koretsune, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sakai, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akashi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Nomura, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sano, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Nonempirical Calculation of Superconducting Transition Temperatures in Light-Element Superconductors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Advanced Materials 29(25), 1602421 (2017) [360] Quan, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickett, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Van Hove singularities and spectral smearing in high-temperature superconducting H3S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 93(10), 104526 (2016) [361] Sano, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Koretsune, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tadano, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Akashi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Arita, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Effect of Van Hove singularities on high-Tc superconductivity in H3S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 93(9), 094525 (2016) [362] Heil, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boeri, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Influence of bonding on superconductivity in high- pressure hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 92, 060508 (2015) [363] Ge, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': First-principles demonstration of supercon- ductivity at 280 K in hydrogen sulfide with low phosphorus substitution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B, 224513 (2016) [364] Ge, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Hole-doped room- temperature superconductivity in H3S(1−x)Zx (Z=C,Si).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Today Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 15, 100330 (2020) [365] Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Su, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Durajski, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Qi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Effect of covalent bonding on the superconducting temperature of the H-S-Se system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B, 174101 (2018) [366] Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Thermodynamics and superconductivity in SxSe1−xH3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Springer Nature 2021 LATEX template Article Title 63 Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 99, 060102 (2019) [367] Guan, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconductivity of H3S doped with light elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 3, 043102 (2021) [368] Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hilleke, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lamichhane, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Dilute Carbon in H3S Under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' npj Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 8, 87 (2022) [369] Cui, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Route to high-Tc superconductivity via CH4-intercalated H3S hydride perovskites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 101(13), 134504 (2020) [370] Lamichhane, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kumar, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ahart, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Salke, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dasenbrock- Gammon, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Snider, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Meng, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lavina, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chariton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': X-ray diffraction and equation of state of the C-S-H room-temperature superconductor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 155, 114703 (2021) [371] Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykova, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chariton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Synthesis and structure of carbon- doped H3S compounds at high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 131, 025902 (2022) [372] Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jiang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Iitaka, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhong, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Computational discovery of a dynamically stable cubic SH3-like high- temperature superconductor at 100 GPa via CH4 intercalation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 101(17), 174102 (2020) [373] Harshman, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fiory, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': The superconducting transition temper- atures of C-S-H based on inter-sublattice S-H4-tetrahedron electronic interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 131, 015105 (2022) [374] Hirsch, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Marsiglio, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unusual width of the superconducting transition in a hydride.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nature 569, 9–10 (2021) [375] Gubler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Flores-Livas, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kozhevnikov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goedecker, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Missing theoretical evidence for conventional room-temperature superconductiv- ity in low-enthalpy structures of carbonaceous sulfur hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 014801 (2022) [376] Bykova, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bykov, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Chariton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Prakapenka, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Glazyrin, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Aslandukov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Aslandukova, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Criniti, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Kurnosov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goncharov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Structure and composition of C-S-H compounds up to 143 GPa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 103, 140105 (2021) [377] Somayazulu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Finger, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Hemley, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Mao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : High- Pressure Compounds in Methane-Hydrogen Mixtures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Science 271, Springer Nature 2021 LATEX template 64 Article Title 1400–1402 (1996) [378] Smith, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Collings, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Snider, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Petitgirard, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Smith, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', White, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Jones, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ellison, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Lawler, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Dias, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Salamat, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Lower pressure phases and metastable states of superconduct- ing photo-induced carbonaceous sulfur hydride (2021) arXiv:2111.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='15051 [cond-mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='supr-con] [379] Fu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Du, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Peng, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhang, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pickard, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Needs, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Singh, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zheng, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': High-pressure phase stability and superconductivity of pnictogen hydrides and chemical trends for compressed hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 28(6), 1746–1755 (2016) [380] Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gao, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tse, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Naumov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Crystal structure and superconductivity of PH3 at high pressures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 120(6), 3458–3461 (2016) [381] Shamp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Terpstra, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Falls, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Avery, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Decomposition products of phosphine under pressure: PH2 stable and superconducting?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Am.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 138(6), 1884–1892 (2016) [382] Flores-Livas, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Amsler, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Heil, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sanna, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Boeri, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pro- feta, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wolverton, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Goedecker, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Gross, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' : Superconductivity in metastable phases of phosphorus-hydride compounds under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' B 93(2), 020508 (2016) [383] Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Miller, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content='P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Shamp, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Superconducting phases of phosphorus hydride under pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Stabilization by mobile molecular hydrogen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 56(34), 10192–10195 (2017) [384] Liu, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wu, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zhou, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' andLu, Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Unravelling decomposition products of phosphine under high pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Raman Spectosc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 49, 721– 727 (2018) [385] Yuan, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Fang, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Pei, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zheng, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yang, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Stoichiometric evolutions of PH3 under high pressure: implication for high-Tc superconducting hydrides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 6, 524– 531 (2019) [386] Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Sun, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Huang, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Chemically Tuning Stability and Superconductivity of P-H Compounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 11, 935–939 (2020) [387] Shao, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Duan, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Ma, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Yu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Song, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Xie, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Li, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Tian, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Liu, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Cui, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Ternary superconducting cophosphorus hydrides stabilized via lithium.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' npj Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' 5, 104 (2019) Springer Nature 2021 LATEX template Article Title 65 [388] Geng, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Bi, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=', Zurek, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=': Structural Diversity and Superconductivity in S-P-H Ternary Hydrides Under Pressure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} +page_content=' C 126, 7208– 7220 (2022)' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/h9AzT4oBgHgl3EQfMvvr/content/2301.01139v1.pdf'} diff --git a/i9FRT4oBgHgl3EQfVze1/content/tmp_files/2301.13541v1.pdf.txt b/i9FRT4oBgHgl3EQfVze1/content/tmp_files/2301.13541v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..491da0555075288882e17c804cc98a6c0b4ba6bb --- /dev/null +++ b/i9FRT4oBgHgl3EQfVze1/content/tmp_files/2301.13541v1.pdf.txt @@ -0,0 +1,4557 @@ +arXiv:2301.13541v1 [cs.DS] 31 Jan 2023 +Singular Value Approximation and +Reducing Directed to Undirected Graph Sparsification +AmirMahdi Ahmadinejad +Amazon∗ +ahmadi@alumni.stanford.edu +John Peebles +Apple +peebles@apple.com +Edward Pyne† +MIT +epyne@mit.edu +Aaron Sidford +Stanford +sidford@stanford.edu +Salil Vadhan‡ +Harvard University +salil vadhan@harvard.edu +February 1, 2023 +Abstract +In this paper, we introduce a new, spectral notion of approximation between directed graphs, +which we call Singular Value (SV) approximation. +SV-approximation is stronger than pre- +vious notions of spectral approximation considered in the literature, including spectral ap- +proximation of Laplacians for undirected graphs [ST04], standard approximation for directed +graphs [CKP+17], and unit-circle approximation for directed graphs [AKM+20]. Moreover, SV +approximation enjoys several useful properties not known to be possessed by previous notions +of approximation, such as being preserved under products of random-walk matrices and with +matrices of bounded norm. +Notably, we show that there is a simple black-box reduction from SV-sparsifying Eulerian +directed graphs to SV-sparsifying undirected graphs. With this reduction in hand, we provide +a nearly linear-time algorithm for SV-sparsifying undirected and hence also Eulerian directed +graphs. This also yields the first nearly linear-time algorithm for unit-circle-sparsifying Eulerian +directed graphs. In addition, we give a nearly linear-time algorithm for SV-sparsifying (and +UC-sparsifying) random-walk polynomials of Eulerian directed graphs with second normalized +singular value bounded away from 1 by 1/poly(n). +Finally, we show that a simple repeated-squaring and sparsification algorithm for solving +Laplacian systems, introduced by Peng and Spielman [PS14] for undirected graphs, also works +for Eulerian digraphs whose random-walk matrix is normal (i.e. unitarily diagonalizable), if we +use SV-sparsification at each step. Prior Laplacian solvers for Eulerian digraphs are significantly +more complicated. +∗The work was started prior to joining Amazon and does not relate to Amazon +†Supported by an Akamai Presidential Fellowship. +‡Supported by a Simons Investigator Award. +1 + +Contents +1 +Introduction +3 +1.1 +Our Contribution: Singular-Value Approximation . . . . . . . . . . . . . . . . . . . . +4 +1.2 +Comparison to Previous Notions of Approximation . . . . . . . . . . . . . . . . . . . +5 +1.3 +Properties of SV Approximation +. . . . . . . . . . . . . . . . . . . . . . . . . . . . . +8 +1.4 +Algorithmic Results +. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +10 +1.5 +Open Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +12 +1.6 +Roadmap +. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +12 +2 +Preliminaries +13 +3 +Singular Value Approximation +13 +3.1 +Matrix Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +13 +3.2 +Equivalent definitions of SV approximation +. . . . . . . . . . . . . . . . . . . . . . . +15 +3.3 +Properties of Singular Value Approximation . . . . . . . . . . . . . . . . . . . . . . . +17 +4 +Sparsification +19 +4.1 +Cycle and Expander Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . +21 +4.2 +Sparsifying Bipartite Expanders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +22 +4.3 +Sparsifying a Constant Fraction of Edges +. . . . . . . . . . . . . . . . . . . . . . . . +24 +4.4 +Nearly Linear-Sized Sparsifers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +25 +4.5 +Derandomized Square Sparsification +. . . . . . . . . . . . . . . . . . . . . . . . . . . +26 +4.6 +Sparsification of Random Walk Polynomials . . . . . . . . . . . . . . . . . . . . . . . +27 +5 +Squaring algorithm for solving normal directed Laplacians +31 +A SV Approximation Proofs +39 +A.1 Equivalences +. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +40 +A.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +44 +A.3 Separations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . +48 +B Facts about Singular Values +49 +2 + +1 +Introduction +Given an undirected graph, its associated Laplacian matrix is L = D − A where D is the diagonal +matrix of degrees of the graph and A is its associated adjacency matrix. +The linear algebraic +properties of Laplacian matrices capture fundamental information about the corresponding graph, +including random walks, cuts, and more. Correspondingly, starting with the seminal work of Spiel- +man and Teng [ST04], nearly-linear time algorithms for solving Laplacian linear systems and related +problems [ST04, BSS12, CKP+16, CKP+17, CKK+18] have been established as an important tool +in algorithmic graph theory. These results have enabled asymptotic improvements for a variety of +problems, such as max-flow and min-cut [CKM+11,LRS13,KLOS14,vdBGJ+22,vdBLL+21], travel- +ing salesman [AGM+10,AG15], sampling random spanning trees [KM09,MST15], graph sparsifica- +tion [SS08,LS13,KMP14,CKM+14,LS18,ABKS21,JS21], multicommodity flow [KMP12,KLOS14], +and solving clustering and partitioning problems [AM85,KVV04,ACL06,OSV12]. +This “Laplacian paradigm” was initially formulated for undirected graphs. However, there is +a natural generalization of the Laplacian matrix to directed graphs, which analogously captures +the information about random walks on the directed graph [CKP+16]. +This generalization is +L = Dout − A⊤ where Dout is now the diagonal matrix of out-degrees and A⊤ is the transpose of +the adjacency matrix A. It is called the directed Laplacian matrix, to emphasize that the graph +corresponding to it is directed. +While the generalization of the definition of the Laplacian matrix to directed graphs is straight- +forward, the generalization of algorithms for Laplacian matrices is not so straightforward. Nonethe- +less, a sequence of papers [CKP+16,CKP+17,CKK+18] obtained nearly linear time algorithms for +sparsifying and solving directed Laplacian linear systems. In order to do so, they needed to over- +come three main difficulties not present in the undirected setting: +Challenge #1: +While the kernel of an undirected Laplacian matrix is the all ones vector, comput- +ing the kernel of a directed Laplacian matrix L corresponds to computing the stationary distribution +π of the random walk on the directed graph (LD−1 +outπ = 0). Without explicitly knowing the kernel, +it is not known how to efficiently perform any kind of useful sparsification or approximately solve +systems in L. This difficulty was overcome in [CKP+16, AJSS19] which provide reductions from +solving general directed Laplacian systems to the case where the graph is Eulerian, meaning that +every vertex has the same in-degree as out-degree. In Eulerian graphs, the stationary distribution +is simply proportional to the vertex degrees and the all ones vector is both the left and right kernels +of the associated directed Laplacian. +Challenge #2: +Undirected Laplacians L are not only symmetric but also positive semidefinite +(PSD), i.e. x⊤Lx ≥ 0 for all x . This leads to a natural definition of multiplicative approximation +between such matrices, namely saying that �L is an ε-approximation of L if (1−ε)L ⪯ �L ⪯ (1+ε)L, +where ⪯ is the L¨owner order on PSD matrices. +However, even though Laplacians of directed +graphs are potentially asymmetric, the quadratic form x⊤Lx depends only on a symmetrization +of the Laplacian (x⊤Lx = x⊤(1/2)(L + L⊤))x. Consequently, the quadratic form discards key +information about the associated directed graph (e.g. the quadratic form of a directed cycle and an +undirected cycle are the same). Thus, defining approximation for directed graphs (even Eulerian +ones) is more challenging than for undirected graphs and more complex notions of approximations +were introduced in [CKP+16, AKM+20]. This additional complexity suggest that one formulate +new sparsification algorithms that take into account the directedness of the graph. +3 + +Challenge #3: +Finally, the recursive identities used for solving undirected Laplacian systems, +while behaving nicely with respect to PSD approximation, do not behave as nicely with respect +to the new directed matrix approximation definitions mentioned in Challenge #2. This necessi- +tates different, more sophisticated recursion with a more involved analysis of the error [CKP+17, +CKK+18,AKM+20,KMG22]. +1.1 +Our Contribution: Singular-Value Approximation +In this paper, we present a stronger and more robust notion for addressing Challenge #2 and use +it make progress in simplifying approaches to Challenge #3. We do not address Challenge #1; +rather the existing reductions from general directed graphs to Eulerian ones allows us to focus on +Eulerian digraphs without loss of generality. +Specifically, we introduce a novel definition of approximation for asymmetric matrices, which +we call singular-value approximation (or SV approximation). For simplicity in the introduction, we +focus on the case of regular directed graphs, i.e. directed graphs where for some value d ≥ 0, every +vertex has in-degree d and out-degree d. (In the case of digraphs with nonnegative edge weights, +we obtain the in- and out-degrees by summing the in-coming or out-going edge weights at each +vertex.) However, all of our results generalize to Eulerian digraphs and some generalize to wider +classes of complex matrices. To introduce SV approximation, let A be the adjacency matrix of +a d-regular digraph, i.e. A is a nonnegative real matrix where every row and column sum equals +d. Then the (in- and out-) degree matrix is simply dI. Dividing by d, it is equivalent to study +approximation of the random-walk matrix W = A/d, which is doubly stochastic, and has degree +matrix I. +Definition 1.1 (SV approximation for doubly stochastic matrices). For doubly stochastic matrices +W, � +W ∈ Rn×n we say that � +W is an ε-singular-value (SV) approximation of W, written � +W +sv≈ε W, +if for all vectors x, y ∈ Rn, we have +���x⊤(� +W − W)y +��� ≤ ε +4 · +� +x⊤(I − WW⊤)x + y⊤(I − W⊤W)y +� +. +(1) +This formulation of SV-approximation is one of several equivalent formulations we provide later +in the paper (Lemma 3.7). We can equivalently define SV approximation between doubly stochastic +matrices by requiring Equation (1) to hold for all complex test vectors x, y ∈ Cn. SV-approximation +can also be defined equivalently by replacing condition (1) with +���x⊤(� +W − W)y +��� ≤ ε +4 · +� +[x⊤(I − WW⊤)x] · [y⊤(I − W⊤W)y]. +(2) +These two formulations, (1) and (2), differ only in using the geometric mean or the arithmetic mean +of the terms involving x and y on the right-hand side. The formulation in terms of the geometric +mean implies the one in terms of the arithmetic mean (since the geometric mean is no larger than +the arithmetic mean); the converse follows by optimizing over scalar multiples of x and y (as was +done in e.g. [CKP+17,AKM+20]). Both formulations can also be rewritten more simply by noting +that +x⊤(I − WW⊤)x = ∥x∥2 − ∥x⊤W∥2 and y⊤(I − W⊤W)y = ∥y∥2 − ∥Wy∥2, +but the description in terms of quadratic forms will be more convenient for comparison with previous +notions of approximation. In Section 3, we provide more general definitions of SV approximation +which also apply to unnormalized directed Laplacians and even to complex matrices. +4 + +We prove that SV approximation is strictly stronger than previous notions of spectral approxi- +mation considered in the literature for both asymmetric and symmetric matrices, and enjoys several +useful properties not possessed by the previous notions. Most notably, there is a simple black-box +reduction from SV-sparsifying Eulerian directed graphs to SV-sparsifying undirected graphs; no +such reduction is known for prior notions of asymmetric spectral approximation. +Furthermore, we give efficient algorithms for working with SV approximation. These include +nearly linear-time algorithms for SV-sparsifying undirected and hence also Eulerian directed graphs +(Theorem 1.8), as well as random-walk polynomials associated with such graphs (Theorem 1.9). We +also show that a simple repeated-squaring and sparsification algorithm for solving Laplacian systems +also works for Eulerian digraphs whose random-walk matrix is normal (i.e. unitarily diagonalizable), +if we use SV-sparsification at each step (Theorem 1.11). Prior Laplacian solvers for Eulerian graphs +are more complex. We elaborate on these results in the next several subsections. +1.2 +Comparison to Previous Notions of Approximation +Let us compare Definition 1.1 to previous definitions of approximation. +Undirected spectral approximation. +Let’s start with the undirected case, where W = W⊤. +In this case, it can be shown that we can without loss of generality restrict to x = y ∈ Rn, and +� +W +sv≈ε W requires that for all x ∈ Rn, +���x⊤(� +W − W)x +��� ≤ ε +2 · +� +x⊤(I − W2)x +� +. +(3) +In contrast, the standard definition of spectral approximation (originally introduced by Spielman +and Teng [ST04]), which we denote by � +W ≈ε W, is equivalent to requiring that for all x ∈ Rn, we +have +���x⊤(� +W − W)x +��� ≤ ε · +� +x⊤(I − W)x +� +. +(4) +To compare inequalities (3) and (4), we write x = � +i civi, where v1, . . . , vn is an orthonormal +eigenbasis for W with associated eigenvalues λ1, . . . , λn. Since W is stochastic, |λi| ≤ 1 for all +i ∈ [n]. Then the right-hand side of SV inequality (3) becomes +ε +2 · +� +i∈[n] +c2 +i · (1 − λ2 +i ), +whereas the right-hand side of ST inequality (4) becomes +ε · +� +i∈[n] +c2 +i · (1 − λi). +Since each |λi| ≤ 1, the fact that SV approximation implies ST approximation then follows from +(1 − λ2 +i ) = (1 − λi)(1 + λi) ≤ 2(1 − λi). +However, we also see that inequality (3) can be much stronger than inequality (4) when W has +eigenvalues λi close to -1 (e.g. in a bipartite graph with poor expansion) because then 1 − λ2 +i is +close to 0, but 1−λi is bigger than 2. More generally, inequality (3) requires that � +W approximates +W very well on every test vector x that is concentrated on the eigenvectors whose eigenvalues +have magnitude close to 1, whereas inequality (4) only requires close approximation on the (signed) +eigenvalues that are close to 1. +5 + +We remark that another way of ensuring � +W preserves unit singular values is to replace W2 +in the SV inequality (3) with the matrix |W| where we replace all eigenvalues of W with their +absolute value rather than their square,1 so that we have: +x⊤(I − |W|)x = +� +i∈[n] +c2 +i · (1 − |λi|). +Using |W| instead of W2 results in an equivalent definition up to a factor of 2 in ε, and W2 +turns out to be convenient to work with.2 This viewpoint also explains why we stop at W2 in the +definition and don’t explicitly use higher powers; it is simply a convenient proxy for |W|, which +captures all powers. Indeed, for all k ∈ N +I − W2 ⪯ 2 · (1 − |W|) ⪯ 2 · +� +I − Wk� +. +Directed spectral approximation. +Turning to previous notions of spectral approximation +for directed graphs, standard approximation [CKP+17] generalizes the definition of Spielman and +Teng [ST04] by saying � +W ≈ε W if for all x, y ∈ Rn +���x⊤(� +W − W)y +��� ≤ ε +2 · +� +x⊤(I − W)x + y⊤(I − W)y +� +. +(5) +Equivalently, we can require that for all x, y ∈ Rn, +���x⊤(� +W − W)y +��� ≤ ε · +� +[x⊤(I − W)x] · [y⊤(I − W)y]. +(6) +It can be shown that for undirected graphs, standard approximation is equivalent to the condition +of Equation 4, so we will also refer to it as standard approximation. +The use of different left and right test vectors x and y on the left-hand side is crucial for capturing +the asymmetric information in � +W and W. As before, if x or y is concentrated on eigenvectors of +W whose eigenvalues are close to 1, then the right-hand side of ST inequality (6) is close to 0 +and � +W must approximate W very well. However, like the standard undirected ST inequality (4), +not much is required on eigenvalues near -1. Moreover, asymmetric matrices can have eigenvalues +that are not real and are equal to or close to complex numbers of magnitude 1. For example, the +eigenvalues of a directed n-cycle are the complex n’th roots of unity. +To address this issue, unit-circle (UC) approximation [AKM+20], written � +W +◦≈ε W, requires +that for all complex test vectors x, y ∈ Cn, we have +���x∗(� +W − W)y +��� ≤ ε +2 · +� +∥x∥2 + ∥y∥2 − |x∗Wx + y∗Wy| +� +. +(7) +That is, we take the complex magnitude of the terms involving W on the right-hand side of ST +inequality (5). That way, if x and y are concentrated on eigenvectors of W that have eigenvalue +near some complex number µ of magnitude 1, we require that � +W approximates W very well. +For example, consider the case where W is normal, i.e. +has an orthonormal basis of complex +eigenvectors v1, . . . , vn and with corresponding complex eigenvalues λ1, . . . , λn. Then if we write +x = � +i civi and y = � +i divi, the right-hand side of UC inequality (7) becomes: +ε +2 · +� +i∈[n] +(|ci|2 + |di|2) − +������ +� +i∈[n] +(|ci|2 + |di|2) · λi +������ +. +(8) +1Another way of describing |W| is as the psd square root of the psd matrix W2. +2|W| = (W2)1/2, i.e. |W| is the PSD square root of W2. In Definition 1.1, we could similarly replace WW⊤ and +W⊤W with their PSD square roots and obtain a definition that is equivalent up to a factor of 2 in ε. +6 + +If x and y are concentrated on eigenvalues λi ≈ µ where |µ| = 1, then this expression will be close +to 0. Unit-circle approximation has valuable properties not enjoyed by standard approximation, in +particular being preserved under powering: If � +W is an ε-UC approximation of W, then for every +positive integer k, � +Wk is an O(ε)-UC approximation of Wk; note that the quality of approximation +does not degrade with k. This property was crucial for the results of [AKM+20]. +However, UC approximation has two limitations compared to SV approximation. First, UC +expression (8) is only small if x and/or y is concentrated on eigenvalues that are all close to the +same point µ on the complex unit circle. Even in the undirected case, if x and y are mixtures of +eigenvectors with eigenvalue close to 1 and eigenvalue close to -1, then there will be cancellations +in the second term of UC expression (8) and the result will not be small. Second, the spectral +properties of asymmetric matrices are better captured by singular values than eigenvalues, since +singular values treat the domain and codomain as distinct. +To see how SV approximation addresses these limitations, let σ1, . . . , σn ≥ 0 be the singu- +lar values of W, let u1, u2, . . . , un ∈ Cn the corresponding left-singular vectors of W, and let +v1, . . . , vn ∈ Cn the corresponding right-singular vectors. If we write x = � +i ciui and y = � +i divi, +then the right-hand side of SV inequality (1) becomes: +ε +4 · + +� +i∈[n] +(|ci|2 + |di|2) · (1 − σ2 +i ) + + . +(9) +Consequently, SV-approximation requires high-quality approximation if x is concentrated on left- +singular vectors of singular value close to 1 and/or y is concentrated on right-singular vectors of +singular value close to 1. (For the “or” interpretation, use the formulation of SV approximation in +terms of inequality (2).) To compare with UC expression (8), let us consider what happens with a +normal matrix, where ui = vi and σi = |λi|. In this case, SV expression (9) amounts to bringing +the absolute value of UC expression (8) inside the summation (and squaring, which only makes a +factor of 2 difference), to avoid cancellations between eigenvalues of different phases. +Furthermore, for non-normal matrices, SV approximation retains the asymmetry of W even +on the right-hand side, by always using x on the left of W (thus relating to its decomposition +into left singular vectors) and y on the right of W (thus relating to its decomposition into right +singular vectors). Indeed, this feature allows us to even extend the definition of SV approximation +to non-square matrices. (See Section 3.) +Following the above intuitions, we prove that SV approximation is indeed strictly stronger than +the previous notions of approximation: +Theorem 1.2. For all doubly stochastic W and � +W, if � +W +sv≈ε W, then � +W +◦≈ε W (and hence +� +W ≈ε W). +On the other hand, for every ε ∈ (0, 1) and ε′ ∈ (ε, ∞), there exist symmetric +W, � +W ∈ R2×2 of spectral norm at most 1 such that � +W +◦≈ε W but it is not the case that � +W +svn +≈ ε′ W. +The separation above relies on a generalization of the definition of SV approximation to ma- +trices that can have negative entries, given in Section 3. It would be interesting to extend the +separation to matrices corresponding to graphs, and even doubly stochastic matrices. We note +that in [AKM+20] UC approximation was separated from standard approximation even for doubly +stochastic matrices, so SV approximation is also strictly stronger than standard approximation, +even for doubly stochastic matrices. +Other Work on SV Approximation. +The definition of SV approximation and some of our re- +sults on it (obtained in collaboration between Jack Murtagh and the authors) were first presented in +7 + +the first author’s PhD thesis [Ahm20] in August 2020. Independently, Kelley [Kel21] used a variant +of SV approximation to present an alternative proof of a result of [HPV21], who used unit-circle +approximation to prove that the Impagliazzo-Nisan–Wigderson pseudorandom generator [INW94] +fools permutation branching programs. Golowich and Vadhan [GV22] used SV approximation and +some of our results (presented in the aforementioned thesis) to prove new pseudorandomness prop- +erties of expander walks against permutation branching programs. Most recently, Chen, Lyu, Tal, +and Wu [CLTW22] have used a form of SV approximation to present alternative proofs of the +results of [AKM+20,PV21]. +1.3 +Properties of SV Approximation +SV approximation enjoys a number of novel properties not known to be possessed by previous +notions of spectral approximation. Most striking is the fact that directed approximation reduces +to undirected approximation. To formulate this, we define the symmetric lift of a matrix: +Definition 1.3. Given W ∈ Cm×n, let the symmetric lift of W be defined as +slift (W) +def += +�0n×n +W∗ +W +0m×m +� +. +Graph theoretically, the symmetric lift of W is the following standard operation: Given our +directed graph G on n vertices with random-walk matrix W, we lift G to an undirected bipartite +graph H with n vertices on each side, where we connect left-vertex i to right-vertex j if there is a +directed edge from i to j in G. Then slift (W) is the random-walk matrix of H. +Theorem 1.4. Let W and � +W be doubly stochastic matrices. +Then W +sv≈ε � +W if and only if +slift +� +� +W +� sv≈ε slift (W). +Thus, for the first time (as far as we know), sparsification of directed graphs reduces directly to +sparsification of undirected graphs. It would be very interesting to obtain a similar reduction for +other algorithmic problems in spectral graph theory, such as solving Laplacian systems. +Another novel property of SV approximation is that it is preserved under products of random- +walk matrices: +Theorem 1.5. Let W1, . . . , Wk and � +W1, . . . , � +Wk be doubly stochastic matrices such that � +Wi +sv≈εWi +for each i ∈ [k]. Then � +W1 � +W2 · · · � +Wk +sv≈ε+O(ε2) W1W2 · · · Wk. +Notably the approximation error does not grow with the number k of matrices being multiplied. +The analogous property was known for UC approximation only in the special case of powering, i.e. +W1 = W2 = · · · = Wk and � +W1 = � +W2 = · · · = � +Wk. +In addition, SV approximation is preserved under multiplication on the left and right by ar- +bitrary matrices of bounded spectral norm. Indeed, it can be seen as the “closure” of standard +approximation under this operation (up to a factor of 2). +Theorem 1.6. +1. Let W and � +W be doubly stochastic matrices such that � +W +sv≈εW. Then for all complex matrices +U and V of spectral norm at most 1, we have U� +WV +sv≈ε UWV, and hence U� +WV ≈ε UWV. +2. Let W and � +W be doubly stochastic matrices such that for all complex matrices U and V of +spectral norm at most 1, we have U� +WV ≈ε UWV. Then � +W +sv≈2ε W. +8 + +Since UWV and U� +WV need not be doubly stochastic matrices, Theorem 1.6 uses the gener- +alization of SV approximation to more general matrices, which can be found in Section 3. +Recall that standard spectral sparsifiers [ST04] are also cut sparsifiers [BK00]. That is, if �G is +an ε-approximation of G, then for every set S of vertices, the weight of the cut S in �G is within a +(1 ± ε) factor of the weight of S in G. Indeed, if we take the test vector x to be the characteristic +vector of the set S in inequality (4), we obtain +���� +Pr +(i,j)∼µedge +[i ∈ S, j /∈ S] − +Pr +(i,j)∼�µedge +[i ∈ S, j /∈ S] +���� ≤ ε · +Pr +(i,j)∼µedge +[i ∈ S, j /∈ S], +(10) +where µedge +def += µedge(W) (resp. +�µedge +def += �µedge(� +W)) denotes a random edge according to the +stationary distribution of W (resp. � +W), namely pick i ∈ [n] uniformly at random and obtain j +as a random neighbor of i using W (resp. � +W). Note that Pr(i,j)∼µedge[i ∈ S, j /∈ S] is simply a +normalized version of the weight of the cut (S, Sc); namely it equals the sum of the edge weights +leaving S divided the sum of all edge weights in the graph. +Similarly, we can obtain a combinatorial consequence of SV approximation, by taking x to be +a characteristic vector of a set S of vertices and taking y to be a characteristic vector of a set T of +vertices. This yields: +Proposition 1.7. Let � +W and W be doubly stochastic n × n random-walk matrices and suppose +that � +W +sv≈ε W. Then for every two subsets S, T ⊆ [n], we have +���� +Pr +(i,j)∼µedge +[i ∈ S, j ∈ T] − +Pr +(i,j)∼�µedge +[i ∈ S, j ∈ T] +���� ≤ ε +2· +� +Pr +(i,j,k)∼µfb +[i ∈ S, k /∈ S] · +Pr +(i,j,k)∼µfb +[i ∈ T, k /∈ T], +where: +• µedge +def += µedge(W) (resp. �µedge +def += �µedge(� +W)) denotes a random edge according to the station- +ary distribution of W (resp. � +W), namely pick i ∈ [n] uniformly at random and obtain j as a +random neighbor of i using W (resp. � +W), and +• µfb +def += µfb(W) denotes a random forward-backward walk, namely pick i ∈ [n] uniformly at +random, obtain j as a random neighbor of i using W, and k as a random neighbor of j using +W⊤. +Let us interpret Proposition 1.7. First, consider the case that W = J, the matrix with every +entry equal to 1/n (the random-walk matrix for the complete graph with self-loops). Then (i, j) ∼ +µedge and (i, j, k) ∼ µfb are all uniform and independent vertices, so Proposition 1.7 says: +����µ(S) · µ(T) − +Pr +(i,j)∼�µedge +[i ∈ S, j ∈ T] +���� ≤ ε +2 · +� +µ(S) · (1 − µ(S)) · µ(T) · (1 − µ(T)), +where µ(S) = |S|/n and µ(T) = |T|/n are the stationary probabilities of S and T, respectively. +This amounts to a restatement of the Expander Mixing Lemma ([Vad12, Lemma 4.15]); indeed +� +W +sv≈ε J if and only if � +W is a spectral expander with all nontrivial singular values at most ε/2. +Next, let’s consider the case that T = Sc. Then, +Pr +(i,j,k)∼µfb +[i ∈ S, k /∈ S] = +Pr +(i,j,k)∼µfb +[i ∈ T, k /∈ T]. +9 + +(In an Eulerian graph, the weight of edges crossing any cut in one direction must equal the weight +of edges crossing it in the other direction.) Thus, SV approximation implies that: +���� +Pr +(i,j)∼µedge +[i ∈ S, j /∈ S] − +Pr +(i,j)∼�µedge +[i ∈ S, j /∈ S] +���� ≤ ε +2 · +Pr +(i,j,k)∼µfb +[i ∈ S, k /∈ S]. +(11) +We claim that (11) is stronger than the standard notion of a cut approximator (10). Indeed, +Pr +(i,j,k)∼µfb +[i ∈ S, k /∈ S] ≤ +Pr +(i,j,k)∼µfb +[i ∈ S, j /∈ S] + +Pr +(i,j,k)∼µfb +[j ∈ S, k /∈ S] = 2 +Pr +(i,j)∼µedge +[i ∈ S, j /∈ S]. +(12) +Similarly, we have +Pr +(i,j,k)∼µfb +[i ∈ S, k /∈ S] ≤ +Pr +(i,j)∼µedge +[i ∈ S, j ∈ S] + +Pr +(i,j)∼µedge +[i /∈ S, j /∈ S] +(13) +Thus, we conclude that an SV-approximator not only approximates every cut to within a small +additive error that is scaled by the weight of the cut edges (as in (10)), but also scaled by the +weight of the uncut edges. +1.4 +Algorithmic Results +Even though SV approximation is stronger than previously considered notions of spectral approx- +imation, we show that it still admits sparsification: +Theorem 1.8. Given a regular directed graph G with n vertices and m edges, integer edge weights +in [0, U], and random-walk matrix W, and ε > 0 +1. There exists a weighted graph �G with at most O(nε−2 log(nU) log11 n) edges such that its +random-walk matrix � +W satisfies � +W +sv≈ε W. +2. There is a randomized nearly-linear time algorithm that whp outputs a weighted graph �G with +at most O(nε−2 log(nU) log19 n) edges such that its random-walk matrix � +W satisfies � +W +sv≈εW. +A more general theorem that also applies to Eulerian digraphs is stated in the main body of the +paper (Corollaries 4.12 and 4.13). Prior to this work, it was open whether or not even UC-sparsifiers +with O(n · poly(log n, ǫ−1)) edges existed for all unweighted regular digraphs. Instead, it was only +known how to UC-sparsify powers of a random walk matrix to bound the sparsity increase.3 +By Theorem 1.4, it suffices to prove Theorem 1.8 for undirected bipartite graphs. We obtain the +latter via an undirected sparsification algorithms based on expander partitioning [ST04]. It remains +an open question whether algorithms based on edge sampling can yield SV approximation or unit +circle approximation, even in undirected graphs. The standard approach to spectral sparsification +of undirected graphs via sampling, namely keeping each edge independently with probability pro- +portional to its effective resistance [SS08], does not work for SV or UC approximation. For example, +this method does not exactly preserve degrees, which we show is necessary for SV sparsification +(Lemma 3.17).4 +However, we remark that the work of Chu, Gao, Peng, Sawlani, and Wang [CGP+18] does +yield something closer to sparsification via degree preserving sampling for standard approxima- +tion [CKP+17] but not unit circle approximation. They show that if one has a directed graph and +3We extend this result to SV approximation (Lemma 4.15). +4Unlike standard spectral approximation, degrees cannot be fixed just by adding self loops; indeed, self-loops ruin +bipartiteness and periodicity, which are properties that UC and SV approximation retain (as they are captured by +eigenvalues like -1 or other roots of unity). +10 + +decomposes it into short “cycles” without regard for the direction of the edges on the cycle, then +one can sparsify by randomly eliminating either the clockwise or counterclockwise edges on each +such cycle. We build on their procedure and use it to obtain SV sparsification (and hence, unit +circle) by showing that this technique obtains SV approximation, even if the cycles are not short, +as long as (a) all the cycles are within expanding subgraphs, and (b) the cycles alternate between +forward and backward edges. (Note that such alternating cycles in a directed graph correspond to +ordinary cycles in the undirected lift given by Theorem 1.4.) +Sparsity +Approximation +Time +Subgraph? +Citation +O(nε−2 logc n) +Standard +O(m logc n) +No +[CKP+17] +O(nε−2 logc n) +Standard +O(m1+o(1)) +Yes +[CGP+18,PY19] +O(nε−2 log4 n) +Standard +O(nm) +Yes +[CGP+18] +O(nε−2 log12 n) +SV +Existential +Yes +Corollary 4.13 +O(nε−2 log20 n) +SV +O(m log7 n) +Yes +Corollary 4.12 +Our results also enable us to give the first nearly linear-time algorithm for sparsifying powers +of random walk matrices of graphs. More precisely, we show the following. +Theorem 1.9 (Theorem 4.17, informally stated). There is a randomized nearly-linear time al- +gorithm that, given a regular digraph G with random walk matrix W and any positive integer +k ≤ poly(n), outputs a nearly linear sized SV-sparsifer of Wk, provided that the second largest +singular value of W is at most 1 − 1/poly(n). +This theorem is stated more generally for Eulerian digraphs and proven as Theorem 4.17. +A regular (or even Eulerian) digraph whose weak mixing time (i.e. the mixing time of the 1/2- +lazy random walk) is at most poly(n) can be made to have second largest singular value at most +1 − 1/poly(n) by making it 1/poly(n)-lazy (Lemma B.5). Thus we obtain the following corollary +that allows us to unconditionally sparsify lazy random walks with even very small amounts of +laziness. +Corollary 1.10 (Corollary 4.19, informally stated). Let G be a regular digraph with poly(n) weak +mixing time. For any positive integer k ≤ poly(n) and γ ≥ 1/poly(n), we can, in nearly linear +time, output a nearly linear sized SV-sparsifer of ((1 − γ)W + γI)k. +Importantly we can take γ ≪ 1/k, so a γ-lazy walk of length k is very close (e.g. in total +variation distance) to an ordinary random walk of length k. +Our algorithm is relatively simple: we show that we can SV-sparsify the square of random +walk matrices of regular graphs, which gives all powers of two, then multiply different such powers +together to get the precise exponent we want. For example, to get a sparsifier for W7, we multiply +sparsifiers for W4, W2, and W1. The use of SV approximation plays an important role in the +analysis of this algorithm, because it has the property that the product of the approximations of +the powers still approximates the product of the true powers (Theorem 1.5). +Finally for Challenge #3, we make progress towards simplifying the recursion and analysis of +solving directed Laplacian linear system in the following way. We show that a simpler recursion, +a variant of the Peng-Spielman repeated squaring based recrusion (Equation (15)) and analysis +suffice if the directed Laplacian is normal (i.e. unitarily diagonalizable). A solver of this form was +given in the undirected case by Peng and Spielman [PS14], but the existing solvers for directed +graphs are all significantly more involved. Note that this result on the squaring recursion is the +only result in this paper that relies on a normality assumption, e.g. aforementioned sparsification +results hold for all Eulerian directed graphs. +11 + +Theorem 1.11. For a doubly stochastic normal matrix W ∈ Rn×n with ∥W∥ ≤ 1, let W = +W0, . . . , Wk−1 be a sequence of matrices such that for ε ≤ 1/4k we have +Wi +sv≈ǫ W2 +i−1 +∀0 < i < k, +(14) +and +Pi = 1 +2 [I + (I + Wi)Pi+1(I + Wi)] +∀0 ≤ i < k +(15) +defining the Peng-Spielman squaring recursion. +Then, for a matrix Pk, such that +���(I − W2k) +1 +2 +� +Pk − (I − W2k)+� +(I − W2k) +1 +2 +��� ≤ O(kǫ), we have +∥P0(I − W) − I∥B = +���(I − W) +1 +2 � +P0 − (I − W)+� +(I − W) +1 +2 +��� ≤ O(k2ǫ) +(16) +where B = ((I − W)1/2)∗(I − W)1/2. +Theorem 1.11 says that that we can compute a good preconditioner P0 for the Laplacian I−W +(eq. 16) by repeatedly computing SV-approximate squares (eq. 14) and use the simple recurrence +(eq. 15, starting with a preconditioner Pk for a sufficiently large power of W. Generally, Pk is easy +to obtain for a large enough k = O(log n) since W2k is well-approximated by a complete graph +(assuming the original graph is connected and aperiodic). +1.5 +Open Problems +One open problem is to determine whether or not it is possible to obtain linear-sized sparsifiers. +Recall that undirected graphs have sparsifiers with respect to standard spectral approximation that +have only O(n/ε2) nonzero edge weights [BSS12]. If this result could be extended to obtain linear- +sized SV-sparsifiers of undirected graphs, we would also have linear-sized sparsifiers for directed +graphs by Theorem 1.4, which would be a new result even for standard approximation [CKP+17]. +1.6 +Roadmap +The rest of the paper is organized as follows: +• Section 2 gives definitions we use throughout the paper. +• Section 3 defines SV-approximation and provides several equivalent characterizations of it +which are useful when analyzing algorithms and also interesting in their own right. These +results immediately imply a reduction from the directed to undirected case for SV approxi- +mation. +• Section 4 gives a direct SV sparsification algorithm based on sparsifying alternating cycles +within expanders, and shows that the derandomized square gives SV square sparsifiers, and +shows how to sparsify powers of random walk matrices with respect to SV approximation. +• Section 5 gives our simpler algorithm and analysis for solving Laplacian linear systems when +the matrix is unitarily diagonalizable (normal). +• In Appendix A, we prove equivalences, properties, and separations, including several deferred +proofs from the body of the paper. +12 + +2 +Preliminaries +We begin by providing notation and definitions that we will use throughout the paper. +Complex Numbers. +Let C denote the set of complex numbers. For z +def += a+bi ∈ C, let z∗ def += a−bi +and ℜ[z] +def += a and ℑ[z] +def += b. Define the magnitude of z as |z| +def += +√ +z∗z. +Matrix Notation. +For x ∈ Cn we let diag(x) ∈ Cn×n denote the diagonal matrix with [diag(x)]i,i = +xi for all i ∈ [n]. For complex matrix M that is not necessarily square, we write M∗ to denote +the conjugate transpose of M and M+ to denote the Moore-Penrose pseudoinverse of M. We let +lker(M) and rker(M) be the left and right kernels of M respectively. We say M ∈ Cn×n is Hermi- +tian if M∗ = M. M is positive semidefinite (PSD) if it is Hermitian and for every vector x ∈ Cn, +we have x∗Mx ≥ 0. M is normal if M∗M = MM∗. For a square matrix M, let SM = (M+M∗)/2 +denote its symmeterization. Note that SM is Hermitian, and if M is a scalar then SM = ℜ[M]. +Given any M ∈ Cn×n and subsets S, T ⊂ [n], we let MS,T ∈ CS×T be the submatrix of M with +rows specified by S and columns specified by T. We let Sc def += [n] \ S be the complement of S. +Schur Complements. +Given an n × n matrix M and S ⊂ [n], we define the Schur Complement +of M onto S as +SCS(M) = MSc,Sc − MSc,SM+ +S,SMS,Sc. +Eigenvalues and Singular Values. +For a symmetric matrix A with nonnegative entries, let +λi(A) denote the ith largest eigenvalue of A. +For a matrix A, let σi(A) +def += λi(A∗A). +Given +an undirected graph G with adjacency matrix A and degree matrix D +def += diag(A⃗1), let λ(G) +def += +λ2(D+/2AD+/2). +Norms. +Throughout we use ∥ · ∥ to denote the spectral norm, where for any A ∈ Cn×m, +∥A∥ +def += +sup +x∈Cm\{⃗0} +∥Ax∥ +∥x∥ . +L¨owner Order. +Given Hermitian A, B ∈ Cn×n we write A ⪯ B if B − A is PSD, i.e. for every +x ∈ Cn we have x∗Ax ≤ x∗Bx. +3 +Singular Value Approximation +In this section we formally define the notions of approximation we work with throughout the paper +(Section 3.1), provide a number of equivalent definitions of SV approximation (Section 3.2), and +give key properties of SV approximation (Section 3.3). All proofs in this section are deferred to +their analogous subsection in Appendix A. +3.1 +Matrix Approximation +First we define a general notion of matrix approximation with respect to arbitrary PSD matrices +(Definition 3.1). This definition and the equivalences are a generalization of those established in +[CKP+17]. +13 + +Intuitively, we say that a matrix �A is an ǫ-approximation of A with respect to error matrices E +and F if we can bound the bilinear form of their difference, i.e. x∗( �A−A)y, by the quadratic forms +of x with E and y with F, i.e. x∗Ex and y∗Fy. We use the term error matrix here to distinguish +this notion of approximation from standard approximation (Definition 3.3) which we define later. +Definition 3.1 (Matrix approximation). Let A, �A ∈ Cm×n, and let E ∈ Cm×m, F ∈ Cn×n be PSD +matrices. For ε ≥ 0, we say that �A is an ε-approximation of A with respect to error matrices E +and F if any of the following equivalent conditions hold: +1. +���x∗( �A − A)y +��� ≤ ε +2 (x∗Ex + y∗Fy) for all x ∈ Cm, y ∈ Cn +2. +���x∗( �A − A)y +��� ≤ ε · +√ +x∗Ex · √y∗Fy for all x ∈ Cm, y ∈ Cn +3. +���E+/2( �A − A)F+/2��� ≤ ε, lker( �A − A) ⊇ ker(E), and rker( �A − A) ⊇ ker(F). +If E = F, we say that �A is an ε-approximation of A with respect to error matrix E. +In the following Lemma 3.2 we specialize and simplify the equivalences presented in Defini- +tion 3.1 to Hermitian matrices. +Lemma 3.2 (Hermitian matrix approximation). Let A, �A ∈ Cm×m be Hermitian matrices and let +E ∈ Cm×m be a PSD matrix. Then the following are equivalent conditions. +1. �A is an ε-approximation of A with respect to error matrix E. +2. For all x ∈ Cm we have +���x∗( �A − A)x +��� ≤ ε · (x∗Ex). +3. −εE ⪯ �A − A ⪯ εE. +Next, for matrices A, �A, D ∈ Cm×m we define a natural notion of approximation between +D − A and D − �A which we call standard approximation (Definition 3.3). The choice of this name +is because when A and �A are adjacency matrices of undirected graphs with the same degrees, +standard approximation coincides with spectral approximation of the Laplacian matrices of the +associated graphs. Further, when A and �A are adjacency matrices of directed Eulerian digraphs +with the same degrees then standard approximation coincides with ǫ-approximation from [CKP+17]. +Standard approximation generalizes these two cases in a natural way even when D, which we call +a degree matrix, is not diagonal (as it was in these two cases and as we will often choose it to +be). +We use the term degree matrix to emphasize when we are using standard approximation +(Definition 3.3) and its variants rather than (Definition 3.1). +We now present a series of definitions and equivalences and properties. For clarity, the reader +can initially think of the case that D = I and A = W is a doubly stochastic matrix. +Definition 3.3 (Standard approximation). Let A, �A ∈ Cm×m and D ∈ Cm×m be a PSD matrix. +For ε ≥ 0, we say �A is a standard ε-approximation of A with respect to degree matrix D if E = +D − SA is PSD and �A is an ε-approximation of A with respect to error matrix E.5 +5Recall by Section 2 that SA +def += (A + A∗)/2. +14 + +With matrix approximation and standard approximation established, we can now define unit- +circle (UC) approximation and singular-value (SV) approximation. UC approximation as we present +it here, is a generalization of UC approximation as it was introduced for random walk matrices +in [AKM+20]. As discussed in the introduction, SV approximation is a new notion of approximation +introduced in this paper; it has a number of natural desirable properties and facilitates our results +on sparsification (Section 4) and linear system solving (Section 5). +Definition 3.4 (UC approximation). Let A, �A ∈ Cm×m and D ∈ Cm×m be PSD. For ε ≥ 0, we +say �A is a unit-circle (UC) ε-approximation of A with respect to degree matrix D if for every z ∈ C +with |z| = 1, z �A is a standard ε-approximation of zA with respect to degree matrix D. If this +holds, we write �A +◦≈ε A with respect to D. We omit “respect to D” when D = I. +Note that for doubly stochastic matrices � +W, W, the above definition (with D = I) corresponds +to requiring for every z ∈ C and x, y ∈ Cn, +���x∗(z� +W − zW)y +��� ≤ ε +2 +� +∥x∥2 + ∥y∥2 − x∗zWx − y∗zWy +� +. +(17) +Optimizing over z in (17) implies Equation (7). +Definition 3.5 (SV approximation). Let A, �A ∈ Cm×n and let Din ∈ Cm×m and Dout ∈ Cn×n be +PSD matrices. For ε ≥ 0, we say �A is a ε-singular-value (SV) approximation of A with respect to +degree matrices Din and Dout, if +1. ker(Din) ⊆ lker(A), and ker(Dout) ⊆ rker(A), +2. E = Din − AD+ +outA∗ and F = Dout − A∗D+ +inA are PSD, and +3. �A is an ε/2-approximation of A with respect to error matrices E and F. +If this holds, we write �A +Din,Dout +≈ε +A. If A ∈ Rn×m +≥0 +and Din = diag(A⃗1n), Dout = diag(⃗1⊤ +mA), then +we write �A +sv≈ε A. If Din = Im and Dout = In, then we write �A +svn +≈ ε A, which we call normalized +SV approximation. If m = n and Din = Dout = D, then we write �A +D≈ε A. +Note that normalized SV approximation does not require the relevant matrices to be non- +negative, whereas �A +sv≈ε A is only defined for real non-negative matrices (such as the adjacency +matrices of graphs). The maximally general definition captures both cases, so we will prove prop- +erties with respect to this notion and note their implications for the specialized notions. +3.2 +Equivalent definitions of SV approximation +Here we give several equivalent formulations of SV approximation. We first give conditions under +which the error matrices E and F in the definition of SV approximation are PSD. +Lemma 3.6 (Conditions for SV approximation to be defined). Let A, �A ∈ Cm×n, and let Din ∈ +Cm×m and Dout ∈ Cn×n be PSD matrices such that ker(Din) ⊆ lker(A) and ker(Dout) ⊆ rker(A). +Then the following are equivalent: +1. σmax(D+/2 +in AD+/2 +out ) ≤ 1. +2. Din − AD+ +outA∗ is PSD. +15 + +3. Dout − A∗D+ +inA is PSD. +4. For some scalar z ∈ C with |z| = 1, +� Din +zA +z∗A∗ +Dout +� +is PSD. +5. For every scalar z ∈ C with |z| ≤ 1, +� Din +zA +z∗A∗ +Dout +� +is PSD. +Suppose further that Din = diag(din) and Din = diag(dout) for din ∈ Rm +≥0, dout ∈ Rn +≥0. Then +Condition 1 below implies Condition 2 below, which implies Conditions 1–5 above. +1. A is nonnegative, dout = A⃗1n, din = ⃗1⊤ +mA. +2. For all i ∈ [n], (dout)i ≥ � +j |Ai,j|, and for all j ∈ [m], (din)j ≥ � +i |Ai,j|. +Next we give several equivalent definitions of SV approximation. +Lemma 3.7 (Equivalent formulations of SV approximation). Let A, �A ∈ Cm×n and let Din ∈ +Cm×m and Dout ∈ Cn×n be PSD matrices. Then the following are equivalent +1. �A +Din,Dout +≈ε +A. +2. slift +� +�A +� D≈ε slift +� +�B +� +, where +slift +� +�A +� += +�0m×m +A +A∗ +0n×n +� +, slift (A) = +� +0m×m +�A +�A∗ +0n×n +� +, D = +� Din +0m×n +0n×m +Dout +� +. +3. For some scalar z ∈ C with |z| = 1, ˜C is ε/2-approximation of C with respect to error matrix +E, where +C = +�0m×m +zA +0n×m +0n×n +� +, ˜C = +� +0m×m +z �A +0n×m +0n×n +� +, E = +� Din +zA +z∗A∗ +Dout +� +4. Item 3 holds for every z ∈ C such that |z| ≤ 1. +Each formulation of SV approximation in Lemma 3.7 has useful properties. Item 2 implies that +SV approximation between directed graphs is equivalent to a natural related statement between +undirected graphs, which we use for sparsification (See Section 4). +A version of Item 2 is not +known to hold for prior definitions of approximation between directed graphs, such as standard +approximation and unit circle approximation. Item 3 and Item 4 have an error matrix E that +is linear in A, Din, and Dout, which enables short proofs of properties such as summability. In +addition, Item 3 and Item 4 characterize SV approximation of A in terms of ε-approximation of +the “asymmetric lift” of A with respect to its symmetrization. This enables us to leverage results +developed for ε-approximation, such as preservation under Schur complements (Theorem A.2). +We next note the relation between SV and normalized SV approximation. +Lemma 3.8. Let A, �A ∈ Cm×n, and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices such that +ker(Din) ⊆ lker(A), and ker(Dout) ⊆ rker(A). Let N = D+/2 +in AD+/2 +out and ˜N = D+/2 +in +�AD+/2 +out Then +�A +Din,Dout +≈ε +A if and only if ˜N +svn +≈ ε N. +16 + +Note that if A is nonnegative and Din = diag(A⃗1n), Dout = diag(⃗1⊤ +mA), then N and ˜N are the +normalized adjacency matrices of A and �A. Furthermore, when A, �A are the adjacency matrices of +regular digraphs, then N and ˜N are the random-walk matrices of A and �A respectively. If instead +A and �A are the adjacency matrices of Eulerian digraphs, we obtain that N and ˜N are similar to +the random-walk matrices of A and �A. +Normalized SV approximation is implied by standard ǫ/2-approximation with respect to the +original degree matrix holding for all unitary multiples of the adjacency matrix. +Theorem 3.9 (Unitary transformation characterization of SV-approximation). For ˜N, N ∈ Cn×n, +we have that ˜N +svn +≈ ε N if for every pair of unitary matrices U, V, U ˜NV is a standard ε/2- +approximation of UNV with respect to degree matrix I. +Moreover, if ˜N +svn +≈ ε N then for every +pair of matrices U, V satisfying ∥U∥ ≤ 1, ∥V∥ ≤ 1, we have that U ˜NV +svn +≈ ε UNV, and hence +U ˜NV is a standard ε-approximation of UNV with respect to degree matrix I. +An important corollary is that SV approximation between adjacency matrices is preserved under +multiplication on each side of the adjacency matrix by (possibly different) permutations: +Corollary 3.10 (SV preservation under multiplication by permutation matrices). Let A, �A ∈ +Cm×n +≥0 +and suppose �A +Din,Dout +≈ε +A. Let U, V be arbitrary permutation matrices. Then U �AV +D′ +in,D′ +out +≈ε +UAV where D′ +in = UDinU∗ and D′ +out = V∗DoutV. Consequently, if �A +sv≈ε A, then U�AV +sv≈ε UAV +and if �A +svn +≈ ε A then U �AV +svn +≈ ε UAV. +An equivalent property was not known and is in fact not true for prior notions such as unit +circle, as otherwise it would show SV and UC approximation are equivalent via Item 4. +3.3 +Properties of Singular Value Approximation +Here we provide several useful properties of SV approximation. +First, we show that SV approximation implies unit-circle approximation: +Lemma 3.11. If A, �A ∈ Cn×n with �A +D≈ε A then �A +◦≈ε A with respect to degree matrix D. +Interestingly, even for 2×2 symmetric matrices, SV approximation can be separated arbitrarily +from UC approximation. +Proposition 3.12. For every α, ε ∈ (0, 1), there are symmetric matrices W, � +W with ∥W∥, ∥� +W∥ ≤ +1 such that � +W +◦≈ε W but � +W is not an ε SV-approximation of W with respect to I − W for any +ε′ < +ε +√ +1−α2 . +In prior work, ([AKM+20, Proposition 4.1]), it was also shown that UC approximation can be +arbitrary separated from standard approximation, even for undirected graphs. +Proposition 3.13 ([AKM+20]). For every ǫ ∈ (0, 1), there exist undirected regular graphs with +random walk matrices � +W, W such that � +W is an ε-approximation of W with respect to I − W but +� +W +◦≈c W does not hold for every c ∈ N. +Further, since UC approximation trivially implies standard approximation (as was also argued +in [AKM+20]) we see that SV approximation can be viewed of a strengthening of both UC and +standard approximation. +17 + +A corollary of Corollary 3.10 is that SV approximation is preserved under embedding in a +block matrix. In particular, it shows the stronger fact that approximation is preserved even if we +use a different block structure for rows and columns (i.e. we are not embedding into principal +submatrices). +In contrast, unit circle approximation is only known to be preserved when the +embedding pattern is a block pattern given by the directed cycle; i.e., tensoring the adjacency matrix +with that of the directed cycle. In fact, the separations we prove imply unit circle approximation +cannot be preserved under all block embeddings, as otherwise it would imply SV approximation by +Item 3 of Lemma 3.7. +Lemma 3.14 (SV preservation under arbitrary lifting). Let A, �A ∈ Cm×n be matrices such that +�A +Din,Dout +≈ε +A. Then for all integers i, j, k, ℓ ≥ 0 + + +0i×j +0i×n +0i×k +0m×j +�A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + +D′ +in,D′ +out +≈ε + + +0i×j +0i×n +0i×k +0m×j +A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + . +Where +D′ +in = + + +0i×i +0i×n +0i×ℓ +0n×i +Din +0n×ℓ +0ℓ×i +0ℓ×n +0ℓ×ℓ + + , +D′ +out = + + +0j×j +0j×m +0j×k +0m×j +Din +0m×k +0k×j +0k×m +0k×k + + . +Consequently, if �A +sv≈ε A then + + +0i×j +0i×n +0i×k +0m×j +�A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + sv≈ε + + +0i×j +0i×n +0i×k +0m×j +A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + . +By taking sums of different such liftings, one can obtain approximation for arbitrary tensorings: +if �A +sv≈ε A, for any M ∈ {0, 1}i×j, we have �A⊗M +sv≈ε A⊗M. A special case of this is concatenation: +Corollary 3.15 (SV preservation under concatenation). Let A1, A2 �A1, �A2 ∈ Cm×n be matrices +such that �A1 +sv≈ε A1 and �A2 +sv≈ε A2 then +� +�A1 +�A2 +� sv≈ε +� +A1 +A2 +� +. +Another corollary is that SV approximation is preserved under products of different walk ma- +trices with essentially no loss in the approximation quality. In contrast, the closest property known +to be achieved by definitions of approximation considered in prior work is that unit circle approxi- +mation is preserved with no loss under only powers of the same walk matrix. +Lemma 3.16 (SV preservation under products). Let (Ni)i∈[ℓ], ( ˜Ni)i∈[ℓ] ∈ Cn×n be such that for +every i, ˜Ni +svn +≈ ε Ni. Then ˜Nℓ · · · ˜N2 ˜N1 +svn +≈ ε+O(ǫ2) Nℓ · · · N2N1. +A useful application of this result is to sparsifying powers of a walk matrix of an Eulerian +digraph (Theorem 4.17). +SV approximation between digraphs requires exact preservation of the degrees: +Lemma 3.17. If �A +sv≈ε A, then �A⃗1 = A⃗1 and �A⊤⃗1 = A⊤⃗1. +SV sparsification is additive. We note that the analgous statement for normalized SV approxi- +mation is that it is preserved under convex combinations. +18 + +Lemma 3.18. If �Ai +(Din)i,(Dout)i +≈ε +Ai for all i ∈ [k], letting Din +def += � +i∈[k](Din)i and Dout +def += +� +i∈[k](Dout)i, then � +i∈[k] �Ai +Din,Dout +≈ε +� +i∈[k] Ai. +Consequently, if �Ai +sv≈ε Ai for every i, then +� +i∈[k] �Ai +sv≈ε +� +i∈[k] Ai. +SV approximation satisfies an approximate triangle inequality: +Lemma 3.19. If A3 +Din,Dout +≈δ +A2 and A2 +Din,Dout +≈ε +A1 then A3 +Din,Dout +≈ε+δ+εδ A1. Consequently, if A3 +sv≈δ A2 +and A2 +sv≈εA1 then A3 +sv≈ε+δ+εδA1, and if A3 +svn +≈ δ A2 and A2 +svn +≈ εA1 then A3 +svn +≈ ε+δ+εδA1. Moreover, +if for δ ∈ (0, 1/2) and A0, . . . , Aℓ we have Ai +sv≈δ/2ℓ Ai−1 for every i, then Aℓ +sv≈δ A0. Moreover, the +equivalent claim holds for normalized SV approximation. +A regular undirected graph SV approximates the complete graph with error equal to its expan- +sion. +Lemma 3.20. Let G be a strongly connected, d-regular directed multigraph on n vertices with +adjacency matrix A and let J ∈ Rn×n be a matrix with 1/n in every entry (i.e. J is the walk +matrix of the complete graph with a self loop on every vertex). Then λ(G) ≤ 1 − λ/2 if and only if +A/d +svn +≈ λ J. +4 +Sparsification +Given the Laplacian of an Eulerian directed graph, we wish to compute a sparsifier with respect +to SV approximation. In this section, we show how to solve a more general problem of sparsi- +fiying a non-negative rectangular matrix A ∈ Rm×n +≥0 +with respect to SV approximation. Further, +in Section 4.5 we show how to construct square sparsifiers via the derandomized square, and in +Section 4.6 we show how to sparsify powers of walk matrices. We first give an informal overview +of our approach. +We give a series of reductions that reduce this problem to degree- and bipartition-preserving +sparsification of expander graphs with respect to ε-approximation. By Lemma 3.7 (Item 2), to +SV-sparsify (possibly directed) A, it suffices to SV-sparsity the undirected bipartite lift +�0 +A⊤ +A +0 +� +with respect to SV approximation. In particular, given a symmetric matrix satisfying slift +� +�A +� sv≈ε +slift (A), we have by Lemma 3.7 that �A +sv≈ε A. Thus, any symmetric bipartition-preserving SV +sparsifier of undirected graphs immediately gives an SV sparsifier for Eulerian digraphs. In fact, +every SV sparsifier of undirected graphs must be bipartition-preserving, and an asymmetric spar- +sifier can be used to read off sparsifiers of the underlying directed graphs. For a formal statement +of this stronger claim, see Lemma A.5. +We show (Lemma 4.3) that for undirected bipartite expanders, degree- and bipartition-preserving +approximation with respect to D implies SV approximation. +From this, we can construct SV sparsifiers (Theorem 4.11) by computing an expander decom- +position of the bipartite lift, then sparsifying the relevant subgraphs in a suitable way. While we +can use existing techniques to obtain this (with worse log factors), we also develop a new sparsifica- +tion approach based on cycle decompositions. We show that in a bipartite expander, decomposing +edges indicent to high-degree vertices into edge-disjoint cycles, then taking the odd or even edges +19 + +of each cycle with probability 1/2 produces a sparsifier with high probability. This sparsification +procedure automatically preserves degrees and bipartitions, which we require to lift approximation +with respect to D to SV approximation. +Prior work [CGP+18] showed an equivalent result in +general Eulerian digraphs, except they required the cycles to be short, resulting in almost-linear +runtime [CGP+18,PY19]. +For convenience, we define notation for the adjacency and degree matrices of graphs: +Definition 4.1. For the remainder of the section, for a weighted undirected graph G we let AG +denote the adjacency matrix of G and DG +def += AG⃗1 denote the degree matrix. As such, an ε-SV +sparsifier of G is a graph H such that AH +sv≈ε AG. +First, we show that for undirected expanders, degree- and bipartition-preserving sparsification +with respect to the degree matrix implies sparsification with respect to SV approximation. +Definition 4.2 (Bipartiteness). We say A ∈ Rn×n +≥0 +is bipartite with bipartition S, T ⊆ [n] if S +and T partition V and AS,S = 0S×S and AT,T = 0T×T . Given a graph G with adjacency matrix +A, we say that a sparsifier H is bipartition-preserving if for every bipartition S, T in G, we have +HS,S = 0S×S and HT,T = 0T×T (where the statement is vacuously satisfied if G is not bipartite). +In the following lemma we show that for non-negative matrices that are bipartite with the same +bipartition, SV approximation is implied by approximation with respect to the error matrix D +(Definition 3.1), up to a loss in the condition number. +Lemma 4.3 (From diagonal approximation to SV approximation). Suppose that symmetric �A ∈ +Rn×n +≥0 +is a degree and bipartition-preserving ǫ/2-approximation of symmetric bipartite A ∈ Rn×n +≥0 +with respect to error matrix D +def += diag(A⃗1) = diag(A⊤⃗1). +Then �A +sv≈γ A for γ +def += +ε +λ2 where +λ +def += 1 − λ2(D+/2AD+/2). +Proof. Recall by Lemma 3.7 that it suffices to show that �A is an ε/2-approximation of A with +respect to E +def += D − AD+A. First, we have +E ⪰ (D − A)D+(D + A) ⪰ 0. +Furthermore by Lemma 3.2 and the assumption that �A is an ε/2-approximation of A with respect +to D we have +−ε/2 · D ⪯ A − �A ⪯ ε/2 · D. +By assumption that A has expansion λ and is bipartite (and hence its spectrum is symmetric), we +have +(D − A)+ ⪯ 1 +λD+, +(D + A)+ ⪯ 1 +λD+. +Furthermore, as A is the adjacency matrix of an undirected graph the only singular values of A of +magnitude 1 can be −1 and 1. Thus since �A matches degrees exactly and preserves bipartitions +we have that ker(A − �A) ⊇ ker(E). Thus we have +���E+/2(A − �A)E+/2��� ≤ +���(D + A)+/2D1/2(D − A)+/2(A − �A)(D − A)+/2D1/2(D + A)+/2��� +≤ 1 +λ +���D+/2D1/2(D − A)+/2(A − �A)(D − A)+/2D1/2D+/2��� +≤ 1 +λ +���(D − A)+/2(A − �A)(D − A)+/2��� +≤ 1 +λ2 +���D+/2(A − �A)D+/2��� ≤ ε +λ2 . +20 + +4.1 +Cycle and Expander Decompositions +We give a new analysis of degree-preserving expander sparsification. +Chu, Gao, Peng, Sachdeva, Sawlani, and Wang [CGP+18] constructed sparsifiers for Eulerian +digraphs, which we now discuss. They decomposed the low-importance edges of a graph into a +series of edge-disjoint cycles C1, . . . , Ct. +Then for each cycle, they chose a random orientation +(clockwise or counterclockwise) and kept every edge in this orientation, and upweighted these by a +factor of two. One can see that this procedure gives a subgraph and maintains Eulerianness. They +bounded the impact of a single such sample in terms of the length of the cycle, and by giving a +sophisticated algorithm for decomposing a graph into many short cycles concluded an almost-linear +time sparsifier for standard approximation. +We follow this approach but make two modifications. First, we require the cycles to be Forward- +Backward (FB), where we always alternate between clockwise and counterclockwise edges (and all +cycles are of even length). This ensures that we exactly preserve degrees, not just Eulerianness, +which is essential for SV approximation. +We take advantage of the reduction from directed to +undirected SV sparsification to perform this decomposition in the undirected bipartite lift, where +undirected cycles correspond to alternating cycles in the original graph. Our second modification +is to prove a bound on the importance of the cycles that is independent of their length, as long as +the cycles lie inside an expander. This enables us to use a simple linear time algorithm for finding +the decomposition. +We first recall the standard procedure that enables us to efficiently decompose an undirected +graph into a union of (potentially long) FB cycles. +Lemma 4.4. Given an unweighted undirected graph G = (V, E), the algorithm CycleDecomp re- +turns a collection of edge-disjoint cycles C1, . . . , CT ⊆ E in time O(|E| + |V |) such that at most n +edges are not contained in some cycle. +Algorithm 1: CycleDecomp (G = (V, E)), G undirected +1 Initialize Eex, C = {}. +2 for v ∈ V do +3 +while Greedily construct a non-backtracking path v = u0, u1, . . . , uk in an arbitrary +fashion and mark visited vertices. do +4 +if uk has no neighbors other than uk−1 then +5 +Add Eex ← Eex ∪ {(uk−1, uk)} and remove (uk−1, uk) from E. +6 +Continue search from uk−1. +7 +end +8 +if uk = uk′ for some k′ < k − 1 then +9 +Let C ← C ∪ (uk′, uk′+1, . . . , uk−1) and let +E ← E \ {(uk′, uk′+1), (uk′+1, uk′+2), . . . , (uk−1, uk)}. +10 +Continue search from uk. +11 +end +12 +end +13 end +14 return C, Eex +Proof. The runtime and the fact that the cycles are edge disjoint is direct from the description. +21 + +Then note that for every edge (u, v) that we add to Eex (and remove from E) we must create an +isolated vertex v. But this can happen at most n times, so we are done. +Furthermore, we recall the definition of an expander partition: +Definition 4.5. Given φ, γ and an undirected graph G, a (φ, γ)-expander partition of G is a +partition V1, . . . , Vt such that at most γ · m edges cross between partitions and for every i ∈ [t], +λ(G[Vi]) ≤ 1 − φ. +We will use algorithms for expander decomposition as a key subroutine in our sparsifier. We re- +quire a small constant fraction of edges to lie between clusters, and thus state the algorithms fixing +γ = 1/16 such that m/16 edges cross subgraphs. In particular, given an algorithm ExpanderDe- +comp computing a (Φ, m/16) expander decomposition, let T(ExpanderDecomp, m) be its run- +time (on graphs with m edges) and Φ(ExpanderDecomp) = Φ. We recall the best known nearly- +linear expander partitioning algorithm, as well as the (optimal) existential result: +Theorem 4.6 ([ADK22]). There is an algorithm ExpanderDecomp(G) running in randomized +time T(ExpanderDecomp, m) = O(m log7 m) that computes a (Φ = log−4 m, 1/16) expander +decomposition of G with high probability. +Theorem 4.7 ([ST04]). There exists a (Φ = log−2 m, 1/16) expander decomposition of G. +4.2 +Sparsifying Bipartite Expanders +The core of our algorithm is a way to sparsify bipartite, unweighted expanders in a degree- and +bipartition- preserving way. +The algorithm decomposes the graph into a series of edge-disjoint cycles, then takes either +twice the odd or twice the even edges with probability 1/2. In the case that the undirected graph +corresponds to the bipartite lift of an Eulerian digraph G, this precisely corresponds to decomposing +G into a union of Forward-Backward cycles and taking the forward or backward edges for each cycle. +We bound the error in terms of the degree matrix D. This is tolerable as we apply this procedure +inside expanders, so by Lemma 4.3 we can transform this into an SV approximation guarantee with +polylog(n) loss in approximation quality. +Lemma 4.8. There is a constant c > 0 such that given δ ∈ (0, 1) and an unweighted, undirected +bipartite graph G with m edges, SparsifyCycleUnweighted returns in time O(m) a graph H +with edge weights in {1, 2} such that with high probability AH is a δ-approximation of AG with re- +spect to error matrix DG, and H has at most max{cnδ−2 log n, (7/8)m} edges and exactly preserves +degrees and bipartiteness. +To prove correctness, we recall a result on matrix concentration of Tropp: +Lemma 4.9 ([Tro12]). Let {Xi} be d × d random PSD matrices such that E[� +i Xi] ⪯ X. For +δ ≤ 1, +Pr +�� +i +Xi − E +�� +i +Xi +� +⪰ δX +� +≤ 1−d·eδ2/3R, +Pr +�� +i +Xi − E +�� +i +Xi +� +⪯ −δX +� +≤ 1−d·eδ2/3R, +provided that R ≥ +��X+/2XiX+/2�� almost surely. +Using this matrix concentration result, we can prove the theorem. +22 + +Algorithm 2: SparsifyCycleUnweighted (δ, G = (V, E)) +1 if δ < +� +cn log n/m then +2 +return G +3 end +4 Let dlow +def += m +2n. +5 Let Elow be the edges adjacent to vertices v with d(v) ≤ dlow. +6 Let Eex, (C1, . . . , Ct) =CycleDecomp (V, E \ Elow). +7 Initialize H = (V, Eex ∪ E \ Elow). +8 for i = 1, . . . , t do +9 +Let (a1, b1), (b1, a2), . . . , (bk, a1) = Ci. +10 +Choose b ← 0, 1 uniformly at random. +11 +if b = 0 then +12 +H ← H ∪ {2 · (ai, bi)}i∈[k] // Add with weight 2 +13 +end +14 +else H ← H ∪ {2 · (bi, ai+1)}i∈[k] ; +15 end +16 return H +Proof of Lemma 4.8. First, note that if we do not sparsify G at all we have cnδ−2 log n < m and +hence the sparsity requirement is already satisfied. Subsequently we assume this is not the case. +Recall Elow is the set of edges adjacent to a vertex with degree at most dlow = (m/2n), and +observe that there are at most dlow · n ≤ m/2 such edges. Further recall C1, . . . , Ct are the cycles +found by CycleDecomp applied to E \ Elow. By assumption that G is bipartite, all such cycles +are of even length. +Let Xi +def += �Di − �Ai be the random variable of the Laplacian of the ith sampled cycle and observe +0 ⪯ Xi ⪯ 2ICi. Furthermore, let X +def += DG ⪰ 0. We have +E +�� +i +Xi +� += +⊤ +� +i=1 +LCi ⪯ X. +Furthermore, for every i ∈ [t] we have +���X+/2XiX+/2��� ≤ 2 +���D+/2ICiD+/2��� ≤ 4n +m +where the final step follows from all cycles being exclusively incident to vertices with degree at +least m/2n. Thus, we apply Lemma 4.9 with {Xi}i∈[t], R = 4n/m, X = X and δ = δ/c. Letting +DC − AC be the random graph obtained by the cycle sampling procedure, with high probability +we have +−δDG ⪯ (DC − AC) − +t +� +i=1 +(DCi − ACi) ⪯ δDG. +By adding the non-random components to we obtain +−δDG ⪯ AH − AG ⪯ δDG +and thus AH is an δ-approximation of AG with respect to DG with high probability. +Furthermore, by Lemma 4.4 at least m/2−n ≥ m/4 edges are included in the cycles C1, . . . , Ct, +and hence H has at most 7m/8 edges under the assumption that m ≥ cn, and H preserves degrees +and bipartiteness by construction. +23 + +4.3 +Sparsifying a Constant Fraction of Edges +We then construct our routine for sparsifying a graph with respect to SV approximation. At each +iteration, we decrease the number of edges by a constant factor. We first bucket edges into powers +of two, then apply SparsifyCycleUnweighted . +Lemma 4.10. Given ε > 0 and b ∈ N and an undirected bipartite graph G with m edges, each with +weight in {1, 2, 22, . . . , 2b}, SparsifyCycle returns a graph H with +O +� +b · nε−2Φ−4 log n +� ++ 15 +16m +edges where Φ = Φ(ExpanderDecomp), each with weight in {1, 2, 22, . . . , 2b+1}, such that with +high probability AH +sv≈ε AG. Moreover, SparsifyCycle runs in randomized time O(m + bn + +T(ExpanderDecomp, m))). +Algorithm 3: SparsifyCycle (ε, b, G = (V, E)) +1 Let δ = ε−1Φ2. +2 Initialize H = (V, ∅). +3 for i = 0, . . . , b do +4 +Initialize (unweighted) Gi as G restricted to edges with weight 2i. +5 +Let Vi,1, . . . , Vi,t, Eex +def +=ExpanderDecomp(Gi). +6 +Let H ← H ∪ 2i · Eex. +7 +for j ∈ 1, . . . , t do +8 +Let Hi,j +def +=SparsifyCycleUnweighted (δ, Gi[Vi,j]) +9 +Let H ← H ∪ 2i · Hi,j. +10 +end +11 end +12 return H +Proof. Let G0, . . . , Gb be the unweighted bipartite graphs where Gi contains an edge (u, v) if there +is an equivalent edge with weight 2i. Thus, observe that +b +� +i=0 +2i · (DGi − AGi) = DG − AG +and it suffices to sparsify unweighted bipartite graphs. Furthermore let mi +def += |E(Gi)| and observe +� +i mi = m. +Next, for every i observe that ExpanderDecomp +computes a decomposition Vi,1, . . . , Vi,t +such that letting Gi,j +def += Gi[Vi,j] we have λ(Gi,j) ≤ 1 − Φ, and at most mi/16 edges do not lie +in one of these subgraphs. Let ni,j +def += |Gi,j| and observe � +j ni,j ≤ n and let mi,j +def += |E(Gi,j)| +and observe � +j mi,j ≤ mi. By Lemma 4.8, we have that for every i, j, AHi,j is a degree and +bipartition-preserving δ-approximation of AGi,j with respect to DGi,j with high probability, and +thus by Lemma 4.3 +AHi,j +sv≈δ·Φ−2 AGi,j. +24 + +Thus by Lemma 3.18 we have +AH +sv≈δ·Φ−2 AG +and so by our choice of δ = εΦ2 we obtain the desired approximation. +We now analyze the number of edges. The number of edges not placed into a subgraph Gi,j is +at most +b +� +i=1 +mi +16 = m +16. +By Lemma 4.8 the number of edges in the sparsified graphs is at most +b +� +i=1 +t +� +j=1 +max{cni,jδ−2 log n, 7 +8mi,j} ≤ +b +� +i=1 +t +� +j=1 +� +cni,jδ−2 log n + 7 +8mi,j +� += O +� +bnε−2Φ−4 log n +� ++ 7 +8 +b +� +i=1 +t +� +j=1 +mi,j += O +� +bnε−2Φ−4 log n +� ++ 7 +8m +and hence the total number of edges is bounded as desired. Finally, we analyze the time. We +can compute every Gi with a single pass through the edge list, so the total work is bounded by +� +i +� +j O(mi,j + ni,j) + �b +i=1 T(ExpanderDecomp, mi) = O(m + bn + T(ExpanderDecomp, m)) +as claimed. +4.4 +Nearly Linear-Sized Sparsifers +We now apply our routine for sparsifying a constant fraction of edges in the natural recursive way. +Theorem 4.11. Given ε ∈ (0, 1) and an undirected bipartite graph G with m edges with integer +edge weights in [1, U], SparsifyGraph returns a graph H with +O +� +log(nU) · nε−2Φ−4 log3 n +� +edges where Φ = Φ(ExpanderDecomp), such that with high probability AH +sv≈ε AG. Moreover, +SparsifyGraph runs in time O(m + nb + T(ExpanderDecomp, m log(U))). +Proof. First observe that in Phase 1 of the algorithm, we obtain H0 such that H0 has at most mb +edges (recalling b = ⌈log(U)⌉) and each edge has weight {1, 2, 22, . . . , 2b}, and AH0 = AG. Then in +each phase we obtain Hj such that +AHj +sv≈ε/2 log n AHj−1 +with high probability by Lemma 4.10. Then applying Lemma 3.19 we obtain +AHt +sv≈ε AG +with high probability as desired. We next analyze the sparsity. Observe that the final edge weights +are bounded by 2b+t = poly(nU) and hence the final sparsity is +O +� +(b + t)nε−2Φ−4 log3 n +� += O +� +log(nU)nε−2Φ−4 log3 n +� +. +25 + +Algorithm 4: SparsifyGraph (ε, G = (V, E)) +1 Initialize H0 = (V, ∅). +2 Let b +def += ⌈log(U)⌉. +3 for i = 0, . . . , b do +4 +for (u, v) ∈ E do +5 +Let ⟨wG(u, v)⟩ be the binary expansion of wG(u, v). +6 +if ⟨wG(u, v)⟩i = 1 then Add (u, v) to H0 with weight 2i.; +7 +end +8 end +9 for j = 1, . . . , t +def += O(log n) do +10 +Let Hj ←SparsifyCycle (ε/2 log n, b + j, Hj−1). +11 end +12 return Ht +Finally, we analyze the time complexity. +Our initial call has mb edges and thus runs in time +O(mb + bn + T(ExpanderDecomp, mb)) by Lemma 4.10, and all subsequent calls are to sparsify +graphs with a constant factor fewer edges, so the time is dominated by the initial term. +By plugging in the algorithmic (Theorem 4.6) and existential (Theorem 4.6) results for expander +partitions, we obtain our sparsifier routines: +Corollary 4.12. Given ε ∈ (0, 1) and an undirected bipartite graph G with m edges with integer +edge weights in [1, U], there is a randomized algorithm that returns a graph H with +O +� +log(nU) · nε−2 log19 n +� +edges such that with high probability AH +sv≈εAG. Moreover, the algorithm runs in time O(log(U)m log7 m). +Corollary 4.13. Given ε ∈ (0, 1) and an undirected bipartite graph G with m edges with integer +edge weights in [1, U], there is a graph H with +O +� +log(nU) · nε−2 log11 n +� +edges such that AH +sv≈ε AG. +4.5 +Derandomized Square Sparsification +Here we show that the derandomized square of Rozenman and Vadhan [RV05] gives an SV sparsi- +fication of the square graph. We first explain the derandomized square. Given a d-regular graph +G, the squaring operation can be though of as placing a d × d complete graph between the in- and +out-neighbors of every vertex v. The derandomized square replaces this complete graph with an +expander. As an expander approximates the complete graph, this sparsification operation produces +an approximate square. +We first formally define the derandomized square. +Definition 4.14. Given a d-regular graph G = ([n], E) with neighbor function ΓG : [n] × [d] → [n] +and c-regular graph H = ([d], E′) with neighbor function ΓH : [d] × [c] → [d], the derandomized +square of G with respect to H, denoted G s○H, is the dc-regular graph on n vertices with neighbor +function +ΓG s○H(v, (i, j)) = ΓG(ΓG(v, i), ΓH(i, j)). +26 + +Prior analyses of the derandomized square have shown that it improves expansion approximately +as well as the true square [RV05] and that it produces unit-circle square sparsifiers [AKM+20]. We +strengthen the latter conclusion to SV square sparsifiers. +Lemma 4.15. Let G = (V, E) be a d-regular directed multigraph with random walk matrix W. Let +H be a c-regular expander on d vertices with λ(H) ≤ 1 − ǫ and let � +W be the adjacency matrix of +G s○H. Then +� +W +svn +≈ ε W2. +Proof. Let WH be the random walk matrix corresponding to H, and J ∈ Rd×d be the random walk +matrix of a complete graph with self loops. By Lemmas 3.14 and 3.20 we have +In ⊗ WH +sv≈ε In ⊗ J. +(18) +Let F be an nd×nd edge-rotation permutation matrix such that F((v, j), (u, i)) = 1 if the ith edge +leaving u is the jth edge entering v. It is straight-forward to verify that +W2 = Q⊤F(I ⊗ J)FQ +and +� +W = Q⊤F(I ⊗ WH)FQ +where Q +def += I ⊗ 1d +√ +d is an nd × n matrix. +Now note that ∥F∥ ≤ 1 and ∥Q∥ ≤ 1 and thus +∥FQ∥, ∥Q⊤F∥ ≤ 1. Thus by Theorem 3.9, and Equation (18) we have +� +W +sv≈ǫ W2. +4.6 +Sparsification of Random Walk Polynomials +We give the first nearly-linear time algorithm for sparsifying random walk polynomials of Eulerian +digraphs with second singular value bounded away from 1. In particular, given an Eulerian digraph +W ∈ Rn×n +≥0 +with second singular value bounded away from 1, we compute a sparse graph such +that the walk matrix � +W approximates Wℓ with respect to SV approximation. For this result, we +crucially require that SV approximation is preserved under products, a feature that is not obtained +by prior notions of approximation. +Definition 4.16. Given a strongly connected Eulerian digraph G, we let σ(G) = σ2(D−1/2 +G +AGD−1/2) +and refer to this as the second normalized singular value of G. +Theorem 4.17. Given ℓ ∈ N and ε ∈ (1/poly(n), 1) and an Eulerian digraph G where σ(G) ≤ +1 − 1/poly(n) with m edges with integer edge weights in [1, poly(n)], SparsifyPower +returns +in time ˜O(m + nε−2 log3(ℓ)) an Eulerian digraph H with ˜O(nε−2 log2(ℓ)) edges with integer edge +weights in [1, poly(n, ℓ, 1/ε)] such that +D+/2 +H AHD+/2 +H +svn +≈ ε (D+/2 +G +AD+/2 +G +)ℓ. +This implies the following corollary. +Definition 4.18. Let A be the adjacency matrix of an Eulerian digraph with diagonal matrix of +degrees D and walk matrix W. The kth random walk power of A is defined as +D1/2(D−1/2AD−1/2)kD1/2 = WkD. +27 + +Corollary 4.19. Let G be an Eulerian digraph with poly(n) weak mixing time, adjacency matrix +A, diagonal degree matrix D, and walk matrix W. Then for any positive integer k ≤ poly(n) and +γ ≥ 1/poly(n), we can, in nearly linear time, output an ˜O(n/ε2) sized ε-SV-sparsifer of kth random +walk power of A′ = (1 − γ)A + γD. Furthermore, the walk matrix of the true kth random walk +power of A′ is given by ((1 − γ)W + γI)k so it is the k step γ-lazy random walk on G. +Proof. Immediate from Lemma B.5 and Theorem 4.17. +To prove these, we first recall the square sparsification algorithm implicit in the work of Cohen +et al. [CKP+17]. +Theorem 4.20 (Implicit in [CKP+17]). Given an Eulerian weighted digraph G with Laplacian +D − AG, let G2 be the graph with Laplacian D − AGD+AG. Let U denote the Laplacian of the +undirected graph corresponding to (AG2 + AT +G2)/2. Then SparsifySquare of [CKP+17] produces +in time ˜O(m), a weighted graph H with the following properties: +1. H has the same weighted in- and out- degree sequences as those of G2. +2. H has ˜O(m/ε2) nonzero edges. +3. AH is an ǫ/2-approximation of AGD−1AG with respect to SD−AGD+AG. +We can use a block embedding and Theorem 4.20 to obtain a nearly-linear time algorithm for +sparsifying products with respect to SV approximation. +Lemma 4.21. Given ε > 0 and Eulerian digraphs G1, G2 each with at most m edges with matching +degree matrix D, where each edge has integer weight in [1, U], SparsifyProduct +returns an +Eulerian digraph H with ˜O(m/ε2) edges such that with high probability +AH +sv≈ε AG2D+AG1 +Moreover, SparsifyProduct runs in time ˜O(m/ε2). +Algorithm 5: SparsifyProduct(ε, G1, G2) +1 Let G be the graph with AG = + + +0 +0 +0 +AG1 +0 +0 +0 +AG2 +0 + + +2 Let H ← [SparsifySquare(ε/3, G)]3,1. +3 return H +Proof. We first analyze correctness. Note by Theorem 4.20 we have that + + +0 +0 +0 +0 +0 +0 +AH0 +0 +0 + + is an ε/6-approximation of + + +0 +0 +0 +0 +0 +0 +AG2D+AG1 +0 +0 + + with respect to R +where +R = + + +D +0 +−(AG2D+AG1)⊤ +0 +0 +0 +−AG2D+AG1 +0 +D + + +This is precisely Item 3 of Lemma 3.7, so we conclude AH +sv≈ε AG2D+AG1. +28 + +Finally, we require an algorithm that removes very small edge weights. This is not trivial, as +we must do so in a way that exactly preserves degrees for SV approximation to hold. +Lemma 4.22. There is an algorithm FixEdgeWeights such that the following holds. Let G be +an Eulerian digraph with m edges where the degree matrix DG can be represented as a subset sum +of {1, . . . , 2b}. Then for t ∈ N, FixEdgeWeights(t, G) outputs in time O(mb) a graph H such +that: +• Ht has m + 2n edges and is Eulerian and DH = DG. +• Edge weights in H can be represented as a subset sum of {2−t, . . . , 1, . . . , 2b}. +• ∥AH − AG∥ ≤ 2n · 2−t. +Proof. First, let ⌊G⌋ be the graph where all edge weights are rounded down to a multiple of 2−t. +Observe that this satisfies +∥A⌊G⌋ − AG∥ ≤ n · 2−t. +Furthermore, note that each vertex v has an in-degree shortfall equal to sin(v) = Dv,v−� +u(A⌊G⌋)v,u. +Note that sin(v) ≥ 0 since we only remove weight from edges. This can be expressed as a sub- +set sum of {2−t, . . . , 1, . . . , 2b} as all numbers in the summand can be. An equivalent statement +holds for the out-degree shortfall. Furthermore, the total in- and out- degree shortfalls must be +equal, since every truncation contributes to both sums. Thus, by greedily matching in- shortfalls +to out- shortfalls, we obtain the desired graph H. It is easy to see this satisfies the approximation +claim. +We are now prepared to state our algorithm for sparsifying an arbitrary power of an Eulerian +digraph. +Algorithm 6: SparsifyPower(ε, G, ℓ) +1 Let t +def += ⌈log ℓ⌉. +2 Let ⟨ℓ⟩ be the binary expansion of ℓ. +3 Let l +def += O(log(nℓ/ε)). +4 Let G1 +def += SparsifyGraph (ε/2, G). +5 for i = 2, . . . , t do +6 +Let Pi +def += SparsifyProduct(ε/t, Gi−1, Gi−1). +7 +Let Hi +def += FixEdgeWeights(l, Pi). +8 +Let Gi +def += SparsifyGraph (ε/t, Hi). +9 +if ⟨ℓ⟩t−i = 1 then +10 +Let Pi ← SparsifyProduct(ε/t, Gi−1, G1). +11 +Let Hi ← FixEdgeWeights(l, Pi). +12 +Let Gi ← SparsifyGraph (ε/t, Hi). +13 +end +14 end +15 return Gt +We are now prepared to prove the result: +29 + +Theorem 4.17. Given ℓ ∈ N and ε ∈ (1/poly(n), 1) and an Eulerian digraph G where σ(G) ≤ +1 − 1/poly(n) with m edges with integer edge weights in [1, poly(n)], SparsifyPower +returns +in time ˜O(m + nε−2 log3(ℓ)) an Eulerian digraph H with ˜O(nε−2 log2(ℓ)) edges with integer edge +weights in [1, poly(n, ℓ, 1/ε)] such that +D+/2 +H AHD+/2 +H +svn +≈ ε (D+/2 +G +AD+/2 +G +)ℓ. +Proof. We first show the approximation guarantee. Note that by the guarantees of Lemma 4.21 +and Corollary 4.12 and Lemma 4.22, for every i, +D +def += DPi = DHi = DGi = DG. +Let NPi +def += D+/2A+/2 +Pi D+/2 and NHi +def += D+/2A+/2 +Hi D+/2 and NGi +def += D+/2A+/2 +Gi D+/2. The proof +of correctness does not change if we assume ℓ = 2t for some t (and hence the if condition in the +main loop never occurs), so we analyze it assuming this for simplicity. The fact that this does +not matter follows from the fact that SV approximation is preserved under arbitrary products, not +merely powers (Lemma 3.16). +We first argue that the singular value of all graphs we sparsify remain bounded below 1. +Claim 4.23. For every i, σ(Hi), σ(Pi), σ(Gi) ≤ 1 − 1/poly(n). +Proof. We show the claim via induction. +By assumption, we have σ(G) = 1 − 1/nc for some +constant c. By Lemma B.4 we have σ(G1) ≤ 1−(1+ε/t)/nc. Subsequently, assume that σ(Gi−1) ≤ +1 − (1 + 2ε/t)2(i−1)/nc. +• By Lemma B.1 we have σ(G2 +i−1) ≤ σ(Gi−1) ≤ 1 − (1 + 2ε/t)2(i−1)/nc. +• By Lemma B.4 we have σ(Pi) ≤ 1 − (1 + ε/t)(1 + 2ε/t)2(i−1)/nc. +• By Lemma B.3 we have σ(Hi) ≤ 1 − (1 + ε/t)(1 + 2ε/t)2(i−1)/nc + 1/n10. +• By Lemma B.4 we have σ(Gi) ≤ 1 − (1 + 2ε/t)2i/nc +So the claim follows assuming ε ≤ 1/c′ for constant c′. +We now show that SV approximation is preserved at every step of the loop. Fixing i and Gi−1: +1. By Lemma 4.21 we have that NPi +svn +≈ ε/3t N2 +Gi−1. +2. By Lemma 4.22 we have ∥AHi −APi∥ ≤ 2n2−l, so AHi is a 2n2−l ≤ ε/poly(tn) approximation +of APi with respect to D. Taking the bipartite lift, +� 0 +A⊤ +Hi +AHi +0 +� +is a degree- and bipartition- +preserving ε/poly(tn)-approximation of +� 0 +A⊤ +Pi +APi +0 +� +with respect to +�D +0 +0 +D +� +(where we use +that Pi is not itself bipartite to ensure that the lift of Hi preserves all bipartitions). Thus by +Lemma 4.3 (using that the bipartite lift has normalized second eigenvalue at least 1/poly(n) +by Claim 4.23) and Lemma 3.7, we have that NHi +sv≈ε/3t NPi. +3. By Corollary 4.12, NGi +svn +≈ ε/3t NHi. +30 + +Applying Lemma 3.19 we obtain NGi +svn +≈ ε/t N2 +Gi−1 for every i. Therefore by Lemma 3.16, +NGt +svn +≈ ε/t +� +NGi−1 +�2 svn +≈ ε/t . . . (NG2)2t−1 svn +≈ ε/t (NG1)2t +and so we conclude by Lemma 3.19 (and taking ε ← ε/10). +To analyze the time, note that by Lemma 4.22 we always call SparsifyGraph on graphs with +(up to rescaling) integer edge weights in [1, poly(nℓ/ε)], and by Theorem 4.20 Pi (and thus Hi) has +˜O(nε−2 log2 ℓ) edges, so we obtain the desired runtime and sparsity by Corollary 4.12. +5 +Squaring algorithm for solving normal directed Laplacians +In [PS14], Peng and Speilman proposed a squaring-based parallel algorithm for computing an +approximate pseudo-inverse (preconditioner) of a Laplacian matrix. +A key ingredient of their +algorithm is the following recursion, which we refer to as the Peng-Spielman squaring recursion +(PS-recursion), for inverting I − W when ∥W∥ < 1.6 +(I − W)−1 = 1 +2 +� +I + (I + W)(I − W2)−1(I + W) +� +(19) +Leveraging the PS-recursion a natural appproach to find an approximate inverse of I − W, +is to compute a sparse matrix � +W such that I − � +W ≈ I − W2, and then compute the inverse of +I− � +W recursively. [PS14] showed that for symmetric Laplacians, using spectral approximation this +recursion can be leveraged to produce a constant preconditioner with logarithmic depth. +Here we prove that the same algorithm gives approximate pre-conditioner for I − W if W is +normal but not necessarily symmetric and the square approximations are normalized SV approx- +imations. +We think Theorem 5.2 gives hope that a similar approach can expand the proof to +non-normal matrices. If proved, this will result in simple but space and time efficient algorithms +for solving directed Laplacian systems. For example, this could simplify the space efficient solver +introduced in [AKM+20] which in turn is used to estimate random walk probabilities to a high +precision. +Below we give the main theorem which formally defines the PS-recursion and states the main +result of this section. This theorem uses a generalization of the spectral norm to a (semi-)norm +induced by a PSD matrix, which we first define. +Definition 5.1. Given PSD F ∈ Cn×n, we let ∥ · ∥F be the (semi)norm on vectors where ∥x∥F +def += +√ +x∗Fx, and define the matrix norm ∥A∥F +def += supx∈Cn\{⃗0} +∥Ax∥F +∥x∥F . +Theorem 5.2. For a normal matrix W ∈ Cn×n with ∥W∥ ≤ 1, let W = W0, . . . , Wk−1 be a +sequence of matrices such that for all i ∈ [k − 1] we have Wi +svn +≈ ǫ W2 +i−1 for ε ≤ 1/(4k). Define +Pi +def += 1 +2 [I + (I + Wi)Pi+1(I + Wi)] +for all +0 ≤ i < k +where Pk is a matrix such that +���(I − W2k) +1 +2 +� +Pk − (I − W2k)+� +(I − W2k) +1 +2 +��� ≤ O(kǫ) . +Then for B +def += ((I − W)1/2)∗(I − W)1/2 +∥P0(I − W) − I∥B = +���(I − W) +1 +2 � +P0 − (I − W)+� +(I − W) +1 +2 +��� ≤ O(k2ǫ) . +(20) +6Technically the recursion was introduced for symmetric W. +31 + +Note that in the above statement, the square root of I − W is well defined by the assumption +that W is normal. In the case of non-normal matrices, which is outside the scope of this paper, +one can consider the Maclaurin series for (1 − z)1/2 to work with (I − W)1/2. +We do not give a full algorithm for solving systems defined by Theorem 5.2 as the main purpose +is to show that the error analysis of the squaring algorithm can be expanded to normal asymmetric +matrices, and we already know space and time efficient algorithms [CKK+18,AKM+20] for solving +a larger class of matrices which includes non-normal W’s. However, one can still get an actual +squaring based solver from Theorem 5.2. For example, for a normal W corresponding to a random +walk matrix of an Eulerian digraph one can carry the same steps in [AKM+20] by setting k = +O(log n) and Pk = I − ⃗1⃗1⊤ +n +to get an actual squaring based solver using Theorem 5.2 and the +derandomized square sparsification described in Section 4.5. +We prove the statement of the theorem, following a few helper lemmas. First, we show the error +of Wi’s approximating W2i’s grows additively. +Lemma 5.3. Let Wi be defined as in Theorem 5.2. Then Wi +svn +≈ 2iǫ W2i for all for i ∈ [k]. +Proof. Recall that Wj +svn +≈ ε W2 +j−1 for every j. Thus, by Lemma 3.16 we have for every j ≤ i, +(Wj)2i−j svn +≈ ε+O(ε2) (Wj−1)2i−j+1 . +The result then follows from Lemma 3.19. +Next we give a general technical tool regarding normal matrices. +Lemma 5.4. For a normal matrix V with ∥V∥ ≤ 1 +∥(I + V) +1 +2 ∥ ≤ +√ +2 +(21) +and if �V +svn +≈ δ V then +���(I − V) +1 +2( �V − V)((I − V2)+) +1 +2 +��� ≤ O(δ). +(22) +Proof. Note that since V is normal, we can write it as V = UDU∗ where U is unitary and D is +diagonal and ∥D∥ ≤ 1. Then it is easy to see +∥(I + V) +1 +2 ∥ = ∥U(I + D) +1 +2 U∗∥ = ∥(I + D) +1 +2 ∥ ≤ +√ +2. +(23) +To prove (22), given the SV approximation definition, for unit vectors x, y we have, +���x∗(I − V) +1 +2( �V − V)((I − V2)+) +1 +2 y +��� ≤ δ +2 · x∗(I − V) +1 +2(I − VV∗)(I − V∗) +1 +2 x ++ δ +2 · y∗((I − (V∗)2)+) +1 +2(I − V∗V)((I − V2)+) +1 +2 y +We bound each term on the right hand side separately. Since V is normal we can prove bounds on +the norm of (I − V) +1 +2 (I − VV∗)(I − V∗) +1 +2 and ((I − (V∗)2)+) +1 +2 (I − V∗V)((I − V2)+) +1 +2 by looking +at complex numbers. Note that for z ∈ C and |z| ≤ 1, +|(1 − z) +1 +2(1 − z · z∗)2(1 − z∗) +1 +2| = |1 − z| · (1 − |z|2) ≤ 2, +thus ∥(I − V) +1 +2 (I − VV∗)(I − V∗) +1 +2∥ ≤ 2. Next, +���(1 − z∗2)− 1 +2(1 − z · z∗)2(1 − z2)− 1 +2 +��� = 1 − |z|2 +|1 − z2| ≤ 1 +32 + +because 1 − |z|2 ≤ |1 − z2| by triangle inequality. Thus, +∥((I − (V∗)2)+) +1 +2(I − V∗V)((I − V2)+) +1 +2 ∥ ≤ 1. +Completing the proof that, +���x∗(I − V) +1 +2 ( �V − V)((I − V2)+) +1 +2y +��� ≤ 3 +2 · δ. +Leveraging these tools we give the central lemma towards proving Theorem 5.2 and conclude +by proving Theorem 5.2. +Lemma 5.5. For ǫ and Pi’s defined as in Theorem 5.2, +���(I − W2i) +1 +2 +� +Pi − (I − W2i)+� +(I − W2i) +1 +2 +��� ≤ ǫi +(24) +where ǫi = (1 + O(i · ǫ))ǫi+1 + O(i · ǫ) for 0 ≤ i < k and ǫk = O(k · ǫ). +Proof. We prove the statement by backward induction. Note that the statement is true for i = k +by the assumption about Pk in Theorem 5.2. Let +Qi = 1 +2 +� +I + (I + Wi)(I − W2i+1)+(I + Wi) +� +, +Ri = 1 +2 +� +I + (I + Wi)(I − W2i+1)+(I + W2i) +� +, +Si = 1 +2 +� +I + (I + W2i)(I − W2i+1)+(I + W2i) +� += (I − W2i)+. +Then by triangle inequality we have, +∥(I − W2i) +1 +2 +� +Pi − (I − W2i)+� +(I − W2i) +1 +2 ∥ ≤ ∥(I − W2i) +1 +2 (Pi − Qi)(I − W2i) +1 +2∥ +(25) ++ ∥(I − W2i) +1 +2 (Qi − Ri)(I − W2i) +1 +2∥ +(26) ++ ∥(I − W2i) +1 +2 (Ri − Si)(I − W2i) +1 +2 ∥. +(27) +Next, we bound each term separately. For (25), we have +���(I − W2i) +1 +2(Pi − Qi)(I − W2i) +1 +2 +��� += 1 +2 +���(I − W2i) +1 +2(I + Wi) +� +Pi+1 − (I − W2i+1)+� +(I + Wi)(I − W2i) +1 +2 +��� +≤ 1 +2 +���(I − W2i) +1 +2(I + Wi)((I − W2i+1) +1 +2 )+��� +· +���(I − W2i+1) +1 +2 +� +Pi+1 − (I − W2i+1)+� +(I − W2i+1) +1 +2 +��� +· +���((I − W2i+1) +1 +2 )+(I + Wi)(I − W2i) +1 +2 +��� . +Note that the middle term above satisfies +���(I − W2i+1) +1 +2 +� +Pi+1 − (I − W2i+1)+� +(I − W2i+1) +1 +2 +��� ≤ ǫi+1 +33 + +by the induction hypothesis. Next, we bound the first and third terms similarly: +����� +(I − W2i) +1 +2 (I + Wi)((I − W2i+1) +1 +2)+ +√ +2 +����� ≤ +����� +(I − W2i) +1 +2 (I + W2i)((I − W2i+1) +1 +2 )+ +√ +2 +����� + +(28) +����� +(I − W2i) +1 +2 (Wi − W2i)((I − W2i+1) +1 +2)+ +√ +2 +����� +≤ 1 + O(i · ǫ) +as +���� +(I−W2i) +1 +2 (I+W2i)((I−W2i+1 ) +1 +2 )+ +√ +2 +���� ≤ 1 by Equation (21) in Lemma 5.4 and +���� +(I−W2i) +1 +2 (Wi−W2i)((I−W2i+1 ) +1 +2 )+ +√ +2 +���� ≤ +O(i · ǫ) by Lemma 5.3 and Equation (22) in Lemma 5.4. Combining all the above, +���(I − W2i) +1 +2(Pi − Qi)(I − W2i) +1 +2 +��� ≤ (1 + O(i · ǫ))2ǫi+1, +giving the bound for (25). Next, for (26), we have +���(I − W2i) +1 +2 (Qi − Ri)(I − W2i) +1 +2 +��� = +1 +2 +���(I − W2i) +1 +2(I + Wi)(I − W2i+1)+(Wi − W2i)(I − W2i) +1 +2 +��� ≤ +����� +(I − W2i) +1 +2 (I + Wi)((I − W2i+1) +1 +2 )+ +√ +2 +����� · +����� +((I − W2i+1) +1 +2 )+(Wi − W2i)(I − W2i) +1 +2 +√ +2 +����� +We can get bounds for the two terms above similar to (28). Thus we get, +���(I − W2i) +1 +2 (Qi − Ri)(I − W2i) +1 +2 +��� ≤ +����� +(I − W2i) +1 +2 (I + Wi)((I − W2i+1) +1 +2)+ +√ +2 +����� · +����� +((I − W2i+1) +1 +2 )+(Wi − W2i)(I − W2i) +1 +2 +√ +2 +����� ≤ +(1 + O(i · ǫ)) · O(i · ǫ) ≤ O(i · ǫ). +Finally, for (27), we have +∥(I − W2i) +1 +2 (Ri − Si)(I − W2i) +1 +2 ∥ = +1 +2 +���(I − W2i) +1 +2 (Wi − W2i)(I − W2i+1)+(I + W2i)(I − W2i) +1 +2 +��� ≤ +����� +(I − W2i) +1 +2(Wi − W2i)((I − W2i+1) +1 +2)+ +√ +2 +����� · +����� +((I − W2i+1) +1 +2 )+(I + W2i)(I − W2i) +1 +2 +√ +2 +����� +Further, similar to the earlier upper bounds, we have +���� +(I−W2i) +1 +2 (Wi−W2i)((I−W2i+1 ) +1 +2 )+ +√ +2 +���� ≤ O(i · ǫ) +and +���� +((I−W2i+1) +1 +2 )+(I+W2i)(I−W2i ) +1 +2 +√ +2 +���� ≤ 1. Thus +∥(I − W2i) +1 +2(Ri − Si)(I − W2i) +1 +2∥ ≤ O(i · ǫ). +34 + +Proof of Theorem 5.2. For ǫ < +1 +k2 , we can solve for ǫ0 from ǫi = (1 + O(i · ǫ))ǫi+1 + O(i · ǫ) in +Lemma 5.5 and get ǫ0 = O(ǫ · k2). To complete the proof, we show +∥P0(I − W) − I∥B = ∥(I − W) +1 +2 � +P0 − (I − W)+� +(I − W) +1 +2∥. +Note that, +∥(I − W) +1 +2 � +P0 − (I − W)+� +(I − W) +1 +2∥ += ∥(I − W)1/2 [P0(I − W) − I] (I − W)+/2∥ += +���� +� +(I − W)∗/2(I − W)1/2�1/2 +[P0(I − W) − I] +� +(I − W)+/2((I − W)∗)+/2�1/2���� += +���� +� +(I − W)∗/2(I − W)1/2�1/2 +[P0(I − W) − I] +� +(I − W)∗/2(I − W)1/2�+/2���� += ∥P0(I − W) − I∥B +where we used Lemma B.6. +Acknowledgements +We thank Jack Murtagh for his helpful collaboration at the early stage of research on this research +and we thank Jonathan Kelner for helpful conversations at various stages of work on this project. +Aaron Sidford was supported by a Microsoft Research Faculty Fellowship, NSF CAREER Award +CCF-1844855, NSF Grant CCF-1955039, a PayPal research award, and a Sloan Research Fellow- +ship. +References +[ABKS21] +Deeksha Adil, Brian Bullins, Rasmus Kyng, and Sushant Sachdeva. Almost-linear- +time weighted lp-norm solvers in slightly dense graphs via sparsification. In Nikhil +Bansal, Emanuela Merelli, and James Worrell, editors, 48th International Colloquium +on Automata, Languages, and Programming, ICALP 2021, July 12-16, 2021, Glasgow, +Scotland (Virtual Conference), volume 198 of LIPIcs, pages 9:1–9:15. Schloss Dagstuhl +- Leibniz-Zentrum f¨ur Informatik, 2021. +[ACL06] +Reid Andersen, Fan R. K. Chung, and Kevin J. Lang. Local graph partitioning using +pagerank vectors. +In 47th Annual IEEE Symposium on Foundations of Computer +Science (FOCS 2006), 21-24 October 2006, Berkeley, California, USA, Proceedings, +pages 475–486. IEEE Computer Society, 2006. +[ADK22] +Daniel Agassy, Dani Dorfman, and Haim Kaplan. Expander decomposition with fewer +inter-cluster edges using a spectral cut player. CoRR, abs/2205.10301, 2022. +[AG15] +Nima Anari and Shayan Oveis Gharan. Effective-resistance-reducing flows, spectrally +thin trees, and asymmetric TSP. In Venkatesan Guruswami, editor, IEEE 56th Annual +Symposium on Foundations of Computer Science, FOCS 2015, Berkeley, CA, USA, +17-20 October, 2015, pages 20–39. IEEE Computer Society, 2015. +35 + +[AGM+10] +Arash Asadpour, Michel X. Goemans, Aleksander Madry, Shayan Oveis Gharan, and +Amin Saberi. An o(log n/ log log n)-approximation algorithm for the asymmetric +traveling salesman problem. In Moses Charikar, editor, Proceedings of the Twenty- +First Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2010, Austin, +Texas, USA, January 17-19, 2010, pages 379–389. SIAM, 2010. +[Ahm20] +AmirMahdi Ahmadinejad. Computing Stationary Distributions: Perron Vectors, Ran- +dom Walks, and Ride-Sharing Competition. Stanford University, 2020. +[AJSS19] +AmirMahdi Ahmadinejad, Arun Jambulapati, Amin Saberi, and Aaron Sidford. +Perron-frobenius theory in nearly linear time: Positive eigenvectors, m-matrices, graph +kernels, and other applications. In Proceedings of the Thirtieth Annual ACM-SIAM +Symposium on Discrete Algorithms, SODA 2019, San Diego, California, USA, Jan- +uary 6-9, 2019, pages 1387–1404. SIAM, 2019. +[AKM+20] +AmirMahdi Ahmadinejad, Jonathan A. Kelner, Jack Murtagh, John Peebles, Aaron +Sidford, and Salil P. Vadhan. High-precision estimation of random walks in small +space. +In Sandy Irani, editor, 61st IEEE Annual Symposium on Foundations of +Computer Science, FOCS 2020, Durham, NC, USA, November 16-19, 2020, pages +1295–1306. IEEE, 2020. +[AM85] +Noga Alon and V. D. Milman. lambda1, isoperimetric inequalities for graphs, and +superconcentrators. J. Comb. Theory, Ser. B, 38(1):73–88, 1985. +[BK00] +Andr´as A. Bencz´ur and David R. Karger. Augmenting undirected edge connectivity +in ˜o(n2) time. J. Algorithms, 37(1):2–36, 2000. +[BSS12] +Joshua D. Batson, Daniel A. Spielman, and Nikhil Srivastava. +Twice-ramanujan +sparsifiers. SIAM J. Comput., 41(6):1704–1721, 2012. +[CGP+18] +Timothy Chu, Yu Gao, Richard Peng, Sushant Sachdeva, Saurabh Sawlani, and Junx- +ing Wang. Graph sparsification, spectral sketches, and faster resistance computation, +via short cycle decompositions. In Mikkel Thorup, editor, 59th IEEE Annual Sympo- +sium on Foundations of Computer Science, FOCS 2018, Paris, France, October 7-9, +2018, pages 361–372. IEEE Computer Society, 2018. +[CKK+18] +Michael B. Cohen, Jonathan A. Kelner, Rasmus Kyng, John Peebles, Richard Peng, +Anup B. Rao, and Aaron Sidford. Solving directed laplacian systems in nearly-linear +time through sparse LU factorizations. In Mikkel Thorup, editor, 59th IEEE Annual +Symposium on Foundations of Computer Science, FOCS 2018, Paris, France, October +7-9, 2018, pages 898–909. IEEE Computer Society, 2018. +[CKM+11] +Paul F. Christiano, Jonathan A. Kelner, Aleksander Madry, Daniel A. Spielman, and +Shang-Hua Teng. +Electrical flows, laplacian systems, and faster approximation of +maximum flow in undirected graphs. In Lance Fortnow and Salil P. Vadhan, editors, +Proceedings of the 43rd ACM Symposium on Theory of Computing, STOC 2011, San +Jose, CA, USA, 6-8 June 2011, pages 273–282. ACM, 2011. +[CKM+14] +Michael B. Cohen, Rasmus Kyng, Gary L. Miller, Jakub W. Pachocki, Richard Peng, +Anup B. Rao, and Shen Chen Xu. Solving SDD linear systems in nearly mlog1/2n +time. In David B. Shmoys, editor, Symposium on Theory of Computing, STOC 2014, +New York, NY, USA, May 31 - June 03, 2014, pages 343–352. ACM, 2014. +36 + +[CKP+16] +Michael B. Cohen, Jonathan A. Kelner, John Peebles, Richard Peng, Aaron Sidford, +and Adrian Vladu. Faster algorithms for computing the stationary distribution, sim- +ulating random walks, and more. In Irit Dinur, editor, IEEE 57th Annual Symposium +on Foundations of Computer Science, FOCS 2016, 9-11 October 2016, Hyatt Regency, +New Brunswick, New Jersey, USA, pages 583–592. IEEE Computer Society, 2016. +[CKP+17] +Michael B. Cohen, Jonathan A. Kelner, John Peebles, Richard Peng, Anup B. Rao, +Aaron Sidford, and Adrian Vladu. Almost-linear-time algorithms for markov chains +and new spectral primitives for directed graphs. In Hamed Hatami, Pierre McKenzie, +and Valerie King, editors, Proceedings of the 49th Annual ACM SIGACT Symposium +on Theory of Computing, STOC 2017, Montreal, QC, Canada, June 19-23, 2017, +pages 410–419. ACM, 2017. +[CLTW22] +Lijie Chen, Xin Lyu, Avishay Tal, and Hongxun Wu. +Personal communication, +November 2022. +[GV22] +Louis Golowich and Salil Vadhan. Pseudorandomness of expander random walks for +symmetric functions and permutation branching programs. In Anindya De, editor, +Proceedings of the 37th Computational Complexity Conference (CCC ‘22), volume 234 +of LIPIcs, pages 27:1–27:15. Schloss Dagstuhl - Leibniz-Zentrum f¨ur Informatik, 2022. +Full version posted as ECCC TR22-024. +[HPV21] +William M. Hoza, Edward Pyne, and Salil P. Vadhan. Pseudorandom generators for +unbounded-width permutation branching programs. In James R. Lee, editor, 12th +Innovations in Theoretical Computer Science Conference, ITCS 2021, January 6-8, +2021, Virtual Conference, volume 185 of LIPIcs, pages 7:1–7:20. Schloss Dagstuhl - +Leibniz-Zentrum f¨ur Informatik, 2021. +[INW94] +Russell Impagliazzo, Noam Nisan, and Avi Wigderson. Pseudorandomness for network +algorithms. In Frank Thomson Leighton and Michael T. Goodrich, editors, Proceedings +of the Twenty-Sixth Annual ACM Symposium on Theory of Computing, 23-25 May +1994, Montr´eal, Qu´ebec, Canada, pages 356–364. ACM, 1994. +[JS21] +Arun Jambulapati and Aaron Sidford. Ultrasparse ultrasparsifiers and faster lapla- +cian system solvers. +In D´aniel Marx, editor, Proceedings of the 2021 ACM-SIAM +Symposium on Discrete Algorithms, SODA 2021, Virtual Conference, January 10 - +13, 2021, pages 540–559. SIAM, 2021. +[Kel21] +Zander Kelley. An alternative proof that inw fools permutation programs. Unpub- +lished manuscript, February 2021. +[KLOS14] +Jonathan A. Kelner, Yin Tat Lee, Lorenzo Orecchia, and Aaron Sidford. An almost- +linear-time algorithm for approximate max flow in undirected graphs, and its multi- +commodity generalizations. In Chandra Chekuri, editor, Proceedings of the Twenty- +Fifth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2014, Portland, +Oregon, USA, January 5-7, 2014, pages 217–226. SIAM, 2014. +[KM09] +Jonathan A. Kelner and Aleksander Madry. Faster generation of random spanning +trees. In 50th Annual IEEE Symposium on Foundations of Computer Science, FOCS +2009, October 25-27, 2009, Atlanta, Georgia, USA, pages 13–21. IEEE Computer +Society, 2009. +37 + +[KMG22] +Rasmus Kyng, Simon Meierhans, and Maximilian Probst Gutenberg. Derandomizing +directed random walks in almost-linear time. CoRR, abs/2208.10959, 2022. +[KMP12] +Jonathan A. Kelner, Gary L. Miller, and Richard Peng. Faster approximate multi- +commodity flow using quadratically coupled flows. In Howard J. Karloff and Toniann +Pitassi, editors, Proceedings of the 44th Symposium on Theory of Computing Con- +ference, STOC 2012, New York, NY, USA, May 19 - 22, 2012, pages 1–18. ACM, +2012. +[KMP14] +Ioannis Koutis, Gary L. Miller, and Richard Peng. Approaching optimality for solving +SDD linear systems. SIAM J. Comput., 43(1):337–354, 2014. +[KVV04] +Ravi Kannan, Santosh S. Vempala, and Adrian Vetta. On clusterings: Good, bad and +spectral. J. ACM, 51(3):497–515, 2004. +[LRS13] +Yin Tat Lee, Satish Rao, and Nikhil Srivastava. A new approach to computing maxi- +mum flows using electrical flows. In Dan Boneh, Tim Roughgarden, and Joan Feigen- +baum, editors, Symposium on Theory of Computing Conference, STOC’13, Palo Alto, +CA, USA, June 1-4, 2013, pages 755–764. ACM, 2013. +[LS13] +Yin Tat Lee and Aaron Sidford. Efficient accelerated coordinate descent methods +and faster algorithms for solving linear systems. In 54th Annual IEEE Symposium on +Foundations of Computer Science, FOCS 2013, 26-29 October, 2013, Berkeley, CA, +USA, pages 147–156. IEEE Computer Society, 2013. +[LS18] +Yin Tat Lee and He Sun. Constructing linear-sized spectral sparsification in almost- +linear time. SIAM J. Comput., 47(6):2315–2336, 2018. +[MST15] +Aleksander Madry, Damian Straszak, and Jakub Tarnawski. Fast generation of ran- +dom spanning trees and the effective resistance metric. In Piotr Indyk, editor, Pro- +ceedings of the Twenty-Sixth Annual ACM-SIAM Symposium on Discrete Algorithms, +SODA 2015, San Diego, CA, USA, January 4-6, 2015, pages 2019–2036. SIAM, 2015. +[OSV12] +Lorenzo Orecchia, Sushant Sachdeva, and Nisheeth K. Vishnoi. Approximating the +exponential, the lanczos method and an ˜o(m)-time spectral algorithm for balanced +separator. In Howard J. Karloff and Toniann Pitassi, editors, Proceedings of the 44th +Symposium on Theory of Computing Conference, STOC 2012, New York, NY, USA, +May 19 - 22, 2012, pages 1141–1160. ACM, 2012. +[PS14] +Richard Peng and Daniel A. Spielman. An efficient parallel solver for SDD linear +systems. STOC, 2014. +[PV21] +Edward Pyne and Salil Vadhan. Pseudodistributions That Beat All Pseudorandom +Generators (Extended Abstract). In Proceedings of the 36th Annual Computational +Complexity Conference (CCC), pages 33:1–33:15, 2021. +[PY19] +Merav Parter and Eylon Yogev. Optimal short cycle decomposition in almost lin- +ear time. In Christel Baier, Ioannis Chatzigiannakis, Paola Flocchini, and Stefano +Leonardi, editors, 46th International Colloquium on Automata, Languages, and Pro- +gramming, ICALP 2019, July 9-12, 2019, Patras, Greece, volume 132 of LIPIcs, pages +89:1–89:14. Schloss Dagstuhl - Leibniz-Zentrum f¨ur Informatik, 2019. +38 + +[RV05] +Eyal Rozenman and Salil P. Vadhan. Derandomized squaring of graphs. In Chandra +Chekuri, Klaus Jansen, Jos´e D. P. Rolim, and Luca Trevisan, editors, Approxima- +tion, Randomization and Combinatorial Optimization, Algorithms and Techniques, +8th International Workshop on Approximation Algorithms for Combinatorial Opti- +mization Problems, APPROX 2005 and 9th InternationalWorkshop on Randomization +and Computation, RANDOM 2005, Berkeley, CA, USA, August 22-24, 2005, Proceed- +ings, volume 3624 of Lecture Notes in Computer Science, pages 436–447. Springer, +2005. +[SS08] +Daniel A. Spielman and Nikhil Srivastava. +Graph sparsification by effective resis- +tances. In Cynthia Dwork, editor, Proceedings of the 40th Annual ACM Symposium +on Theory of Computing, Victoria, British Columbia, Canada, May 17-20, 2008, +pages 563–568. ACM, 2008. +[ST04] +Daniel A. Spielman and Shang-Hua Teng. Nearly-linear time algorithms for graph +partitioning, graph sparsification, and solving linear systems. In L´aszl´o Babai, editor, +Proceedings of the 36th Annual ACM Symposium on Theory of Computing, Chicago, +IL, USA, June 13-16, 2004, pages 81–90. ACM, 2004. +[Tro12] +Joel A. Tropp. User-friendly tail bounds for sums of random matrices. Found. Comput. +Math., 12(4):389–434, 2012. +[Vad12] +Salil P. Vadhan. Pseudorandomness. Found. Trends Theor. Comput. Sci., 7(1-3):1– +336, 2012. +[vdBGJ+22] Jan van den Brand, Yu Gao, Arun Jambulapati, Yin Tat Lee, Yang P. Liu, Richard +Peng, and Aaron Sidford. Faster maxflow via improved dynamic spectral vertex sparsi- +fiers. In Stefano Leonardi and Anupam Gupta, editors, STOC ’22: 54th Annual ACM +SIGACT Symposium on Theory of Computing, Rome, Italy, June 20 - 24, 2022, pages +543–556. ACM, 2022. +[vdBLL+21] Jan van den Brand, Yin Tat Lee, Yang P. Liu, Thatchaphol Saranurak, Aaron Sidford, +Zhao Song, and Di Wang. Minimum cost flows, mdps, and l1-regression in nearly linear +time for dense instances. In Samir Khuller and Virginia Vassilevska Williams, editors, +STOC ’21: 53rd Annual ACM SIGACT Symposium on Theory of Computing, Virtual +Event, Italy, June 21-25, 2021, pages 859–869. ACM, 2021. +A +SV Approximation Proofs +In this section we provide proofs of equivalences for definitions related to SV-approximation (Ap- +pendix A.1), properties of SV-approximation (Appendix A.2), and separations of SV-approximation +from other notions of approximation (Appendix A.3). +We recall results useful in the proofs: +Fact A.1 ([CKP+17]). Let A ∈ Cn×n be a Hermitian matrix and I ⊂ [n] be arbitrary. Then A is +PSD if and only if AI,I is PSD and SCI(A) is PSD. +We require a result that matrix approximation is approximately preserved under Schur comple- +ments: +39 + +Theorem A.2 (Theorem 4.8 [AKM+20]). Given N, ˜N ∈ Cn×n such that ∥N∥ ≤ 1, suppose ˜N is +an ε-approximation of N with respect to SN. Then for F ⊂ [n] such that IF,F − NF,F is invertible, +we have that IF c − SCF(I − ˜N) is an ε + O(ε2) approximation of IF c − SCF(I − N) with respect to +SIF c−SCF (I−N). +We also prove a useful property of matri approximation: +Lemma A.3 (Manipulating matrix approximation). Let A, �A ∈ Cm×n, let E ∈ Cm×m, F ∈ Cn×n +be PSD and Hermitian, and assume that A is an ε-approximation of �A with respect to error matrices +E and F, for some ε ≥ 0. +1. For U ∈ Cm′×m, V ∈ Cn×n′, U�AV is an ε-approximation of UAV with respect to error +matrices E′ and F′, where E′ = UEU∗ and F′ = V∗FV. +2. For PSD Hermitian E′ ∈ Cm×m and F′ ∈ Cn×n a with E ⪯ cE′ and F ⪯ cF′ for a constant +c ≥ 0, then �A is a cε-approximation of A with respect to error matrices E′ and F′. +Proof. +1. For arbitrary x ∈ Cm and y ∈ Cn the definition of ε-approximation implies, +���x∗(U �AV − UAV)y +��� = +���x∗U( �A − A)Vy +��� ≤ ε +2 (x∗UEU∗x + y∗V∗FVy) . +2. For arbitrary x ∈ Cm, y ∈ Cn, +���x∗( �A − A)y +��� ≤ ε +2 (x∗Ex + y∗Fy) ≤ cε +2 +� +x∗E′x + y∗F′y +� +. +A.1 +Equivalences +Lemma 3.6 (Conditions for SV approximation to be defined). Let A, �A ∈ Cm×n, and let Din ∈ +Cm×m and Dout ∈ Cn×n be PSD matrices such that ker(Din) ⊆ lker(A) and ker(Dout) ⊆ rker(A). +Then the following are equivalent: +1. σmax(D+/2 +in AD+/2 +out ) ≤ 1. +2. Din − AD+ +outA∗ is PSD. +3. Dout − A∗D+ +inA is PSD. +4. For some scalar z ∈ C with |z| = 1, +� Din +zA +z∗A∗ +Dout +� +is PSD. +5. For every scalar z ∈ C with |z| ≤ 1, +� Din +zA +z∗A∗ +Dout +� +is PSD. +Suppose further that Din = diag(din) and Din = diag(dout) for din ∈ Rm +≥0, dout ∈ Rn +≥0. Then +Condition 1 below implies Condition 2 below, which implies Conditions 1–5 above. +1. A is nonnegative, dout = A⃗1n, din = ⃗1⊤ +mA. +2. For all i ∈ [n], (dout)i ≥ � +j |Ai,j|, and for all j ∈ [m], (din)j ≥ � +i |Ai,j|. +Proof. We first show Conditions 1 − 5 are equivalent. +40 + +1→ 3 Condition 1 implies for all y, y∗D+/2 +out A∗D+ +inAD+/2 +out y ≤ y∗y and letting y ← D1/2 +outy and using +that ker(Dout) ⊆ ker(A) implies Condition 3. +3 → 1 Condition 3 implies for all y we have y∗A∗D+ +inAy ≤ y∗Douty and taking y ← D+/2 +out y we have +y∗D+/2 +out A∗D+ +inAD+/2 +out y ≤ y∗ΠDouty ≤ y∗y which implies Condition 1. +3→ 5 Letting z with |z| ≤ 1 be arbitrary, we have that by Condition 3 Din − z∗A∗D+ +outzA = +Din − A∗D+ +outA ⪰ 0, and since Din ⪰ 0 we apply Fact A.1 and conclude Condition 5. An +analogous argument shows 2→ 5. The fact that Fact A.1 is an equivalence then implies 5→2 +and 5→3. +5 → 4 Immediate. +4 → 2 By Fact A.1 and the fact that z∗z = 1 we have Din − z∗A∗D+ +outzA = Din − A∗D+ +outA ⪰ 0. +We next show the latter set of conditions. +1→2 For all i ∈ [n] we have [dout]i = � +j Ai,j = � +j |Ai,j|. +2 → 5 We use that a Hermitian matrix that is diagonally dominant is positive semidefinite. We have +(Din)i,i ≥ � +j |zAi,j| and (Dout)i,i ≥ � +j |z∗A∗ +j,i| so +�Din +A +A∗ +Dout +� +is diagonally dominant and +thus PSD. +Lemma 3.7 (Equivalent formulations of SV approximation). Let A, �A ∈ Cm×n and let Din ∈ +Cm×m and Dout ∈ Cn×n be PSD matrices. Then the following are equivalent +1. �A +Din,Dout +≈ε +A. +2. slift +� +�A +� D≈ε slift +� +�B +� +, where +slift +� +�A +� += +�0m×m +A +A∗ +0n×n +� +, slift (A) = +� +0m×m +�A +�A∗ +0n×n +� +, D = +� Din +0m×n +0n×m +Dout +� +. +3. For some scalar z ∈ C with |z| = 1, ˜C is ε/2-approximation of C with respect to error matrix +E, where +C = +�0m×m +zA +0n×m +0n×n +� +, ˜C = +� +0m×m +z �A +0n×m +0n×n +� +, E = +� Din +zA +z∗A∗ +Dout +� +4. Item 3 holds for every z ∈ C such that |z| ≤ 1. +Proof. We prove 1 ⇐⇒ 2 and 1 → 4 → 3 → 1. +1 ⇐⇒ 2 Suppose Item 1 holds. Then for arbitrary test vectors x = +�x1 +x2 +� +, y = +�y1 +y2 +� +where x1, y1 ∈ Cm +and x2, y2 ∈ Cn we have +���x∗(B − �B)y +��� ≤ +���x∗ +1(A − �A)y2 +��� + +���x∗ +2(A∗ − �A∗)y1 +��� +≤ ε +4 +� +x∗ +1Dinx1 + y∗ +2Douty2 + x∗ +1AD+ +outA∗x1 + y∗ +2A∗D+ +inAy2 +� ++ ε +4 +� +x∗ +2Doutx2 + y∗ +1Diny1 + x∗ +2A∗D+ +inAx2 + y∗ +1AD+ +outA∗y1 +� += ε +4 +� +x∗Dx + y∗Dy + x∗BD−1B∗x + y∗B∗D−1By +� +. +41 + +Furthermore, ker(D) ⊆ lker(B), ker(D) ⊆ rker(B) and D − BD+B ⪰ 0 by Item 5 of +Lemma 3.6. In the other direction, we obtain 2 → 1 by considering the set of test vectors +�x1 +0n +� +and +�0m +y2 +� +. +1 → 4: Fix arbitrary z with |z| ≤ 1. We have that E ⪰ 0 by Item 5 of Lemma 3.6. Then for arbitrary +test vectors x = +�x1 +x2 +� +, y = +�y1 +y2 +� +, let x′ = +�x1 +x′ +2 +� +, y′ = +�y1 +y′ +2 +� +where +x′ +2 = argminv∈Cn +� +x∗ +1 +v∗� � Din +zA +z∗A∗ +Dout +� �x1 +v +� +, +y′ +1 = argminv∈Cm +� +y∗ +2 +v∗� �Dout +z∗A∗ +zA +Din +� �y2 +v +� +. +Thus +x′∗ +� Din +zA +z∗A∗ +Dout +� +x′ = x∗ +1SC +�� Din +zA +z∗A∗ +Dout +�� +x1, +y′∗ +� Din +zA +z∗A∗ +Dout +� +y′ = y∗ +2SC +��Dout +z∗A∗ +zA +Din +�� +y2. +Then we have +���x∗(˜C − C)y +��� = +���(x1z∗)∗( �A − A)(y2z) +��� +≤ ε +4 +� +x∗ +1(Din − zAD+ +outA∗z∗)x1 + y∗ +2(Dout − z∗A∗D+ +inAz)y2 +� += ε +4 +� +x∗ +1SC +��Din +z∗A∗ +zA +Dout +�� +x1 + y∗ +2SC +��Dout +zA +z∗A∗ +Din +�� +y2 +� += ε +4 +� +x′∗ +� Din +zA +z∗A∗ +Dout +� +x′ + y′∗ +� Din +zA +z∗A∗ +Dout +� +y′ +� +≤ ε +4 (x∗Ex + y∗Ey) . +4 → 3 Immediate. +3 → 1 For arbitrary test vectors x1 ∈ Cm, y2 ∈ Cn, let x = +�zx1 +x2 +� +, y = +� y1 +z∗y2 +� +where +x2 = argminv∈Cn +� +z∗x∗ +1 +v∗� � Din +zA +z∗A∗ +Dout +� �zx1 +v +� +, +y1 = argminv∈Cm +� +zy∗ +2 +v∗� �Dout +z∗A∗ +zA +Din +� �z∗y2 +v +� +Then +���x∗ +1( �A − A)y2 +��� = +���x∗(˜C − C)x +��� +≤ ε +4 (x∗Ex + y∗Ey) += ε +4 +� +z∗x∗ +1SC +��Din +z∗A∗ +zA +Dout +�� +x1z + zy∗ +2SC +��Dout +zA +z∗A∗ +Din +�� +z∗y2 +� += ε +4 +� +z∗x∗ +1(Din − zAD+ +outA∗z∗)zx1 + zy∗ +2(Dout − z∗A∗D+ +inAz)z∗y2 +� += ε +4 +� +x∗ +1(Din − AD+ +outA∗)x1 + y∗ +2(Dout − A∗D+ +inA)y2 +� +. +42 + +Lemma 3.8. Let A, �A ∈ Cm×n, and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices such that +ker(Din) ⊆ lker(A), and ker(Dout) ⊆ rker(A). Let N = D+/2 +in AD+/2 +out and ˜N = D+/2 +in +�AD+/2 +out Then +�A +Din,Dout +≈ε +A if and only if ˜N +svn +≈ ε N. +Proof. Suppose �A +Din,Dout +≈ε +A, i.e. �A is an ε/2-approximation of A with respect to +E = Din − AD+ +outA∗, +F = Dout − AD+ +inA∗. +Applying Lemma A.3 with U = D+/2 +in , V = D+/2 +out we obtain that ˜N is an ε/2-approximation of N +with respect to error matrices +D+/2 +in (Din − AD+ +outA∗)D+/2 +in +⪯ I − (D+/2 +in AD+/2 +out )(D+/2 +out A∗D+/2 +in ) = I − NN∗ +and +D+/2 +out (Dout − A∗D+ +inA)D+/2 +out ⪯ I − (D+/2 +out A∗D+/2 +in )(D+/2 +in AD+/2 +out ) = I − N∗N +and hence we obtain that ˜N +svn +≈ ε N. The other direction is analogous, and the kernel properties +follow as ker(Din) ⊆ lker(D1/2 +in D+/2 +in AD+/2 +out D1/2 +out) and ker(Dout) ⊆ rker(D1/2 +in D+/2 +in AD+/2 +out D1/2 +out). +Theorem 3.9 (Unitary transformation characterization of SV-approximation). For ˜N, N ∈ Cn×n, +we have that ˜N +svn +≈ ε N if for every pair of unitary matrices U, V, U ˜NV is a standard ε/2- +approximation of UNV with respect to degree matrix I. +Moreover, if ˜N +svn +≈ ε N then for every +pair of matrices U, V satisfying ∥U∥ ≤ 1, ∥V∥ ≤ 1, we have that U ˜NV +svn +≈ ε UNV, and hence +U ˜NV is a standard ε-approximation of UNV with respect to degree matrix I. +Proof. +• Suppose for every pair of unitary matrices U, V, we have for all x, y ∈ Cn that +|x∗(U�AV − UAV)y| ≤ ε +2 · +� +x∗(I − SUAV)x · +� +y∗(I − SUAV)y. +We will pick u, v, U, V to depend on x, y and the result will follow immediately from invoking +the above equation with x, y ← u, v. +Set U = I, u = x/∥x∥, and v = Ay/∥Ay∥. Note that u∗v = (x∗Ay)/(∥x∥ · ∥y∥), so u and +v have the same angle between them as A∗x and y. Hence, there exists a unitary matrix V +such that Vu = A∗x/∥A∗x∥ and Vv = y/∥y∥. Substituting gives +���x∗( �A − A)y +��� +∥x∥∥y∥ +≤ ε +2 · +� +1 − x∗(AA∗)x +∥x∥∥A∗x∥ · +� +1 − y∗(A∗A)y +∥y∥∥Ay∥ . +Applying ∥A∗∥ = ∥A∥ ≤ 1 gives the desired result. +• In the other direction, we start with Item 3 of Lemma 3.7 with z = 1 applied to the approxi- +mation statement ˜N +svn +≈ ε N and apply Lemma A.3 with left and right hand side matrices +U′ ← +�U +0 +0 +V∗ +� +and +V′ ← +�U∗ +0 +0 +V +� +43 + +This results in the approximation statement that for the matrices given below, R, �R ǫ/2- +approximate each other with respect to E where +R = +�0 +UNV +0 +0 +� +�R = +� +0 +U ˜NV +0 +0 +� +, and +E = +� UIU∗ +UNV +(UNV)∗ +V∗IV +� +⪯ +� +I +UNV +(UNV)∗ +I +� +. +Hence, by Item 3 of Lemma 3.7 we conclude UNV +svn +≈ ε U ˜NV. +Corollary 3.10 (SV preservation under multiplication by permutation matrices). Let A, �A ∈ +Cm×n +≥0 +and suppose �A +Din,Dout +≈ε +A. Let U, V be arbitrary permutation matrices. Then U �AV +D′ +in,D′ +out +≈ε +UAV where D′ +in = UDinU∗ and D′ +out = V∗DoutV. Consequently, if �A +sv≈ε A, then U�AV +sv≈ε UAV +and if �A +svn +≈ ε A then U �AV +svn +≈ ε UAV. +We start with Item 3 of Lemma 3.7 with z = 1 applied to the approximation statement we are +given in the lemma statement and apply Lemma A.3 with left and right hand side matrices +U′ ← +�U +0 +0 +V∗ +� +and +V′ ← +�U∗ +0 +0 +V +� +This results in the approximation statement that for the matrices given below, R, �R ǫ/2-approximate +each other with respect to E where +R = +�0 +UAV +0 +0 +� +�R = +� +0 +U �AV +0 +0 +� +, and +E = +�UDinU∗ +UAV +(UAV)∗ +V∗DoutV +� +. +Hence, Item 3 of Lemma 3.7 is satisfied for SV approximation of UAV and U �AV with respect to +UDinU∗ and V∗DoutV. +For the first “consequently” claim, observe that UAV⃗1 = UA⃗1 since V is a permutation. +Hence, +diag(UAV⃗1) = UDinU∗. +A similar analysis shows +diag(⃗1⊤UAV) = V∗DinV. +the second statement follows from UU∗ = V∗V = I. +A.2 +Properties +We prove that SV approximation implies standard approximation: +Lemma A.4. Let A, �A ∈ Cn×n and suppose �A +Din,Dout +≈ε +A. Then, �A is an ε-approximation of A +with respect to E = Din − A, F = Dout − A∗. +Proof. This follows immediately from specializing Item 3 of Lemma 3.7 with z = −1 to test vectors +of the form +�x +x +� +, +�y +y +� +It likewise implies UC approximation: +Lemma 3.11. If A, �A ∈ Cn×n with �A +D≈ε A then �A +◦≈ε A with respect to degree matrix D. +44 + +Proof. By Item 4 of Lemma 3.7 we have that for every pair of test vectors +�x +x +� +, +�y +y +� +and every unit +magnitude z, we have +���x∗(zA − z �A)y +��� ≤ ε +4 · 2 (x∗Dx + y∗Dy + ℜ(zx∗Ax + zyAy∗)) +Then choosing z to minimize ℜ(zx∗Ax + zyAy∗), we obtain +���x∗(A − �A)y +��� ≤ 2ε +4 (x∗Dx + y∗Dy − |x∗Ax + yAy∗|) +Which is precisely the condition for �A +◦≈ε A with respect to D. +Lemma 3.14 (SV preservation under arbitrary lifting). Let A, �A ∈ Cm×n be matrices such that +�A +Din,Dout +≈ε +A. Then for all integers i, j, k, ℓ ≥ 0 + + +0i×j +0i×n +0i×k +0m×j +�A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + +D′ +in,D′ +out +≈ε + + +0i×j +0i×n +0i×k +0m×j +A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + . +Where +D′ +in = + + +0i×i +0i×n +0i×ℓ +0n×i +Din +0n×ℓ +0ℓ×i +0ℓ×n +0ℓ×ℓ + + , +D′ +out = + + +0j×j +0j×m +0j×k +0m×j +Din +0m×k +0k×j +0k×m +0k×k + + . +Consequently, if �A +sv≈ε A then + + +0i×j +0i×n +0i×k +0m×j +�A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + sv≈ε + + +0i×j +0i×n +0i×k +0m×j +A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + . +Proof. Observe by expanding the definition of SV approximation that it is preserved under em- +bedding into the first principal diagonal block of any larger matrix that is zeros elsewhere. That +is, +� +�A +0m×(j+k) +0(i+ℓ)×n +0(i+ℓ)×(j+k) +� D′ +in,D′ +out +≈ε +� +A +0m×(j+k) +0(i+ℓ)×n +0(i+ℓ)×(j+k) +� +(29) +where D′ +in, D′ +out are the respective principle embeddings of the degree matrices. +Let U, V be permutation matrices such that +U +� +A +0m×(j+k) +0(i+ℓ)×n +0(i+ℓ)×(j+k) +� +V = + + +0i×j +0i×n +0i×k +0m×j +A +0m×k +0ℓ×j +0ℓ×n +0ℓ×k + + . +It suffices to argue that we can apply this transformation to both sides of Equation (29) and have +the approximation still hold. This is implied by Corollary 3.10 with D′ +in ← UD′ +inU∗ and D′ +out ← +V∗DoutV. The “consequently” claim holds from inspecting the resulting block structure. +Lemma 3.16 (SV preservation under products). Let (Ni)i∈[ℓ], ( ˜Ni)i∈[ℓ] ∈ Cn×n be such that for +every i, ˜Ni +svn +≈ ε Ni. Then ˜Nℓ · · · ˜N2 ˜N1 +svn +≈ ε+O(ǫ2) Nℓ · · · N2N1. +45 + +Proof. Using preservation under lifting and sums (Lemmas 3.14 and 3.18) and that SV approx- +imation implies standard approximation for square matrices (Lemma A.4), we can obtain the +approximation statement that the matrices R, �R given below ǫ/2-standard-approximate each other +with respect to the symmetrization SR of the first one: +R = + + +I/2 +Nℓ +0 +0 +I +Nℓ−1 +... +... +... +... +... +I +N1 +0n×n +0 +I/2 + + +and +�R = + + +I/2 +˜Nℓ +0 +0 +I +˜Nℓ−1 +... +... +... +... +... +I +˜N1 +0n×n +0 +I/2 + + +. +Taking the Schur complement by eliminating the center n(l −1)×n(l −1) block gives the following +two matrices +R = +� I/2 +Nℓ · · · N2N1 +0n×n +I/2 +� +and +�R = +� +I/2 +˜Nℓ · · · ˜N2 ˜N1 +0n×n +I/2 +� +. +And we obtain by Theorem A.2 that �R is an ǫ/2 + O(ǫ2)-approximation of R with respect to SR. +Hence, by Item 3 of Lemma 3.7, we have +˜Nℓ · · · ˜N2 ˜N1 +svn +≈ ε+O(ǫ2) Nℓ · · · N2N1. +Lemma 3.17. If �A +sv≈ε A, then �A⃗1 = A⃗1 and �A⊤⃗1 = A⊤⃗1. +Proof. Let x ← δx be an arbitrary test vector and let y = ⃗1. Then the approximation condition +implies +���δx∗( �A − A)⃗1 +��� ≤ ε +4 +� +δ2x∗Doutx − δ2x∗AD+ +inA∗x +⃗1⊤Din⃗1 −⃗1⊤A⊤D+ +outA⃗1 +� += ε +4 +� +δ2x∗Doutx − δ2x∗AD+ +inA∗x +� +and so taking δ → 0 we obtain that A⃗1 = �A⃗1. An analogous argument setting x = ⃗1 shows that +⃗1⊤A = ⃗1⊤ �A. +Lemma 3.18. If �Ai +(Din)i,(Dout)i +≈ε +Ai for all i ∈ [k], letting Din +def += � +i∈[k](Din)i and Dout +def += +� +i∈[k](Dout)i, then � +i∈[k] �Ai +Din,Dout +≈ε +� +i∈[k] Ai. +Consequently, if �Ai +sv≈ε Ai for every i, then +� +i∈[k] �Ai +sv≈ε +� +i∈[k] Ai. +Proof. This follows immediately from Item 3 of Lemma 3.7 being linear over A and D. +The +“consequently” claim follows immediately from noting + +� +i∈[k] +A + +⃗1 = +� +i∈[k] +(A⃗1) and + +� +i∈[k] +A⊤ + +⃗1 = +� +i∈[k] +(A⊤⃗1). +Lemma 3.19. If A3 +Din,Dout +≈δ +A2 and A2 +Din,Dout +≈ε +A1 then A3 +Din,Dout +≈ε+δ+εδ A1. Consequently, if A3 +sv≈δ A2 +and A2 +sv≈εA1 then A3 +sv≈ε+δ+εδA1, and if A3 +svn +≈ δ A2 and A2 +svn +≈ εA1 then A3 +svn +≈ ε+δ+εδA1. Moreover, +if for δ ∈ (0, 1/2) and A0, . . . , Aℓ we have Ai +sv≈δ/2ℓ Ai−1 for every i, then Aℓ +sv≈δ A0. Moreover, the +equivalent claim holds for normalized SV approximation. +46 + +Proof. By Item 3 of Lemma 3.7 we have +�Din +A∗ +2 +A2 +Dout +� +⪯ (1 + ε) +�Din +A∗ +1 +A1 +Dout +� +. +Thus for arbitrary x ∈ Cm, y ∈ Cn we have +����x∗ +��0 +A3 +0 +0 +� +− +�0 +A1 +0 +0 +�� +y +���� +≤ +����x∗ +��0 +A3 +0 +0 +� +− +�0 +A2 +0 +0 +�� +y +���� + +����x∗ +��0 +A2 +0 +0 +� +− +�0 +A1 +0 +0 +�� +y +���� +≤ δ +4 +� +x∗ +�Din +A∗ +2 +A2 +Dout +� +x + y∗ +�Din +A∗ +2 +A2 +Dout +� +y +� ++ ε +4 +� +x∗ +�Din +A∗ +1 +A1 +Dout +� +x + y∗ +�Din +A∗ +1 +A1 +Dout +� +y +� +≤ δ(1 + ε) +4 +� +x∗ +�Din +A∗ +1 +A1 +Dout +� +x + y∗ +�Din +A∗ +1 +A1 +Dout +� +y +� ++ ε +4 +� +x∗ +�Din +A∗ +1 +A1 +Dout +� +x + y∗ +�Din +A∗ +1 +A1 +Dout +� +y +� += δ + ε + εδ +4 +� +x∗ +�Din +A∗ +1 +A1 +Dout +� +x + y∗ +�Din +A∗ +1 +A1 +Dout +� +y +� +and so we conclude by Item 3 of Lemma 3.7. The first “consequently” claim follows from the fact +that by Lemma 3.17 we have A2⃗1 = A⃗1 and A⊤ +2 ⃗1 = A⊤ +1 ⃗1, and the second follows immediately. +For the “moreover” claim, we prove by induction on i that Ai +sv≈iδ/2ℓ+iδ2/ℓ A0, which suffices as +ℓδ/2ℓ + ℓδ2/ℓ ≤ δ. Assuming this holds for i, we have +Ai+1 +sv≈δ/2ℓ Ai +sv≈iδ/2ℓ+iδ2/ℓ A0 +and thus we have +Ai+1 +sv≈(i+1)δ/2ℓ+γ A0 +for +γ = iδ2 +ℓ + δ +2ℓ +� iδ +2ℓ + iδ2 +ℓ +� +≤ iδ2 +ℓ + δ2/2ℓ + δ3/2ℓ2 ≤ (i + 1)δ2 +ℓ +as desired. The proof for normalized SV approximation is essentially identical, so we omit it. +Lemma 3.20. Let G be a strongly connected, d-regular directed multigraph on n vertices with +adjacency matrix A and let J ∈ Rn×n be a matrix with 1/n in every entry (i.e. J is the walk +matrix of the complete graph with a self loop on every vertex). Then λ(G) ≤ 1 − λ/2 if and only if +A/d +svn +≈ λ J. +Proof. Note that by definition of A +svn +≈ λ J, +|x∗(A/d − J)y| ≤ λ +4 (x∗(I − JJ∗)x + y∗(I − J∗J)y) += λ +4 (x∗(I − J)x + y∗(I − J)y) +Therefore, A/d +svn +≈ λ J if and only if A/d +◦≈λ/2 J. By [AKM+20, Lemma 5.2] A/d +◦≈λ/2 J if and +only if λ(G) ≤ 1 − λ/2, so the result follows. +47 + +We show a slightly stronger version of Item 2 of Lemma 3.7 for non-negative matrices, in that +we show SV sparsification of the bipartite lift of a graph implies SV sparsification of the undirlying +directed graph, even if the sparsifier is not promised to be bipartite-preserving, and even if the +sparsifier is not itself undirected. +Lemma A.5. Given non-negative A ∈ Rm×n +≥0 +, suppose there is nonnegative �A = +� +�AF,F +�AF,F c +�AF c,F +�AF c,F c +� +such that �A +sv≈ε +�0 +A⊤ +A +0 +� +. Then +1. �AF,F = 0 and �AF c,F c = 0. +2. �AF c,F +sv≈ε A and �AF,F c +sv≈ε A⊤. +Proof. Let Dout = diag(A⃗1) and Din = diag(⃗1⊤A) and observe ⃗1⊤Dout⃗1−⃗1⊤AD+ +inA⃗1 = ⃗1⊤Din⃗1− +⃗1⊤A⊤D+ +outA⊤⃗1 = 0. +1. Let x ∈ Cn be arbitrary, and consider the test vectors x′ = δ +� x +0m +� +, y′ = +�⃗1 +⃗1 +� +. Then from the +definition of SV approximation we obtain +���δx⊤ �AF,F⃗1 + δx∗ �AF,F c⃗1 − δxA⊤⃗1 +��� = +����x′∗ +��0 +A⊤ +A +0 +� +− �A +� +y +���� +≤ ε +2 · δ2x∗(Dout − AD+ +inA)x +and hence �AF,F⃗1 + x �AF,F c⃗1 = xA⊤⃗1 by taking δ → 0. Considering the test vectors x′ = +δ +� x +0m +� +, y′′ = +� ⃗1 +−⃗1 +� +we likewise obtain x �AF,F⃗1 − x �AF,F c⃗1 = −xA⊤⃗1 and hence x �AF,F⃗1 = 0. +By an analogous argument we obtain �AF c,F c = 0. +2. For arbitrary x ∈ Cm, y ∈ Cn, apply the test vectors [0n, x], [y, 0m]. Then we obtain +���x∗(A − �AF c,F )y +��� ≤ ε +2 +� +x∗(D − AD+AT )x + y∗(D − AT D+A)y +� +thus �AF c,F +sv≈ε A and an analogous argument shows �AF,F c +sv≈ε AT . +A.3 +Separations +Proposition 3.12. For every α, ε ∈ (0, 1), there are symmetric matrices W, � +W with ∥W∥, ∥� +W∥ ≤ +1 such that � +W +◦≈ε W but � +W is not an ε SV-approximation of W with respect to I − W for any +ε′ < +ε +√ +1−α2 . +Proof. Given α, we define +W = +�α +0 +0 +−α +� +and +� +W = · +� +α +ε +√ +1 − α2 +ε +√ +1 − α2 +−α +� +. +48 + +We first show that � +W +◦≈ε W. Since both matrices are symmetric, suffices [AKM+20, Lemma 3.7] +to show the equivalent statement for PSD approximation between I− � +W, I−W and I+ � +W, I+W, +i.e. +(1 − ε)(I − W) ⪯ I − � +W ⪯ (1 + ε)(I − W) +(1 − ε)(I + W) ⪯ I + � +W ⪯ (1 + ε)(I + W). +All four inequalities are implied by +� ε(1 − α) +ε +√ +1 − α2 +ε +√ +1 − α2 +ε(1 + α) +� +⪰ 0. +Using Fact A.1 this is equivalent to ε(1 − α) − ε(1 − α2)(1 + α)−1 ≥ 0 which holds with equality, +so � +W +◦≈ε W. Now assume that � +W +svn +≈ ε′ W for some ε′. From the definition of SV approximation +applied to test vector z = +�x +x +� +, we have +±(� +W − W) ⪯ ε′ � +I − W2� += ε′(1 − α2)I. +This is equivalent to +ǫ +� +1 − α2 ≤ ε′(1 − α2). +Thus for � +W +svn +≈ ε W, a necessary condition is for ε′ ≥ +ε +√ +1−α2 . +B +Facts about Singular Values +Lemma B.1. Let σi(·) denote the ith largest singular value. For any matrix A ∈ Cn×n and any +positive integer k, +σ2(Ak) ≤ σ2(A) · σ1(A)k−1. +Proof. By the variational characterization of singular values, we have +σ2(Ak) = min +v∈Cn max +x⊥v +∥Akx∥ +∥x∥ +≤ min +v∈Cn max +x⊥v +∥Ak−1∥ · ∥Ax∥ +∥x∥ +≤ ∥A∥k−1·min +v∈Cn max +x⊥v +∥Ax∥ +∥x∥ += σ2(A)·σ1(A)k−1. +Lemma B.2. Let σi(·) denote the ith largest singular value. For any matrix A, B ∈ Cn×n, +σ2(A + B) ≤ σ2(A) + σ1(B). +Furthermore, if A, B share a common right singular vector that achieves the maximum singular +value in each of them respectively, then +σ2(A + B) ≤ σ2(A) + σ2(B) +Proof. For the first part of the statement, by the variational characterization of singular values, we +have +σ2(A + B) = min +v∈Cn max +x⊥v +∥(A + B)x∥ +∥x∥ +≤ min +v∈Cn max +x⊥v +∥Ax∥ + ∥Bx∥ +∥x∥ +≤ min +v∈Cn max +x⊥v +∥Ax∥ +∥x∥ + ∥B∥. +49 + +For the second part of the statement, let v denote a common right singular vector of A, B that +respectively achieves the maximum singular value in each of them. Then we have +σ2(A + B) = min +v′∈Cn max +x⊥v′ +∥(A + B)x∥ +∥x∥ +≤ max +x⊥v +∥(A + B)x∥ +∥x∥ +≤ max +x⊥v +∥Ax∥ + ∥Bx∥ +∥x∥ +≤ max +x⊥v +∥Ax∥ +∥x∥ + max +x⊥v +∥Bx∥ +∥x∥ = σ2(A) + σ2(B). +The third part of the statement has essentially the same proof as the second. +We now use these properties to show that small perturbations preserve the smallest singular +value: +Lemma B.3. Let σ2(·) denote the second largest singular value. Let G be an Eulerian graph such +that σ2(D−1/2 +G +AGD−1/2) ≤ 1 − 1/γ. For every H such that DH = DG and ∥AH − AG∥ ≤ δ, we +have σ2(D−1/2 +H +AHD−1/2 +H +) ≤ 1 − 1/γ + δ/dmin where dmin is the minimum diagonal entry of DG. +Proof. By lemma B.2, +σ2(D−1/2 +H +AHD−1/2 +H +) ≤ σ2(D−1/2 +G +AGD−1/2) + ∥D−1/2 +G +(AH − AG)D−1/2∥ +≤ 1 − 1/γ + ∥D−1/2 +G +∥2 · ∥AH − AG∥ += 1 − 1/γ + δ/dmin. +Lemma B.4. Let A, �A ∈ Rn×n be the adjacency matrices of Eulerian graphs with no isolated +vertices. Suppose �A +sv≈ε A. Let D be their diagonal matrix of degrees. Then +1 − σ2(D−1/2 �AD−1/2) ≤ (1 + 2ε) · (1 − σ2(D−1/2AD−1/2)). +Proof. Define N = D−1/2AD−1/2 and ˜N = D−1/2 �AD−1/2. +It suffices to show +±(σ2( ˜N) − σ2(N)) ≤ σ2( ˜N − N) ≤ 2ε · (1 − σ2(N)). +The first inequality follows from lemma B.2. For the second inequality, note that the definition of +SV approximation can be rewritten as for all x ∈ Cn, +∥( ˜N − N)x∥ ≤ ε · (1 − ∥Nx∥2). +Note that ˜N, N have a common left and right eigenvector given by v = D1/2⃗1, and this achieves +their respective maximum singular values of 1.7 Hence, if we wish to maximize the LHS over all +unit vectors x, it suffices to maximize over all unit vectors x ⊥ v. Doing so, we obtain +∥ ˜N − N∥ ≤ ε · (1 − σ2(N)2) ≤ 2ε · (1 − σ2(N)) +where the last inequality is because ∥N∥ ≤ 1 ([CKP+17] lemma B.4). +7We know it corresponds to a singular value of 1 and this its not possible to have a singular value larger than 1 +for this matrix ([CKP+17] lemma B.4). +50 + +Lemma B.5. Let σi(·) denote the ith largest singular value. Let A ∈ Rn×n be the adjacency matrix +of a strongly connected Eulerian graph G with diagonal degree matrix D and no isolated vertices. +Let N = D−1/2AD−1/2. Let λ denote the second largest eigenvalue of SN.8 Then we have +σ2((1 − γ)N + γI) ≤ 1 − (1 − λ)γ + O(γ2). +In particular, if γ ≥ 1/poly(n) and A has poly(n) weak mixing time and all weights are specified +using O(log n) bits of precision, then +σ2((1 − γ)N + γI) ≤ 1 − 1/poly(n). +Proof. Let v denote a singular vector of (1 − γ)N + γI corresponding to its second-largest singular +value. We then have +[σ2((1 − γ)N + γI)]2 = (1 − γ)2v∗N∗Nv + 2γ(1 − γ)v∗SNv + γ2 +≤ (1 − γ)2 + 2γ(1 − γ)λ + γ2 +≤ 1 − 2(1 − λ)γ + O(γ2). +In the above, we used the fact that ∥N∥ ≤ 1 ([CKP+17] lemma B.4). +To show the second part of the statement, let τ(M) denote the weak mixing time of the random +walk on the graph corresponding to any Eulerian adjacency matrix M. Define L = I − N. Note +that L is related to I − AD−1 by a change of basis with polynomial condition number. We may +assume the graph is strongly connected as this is necessary for the weak mixing time to be finite. +It suffices to prove +1 +1 − λ ≤ poly(n) · τ(N). +We do so by showing +1 +1 − λ ≤ poly(n) · τ(SN) ≤ poly(n) · ∥S+ +L∥ ≤ poly(n) · ∥L+∥ ≤ poly(n) · τ(N). +The first inequality is just the folklore result relating spectral gap to weak mixing time with an +O(log n)-factor loss. The second and fourth inequalities are from ([CKP+16] Theorem 21). +We now show the third inequality, that ∥S+ +L∥ ≤ poly(n) · ∥L+∥. We have +∥S+ +L∥ ≤ poly(n) · ∥(LT S+ +LL)+∥ = poly(n) · ∥SL+∥ ≤ poly(n) · (∥L+∥ + ∥LT+∥)/2 = poly(n) · ∥L+∥ +where the first inequality in the line immediately above is a corollary of ([CKP+16] Lemma 13). +Lemma B.6. For matrices A, B ∈ Cn×n, +∥AB∥ = ∥(A∗A)1/2B∥, ∥BA∥ = ∥B(AA∗)1/2∥. +Proof. For any matrix M, +∥M∥ = +� +λmax(M∗M) = +max +x∈Cm−{⃗0} +x∗M∗Mx +x∗x +. +Thus, +∥BA∥ = +max +x∈Cm−{⃗0} +x∗B∗A∗ABx +x∗x += +max +x∈Cm−{⃗0} +x∗B∗(A∗A)1/2(A∗A)1/2Bx +x∗x += ∥(A∗A)1/2B∥. +The other equality is proved similarly. +8Here we mean largest according to the actual eigenvalues, not their magnitudes. +51 + diff --git a/i9FRT4oBgHgl3EQfVze1/content/tmp_files/load_file.txt b/i9FRT4oBgHgl3EQfVze1/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..a253d42654a9bda490ee84c2c279b48405617f7b --- /dev/null +++ b/i9FRT4oBgHgl3EQfVze1/content/tmp_files/load_file.txt @@ -0,0 +1,2392 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf,len=2391 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='13541v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='DS] 31 Jan 2023 Singular Value Approximation and Reducing Directed to Undirected Graph Sparsification AmirMahdi Ahmadinejad Amazon∗ ahmadi@alumni.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='stanford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='edu John Peebles Apple peebles@apple.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='com Edward Pyne† MIT epyne@mit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='edu Aaron Sidford Stanford sidford@stanford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='edu Salil Vadhan‡ Harvard University salil vadhan@harvard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='edu February 1, 2023 Abstract In this paper, we introduce a new, spectral notion of approximation between directed graphs, which we call Singular Value (SV) approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SV-approximation is stronger than pre- vious notions of spectral approximation considered in the literature, including spectral ap- proximation of Laplacians for undirected graphs [ST04], standard approximation for directed graphs [CKP+17], and unit-circle approximation for directed graphs [AKM+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, SV approximation enjoys several useful properties not known to be possessed by previous notions of approximation, such as being preserved under products of random-walk matrices and with matrices of bounded norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Notably, we show that there is a simple black-box reduction from SV-sparsifying Eulerian directed graphs to SV-sparsifying undirected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' With this reduction in hand, we provide a nearly linear-time algorithm for SV-sparsifying undirected and hence also Eulerian directed graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This also yields the first nearly linear-time algorithm for unit-circle-sparsifying Eulerian directed graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In addition, we give a nearly linear-time algorithm for SV-sparsifying (and UC-sparsifying) random-walk polynomials of Eulerian directed graphs with second normalized singular value bounded away from 1 by 1/poly(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Finally, we show that a simple repeated-squaring and sparsification algorithm for solving Laplacian systems, introduced by Peng and Spielman [PS14] for undirected graphs, also works for Eulerian digraphs whose random-walk matrix is normal (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' unitarily diagonalizable), if we use SV-sparsification at each step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Prior Laplacian solvers for Eulerian digraphs are significantly more complicated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ∗The work was started prior to joining Amazon and does not relate to Amazon †Supported by an Akamai Presidential Fellowship.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ‡Supported by a Simons Investigator Award.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1 Contents 1 Introduction 3 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Our Contribution: Singular-Value Approximation .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Comparison to Previous Notions of Approximation .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Properties of SV Approximation .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 Algorithmic Results .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 10 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 Open Problems .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 12 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 Roadmap .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 12 2 Preliminaries 13 3 Singular Value Approximation 13 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Matrix Approximation .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 13 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Equivalent definitions of SV approximation .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 15 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Properties of Singular Value Approximation .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 17 4 Sparsification 19 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Cycle and Expander Decompositions .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 21 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Sparsifying Bipartite Expanders .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 22 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Sparsifying a Constant Fraction of Edges .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 24 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 Nearly Linear-Sized Sparsifers .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 25 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 Derandomized Square Sparsification .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 26 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 Sparsification of Random Walk Polynomials .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 27 5 Squaring algorithm for solving normal directed Laplacians 31 A SV Approximation Proofs 39 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Equivalences .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 40 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Properties .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 44 A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Separations .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 48 B Facts about Singular Values 49 2 1 Introduction Given an undirected graph, its associated Laplacian matrix is L = D − A where D is the diagonal matrix of degrees of the graph and A is its associated adjacency matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The linear algebraic properties of Laplacian matrices capture fundamental information about the corresponding graph, including random walks, cuts, and more.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Correspondingly, starting with the seminal work of Spiel- man and Teng [ST04], nearly-linear time algorithms for solving Laplacian linear systems and related problems [ST04, BSS12, CKP+16, CKP+17, CKK+18] have been established as an important tool in algorithmic graph theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' These results have enabled asymptotic improvements for a variety of problems, such as max-flow and min-cut [CKM+11,LRS13,KLOS14,vdBGJ+22,vdBLL+21], travel- ing salesman [AGM+10,AG15], sampling random spanning trees [KM09,MST15], graph sparsifica- tion [SS08,LS13,KMP14,CKM+14,LS18,ABKS21,JS21], multicommodity flow [KMP12,KLOS14], and solving clustering and partitioning problems [AM85,KVV04,ACL06,OSV12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This “Laplacian paradigm” was initially formulated for undirected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' However, there is a natural generalization of the Laplacian matrix to directed graphs, which analogously captures the information about random walks on the directed graph [CKP+16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This generalization is L = Dout − A⊤ where Dout is now the diagonal matrix of out-degrees and A⊤ is the transpose of the adjacency matrix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It is called the directed Laplacian matrix, to emphasize that the graph corresponding to it is directed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' While the generalization of the definition of the Laplacian matrix to directed graphs is straight- forward, the generalization of algorithms for Laplacian matrices is not so straightforward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Nonethe- less, a sequence of papers [CKP+16,CKP+17,CKK+18] obtained nearly linear time algorithms for sparsifying and solving directed Laplacian linear systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In order to do so, they needed to over- come three main difficulties not present in the undirected setting: Challenge #1: While the kernel of an undirected Laplacian matrix is the all ones vector, comput- ing the kernel of a directed Laplacian matrix L corresponds to computing the stationary distribution π of the random walk on the directed graph (LD−1 outπ = 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Without explicitly knowing the kernel, it is not known how to efficiently perform any kind of useful sparsification or approximately solve systems in L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This difficulty was overcome in [CKP+16, AJSS19] which provide reductions from solving general directed Laplacian systems to the case where the graph is Eulerian, meaning that every vertex has the same in-degree as out-degree.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Eulerian graphs, the stationary distribution is simply proportional to the vertex degrees and the all ones vector is both the left and right kernels of the associated directed Laplacian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Challenge #2: Undirected Laplacians L are not only symmetric but also positive semidefinite (PSD), i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' x⊤Lx ≥ 0 for all x .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This leads to a natural definition of multiplicative approximation between such matrices, namely saying that �L is an ε-approximation of L if (1−ε)L ⪯ �L ⪯ (1+ε)L, where ⪯ is the L¨owner order on PSD matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' However, even though Laplacians of directed graphs are potentially asymmetric, the quadratic form x⊤Lx depends only on a symmetrization of the Laplacian (x⊤Lx = x⊤(1/2)(L + L⊤))x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, the quadratic form discards key information about the associated directed graph (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' the quadratic form of a directed cycle and an undirected cycle are the same).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, defining approximation for directed graphs (even Eulerian ones) is more challenging than for undirected graphs and more complex notions of approximations were introduced in [CKP+16, AKM+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This additional complexity suggest that one formulate new sparsification algorithms that take into account the directedness of the graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 Challenge #3: Finally, the recursive identities used for solving undirected Laplacian systems, while behaving nicely with respect to PSD approximation, do not behave as nicely with respect to the new directed matrix approximation definitions mentioned in Challenge #2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This necessi- tates different, more sophisticated recursion with a more involved analysis of the error [CKP+17, CKK+18,AKM+20,KMG22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Our Contribution: Singular-Value Approximation In this paper, we present a stronger and more robust notion for addressing Challenge #2 and use it make progress in simplifying approaches to Challenge #3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We do not address Challenge #1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' rather the existing reductions from general directed graphs to Eulerian ones allows us to focus on Eulerian digraphs without loss of generality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Specifically, we introduce a novel definition of approximation for asymmetric matrices, which we call singular-value approximation (or SV approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For simplicity in the introduction, we focus on the case of regular directed graphs, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' directed graphs where for some value d ≥ 0, every vertex has in-degree d and out-degree d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (In the case of digraphs with nonnegative edge weights, we obtain the in- and out-degrees by summing the in-coming or out-going edge weights at each vertex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=') However, all of our results generalize to Eulerian digraphs and some generalize to wider classes of complex matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To introduce SV approximation, let A be the adjacency matrix of a d-regular digraph, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A is a nonnegative real matrix where every row and column sum equals d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the (in- and out-) degree matrix is simply dI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Dividing by d, it is equivalent to study approximation of the random-walk matrix W = A/d, which is doubly stochastic, and has degree matrix I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 (SV approximation for doubly stochastic matrices).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For doubly stochastic matrices W, � W ∈ Rn×n we say that � W is an ε-singular-value (SV) approximation of W, written � W sv≈ε W, if for all vectors x, y ∈ Rn, we have ���x⊤(� W − W)y ��� ≤ ε 4 · � x⊤(I − WW⊤)x + y⊤(I − W⊤W)y � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (1) This formulation of SV-approximation is one of several equivalent formulations we provide later in the paper (Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We can equivalently define SV approximation between doubly stochastic matrices by requiring Equation (1) to hold for all complex test vectors x, y ∈ Cn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SV-approximation can also be defined equivalently by replacing condition (1) with ���x⊤(� W − W)y ��� ≤ ε 4 · � [x⊤(I − WW⊤)x] · [y⊤(I − W⊤W)y].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (2) These two formulations, (1) and (2), differ only in using the geometric mean or the arithmetic mean of the terms involving x and y on the right-hand side.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The formulation in terms of the geometric mean implies the one in terms of the arithmetic mean (since the geometric mean is no larger than the arithmetic mean);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' the converse follows by optimizing over scalar multiples of x and y (as was done in e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CKP+17,AKM+20]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Both formulations can also be rewritten more simply by noting that x⊤(I − WW⊤)x = ∥x∥2 − ∥x⊤W∥2 and y⊤(I − W⊤W)y = ∥y∥2 − ∥Wy∥2, but the description in terms of quadratic forms will be more convenient for comparison with previous notions of approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Section 3, we provide more general definitions of SV approximation which also apply to unnormalized directed Laplacians and even to complex matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4 We prove that SV approximation is strictly stronger than previous notions of spectral approxi- mation considered in the literature for both asymmetric and symmetric matrices, and enjoys several useful properties not possessed by the previous notions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Most notably, there is a simple black-box reduction from SV-sparsifying Eulerian directed graphs to SV-sparsifying undirected graphs;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' no such reduction is known for prior notions of asymmetric spectral approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, we give efficient algorithms for working with SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' These include nearly linear-time algorithms for SV-sparsifying undirected and hence also Eulerian directed graphs (Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8), as well as random-walk polynomials associated with such graphs (Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We also show that a simple repeated-squaring and sparsification algorithm for solving Laplacian systems also works for Eulerian digraphs whose random-walk matrix is normal (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' unitarily diagonalizable), if we use SV-sparsification at each step (Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Prior Laplacian solvers for Eulerian graphs are more complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We elaborate on these results in the next several subsections.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Comparison to Previous Notions of Approximation Let us compare Definition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 to previous definitions of approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Undirected spectral approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let’s start with the undirected case, where W = W⊤.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In this case, it can be shown that we can without loss of generality restrict to x = y ∈ Rn, and � W sv≈ε W requires that for all x ∈ Rn, ���x⊤(� W − W)x ��� ≤ ε 2 · � x⊤(I − W2)x � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (3) In contrast, the standard definition of spectral approximation (originally introduced by Spielman and Teng [ST04]), which we denote by � W ≈ε W, is equivalent to requiring that for all x ∈ Rn, we have ���x⊤(� W − W)x ��� ≤ ε · � x⊤(I − W)x � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (4) To compare inequalities (3) and (4), we write x = � i civi, where v1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , vn is an orthonormal eigenbasis for W with associated eigenvalues λ1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , λn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Since W is stochastic, |λi| ≤ 1 for all i ∈ [n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the right-hand side of SV inequality (3) becomes ε 2 · � i∈[n] c2 i · (1 − λ2 i ), whereas the right-hand side of ST inequality (4) becomes ε · � i∈[n] c2 i · (1 − λi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Since each |λi| ≤ 1, the fact that SV approximation implies ST approximation then follows from (1 − λ2 i ) = (1 − λi)(1 + λi) ≤ 2(1 − λi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' However, we also see that inequality (3) can be much stronger than inequality (4) when W has eigenvalues λi close to -1 (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' in a bipartite graph with poor expansion) because then 1 − λ2 i is close to 0, but 1−λi is bigger than 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' More generally, inequality (3) requires that � W approximates W very well on every test vector x that is concentrated on the eigenvectors whose eigenvalues have magnitude close to 1, whereas inequality (4) only requires close approximation on the (signed) eigenvalues that are close to 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 We remark that another way of ensuring � W preserves unit singular values is to replace W2 in the SV inequality (3) with the matrix |W| where we replace all eigenvalues of W with their absolute value rather than their square,1 so that we have: x⊤(I − |W|)x = � i∈[n] c2 i · (1 − |λi|).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Using |W| instead of W2 results in an equivalent definition up to a factor of 2 in ε, and W2 turns out to be convenient to work with.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 This viewpoint also explains why we stop at W2 in the definition and don’t explicitly use higher powers;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' it is simply a convenient proxy for |W|, which captures all powers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Indeed, for all k ∈ N I − W2 ⪯ 2 · (1 − |W|) ⪯ 2 · � I − Wk� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Directed spectral approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Turning to previous notions of spectral approximation for directed graphs, standard approximation [CKP+17] generalizes the definition of Spielman and Teng [ST04] by saying � W ≈ε W if for all x, y ∈ Rn ���x⊤(� W − W)y ��� ≤ ε 2 · � x⊤(I − W)x + y⊤(I − W)y � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (5) Equivalently, we can require that for all x, y ∈ Rn, ���x⊤(� W − W)y ��� ≤ ε · � [x⊤(I − W)x] · [y⊤(I − W)y].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (6) It can be shown that for undirected graphs, standard approximation is equivalent to the condition of Equation 4, so we will also refer to it as standard approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The use of different left and right test vectors x and y on the left-hand side is crucial for capturing the asymmetric information in � W and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' As before, if x or y is concentrated on eigenvectors of W whose eigenvalues are close to 1, then the right-hand side of ST inequality (6) is close to 0 and � W must approximate W very well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' However, like the standard undirected ST inequality (4), not much is required on eigenvalues near -1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, asymmetric matrices can have eigenvalues that are not real and are equal to or close to complex numbers of magnitude 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For example, the eigenvalues of a directed n-cycle are the complex n’th roots of unity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To address this issue, unit-circle (UC) approximation [AKM+20], written � W ≈ε W, requires that for all complex test vectors x, y ∈ Cn, we have ���x∗(� W − W)y ��� ≤ ε 2 · � ∥x∥2 + ∥y∥2 − |x∗Wx + y∗Wy| � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (7) That is, we take the complex magnitude of the terms involving W on the right-hand side of ST inequality (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' That way, if x and y are concentrated on eigenvectors of W that have eigenvalue near some complex number µ of magnitude 1, we require that � W approximates W very well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For example, consider the case where W is normal, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' has an orthonormal basis of complex eigenvectors v1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , vn and with corresponding complex eigenvalues λ1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , λn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then if we write x = � i civi and y = � i divi, the right-hand side of UC inequality (7) becomes: ε 2 · � i∈[n] (|ci|2 + |di|2) − ������ � i∈[n] (|ci|2 + |di|2) · λi ������ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (8) 1Another way of describing |W| is as the psd square root of the psd matrix W2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2|W| = (W2)1/2, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' |W| is the PSD square root of W2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Definition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1, we could similarly replace WW⊤ and W⊤W with their PSD square roots and obtain a definition that is equivalent up to a factor of 2 in ε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 6 If x and y are concentrated on eigenvalues λi ≈ µ where |µ| = 1, then this expression will be close to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Unit-circle approximation has valuable properties not enjoyed by standard approximation, in particular being preserved under powering: If � W is an ε-UC approximation of W, then for every positive integer k, � Wk is an O(ε)-UC approximation of Wk;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' note that the quality of approximation does not degrade with k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This property was crucial for the results of [AKM+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' However, UC approximation has two limitations compared to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, UC expression (8) is only small if x and/or y is concentrated on eigenvalues that are all close to the same point µ on the complex unit circle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Even in the undirected case, if x and y are mixtures of eigenvectors with eigenvalue close to 1 and eigenvalue close to -1, then there will be cancellations in the second term of UC expression (8) and the result will not be small.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Second, the spectral properties of asymmetric matrices are better captured by singular values than eigenvalues, since singular values treat the domain and codomain as distinct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To see how SV approximation addresses these limitations, let σ1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , σn ≥ 0 be the singu- lar values of W, let u1, u2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , un ∈ Cn the corresponding left-singular vectors of W, and let v1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , vn ∈ Cn the corresponding right-singular vectors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If we write x = � i ciui and y = � i divi, then the right-hand side of SV inequality (1) becomes: ε 4 · \uf8ee \uf8f0� i∈[n] (|ci|2 + |di|2) · (1 − σ2 i ) \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (9) Consequently, SV-approximation requires high-quality approximation if x is concentrated on left- singular vectors of singular value close to 1 and/or y is concentrated on right-singular vectors of singular value close to 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (For the “or” interpretation, use the formulation of SV approximation in terms of inequality (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=') To compare with UC expression (8), let us consider what happens with a normal matrix, where ui = vi and σi = |λi|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In this case, SV expression (9) amounts to bringing the absolute value of UC expression (8) inside the summation (and squaring, which only makes a factor of 2 difference), to avoid cancellations between eigenvalues of different phases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, for non-normal matrices, SV approximation retains the asymmetry of W even on the right-hand side, by always using x on the left of W (thus relating to its decomposition into left singular vectors) and y on the right of W (thus relating to its decomposition into right singular vectors).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Indeed, this feature allows us to even extend the definition of SV approximation to non-square matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (See Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=') Following the above intuitions, we prove that SV approximation is indeed strictly stronger than the previous notions of approximation: Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For all doubly stochastic W and � W, if � W sv≈ε W, then � W ≈ε W (and hence � W ≈ε W).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' On the other hand, for every ε ∈ (0, 1) and ε′ ∈ (ε, ∞), there exist symmetric W, � W ∈ R2×2 of spectral norm at most 1 such that � W ≈ε W but it is not the case that � W svn ≈ ε′ W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The separation above relies on a generalization of the definition of SV approximation to ma- trices that can have negative entries, given in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It would be interesting to extend the separation to matrices corresponding to graphs, and even doubly stochastic matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We note that in [AKM+20] UC approximation was separated from standard approximation even for doubly stochastic matrices, so SV approximation is also strictly stronger than standard approximation, even for doubly stochastic matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Other Work on SV Approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The definition of SV approximation and some of our re- sults on it (obtained in collaboration between Jack Murtagh and the authors) were first presented in 7 the first author’s PhD thesis [Ahm20] in August 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Independently, Kelley [Kel21] used a variant of SV approximation to present an alternative proof of a result of [HPV21], who used unit-circle approximation to prove that the Impagliazzo-Nisan–Wigderson pseudorandom generator [INW94] fools permutation branching programs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Golowich and Vadhan [GV22] used SV approximation and some of our results (presented in the aforementioned thesis) to prove new pseudorandomness prop- erties of expander walks against permutation branching programs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Most recently, Chen, Lyu, Tal, and Wu [CLTW22] have used a form of SV approximation to present alternative proofs of the results of [AKM+20,PV21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Properties of SV Approximation SV approximation enjoys a number of novel properties not known to be possessed by previous notions of spectral approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Most striking is the fact that directed approximation reduces to undirected approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To formulate this, we define the symmetric lift of a matrix: Definition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given W ∈ Cm×n, let the symmetric lift of W be defined as slift (W) def = �0n×n W∗ W 0m×m � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Graph theoretically, the symmetric lift of W is the following standard operation: Given our directed graph G on n vertices with random-walk matrix W, we lift G to an undirected bipartite graph H with n vertices on each side, where we connect left-vertex i to right-vertex j if there is a directed edge from i to j in G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then slift (W) is the random-walk matrix of H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let W and � W be doubly stochastic matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then W sv≈ε � W if and only if slift � � W � sv≈ε slift (W).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, for the first time (as far as we know), sparsification of directed graphs reduces directly to sparsification of undirected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It would be very interesting to obtain a similar reduction for other algorithmic problems in spectral graph theory, such as solving Laplacian systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Another novel property of SV approximation is that it is preserved under products of random- walk matrices: Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let W1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Wk and � W1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , � Wk be doubly stochastic matrices such that � Wi sv≈εWi for each i ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then � W1 � W2 · · · � Wk sv≈ε+O(ε2) W1W2 · · · Wk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Notably the approximation error does not grow with the number k of matrices being multiplied.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The analogous property was known for UC approximation only in the special case of powering, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' W1 = W2 = · · · = Wk and � W1 = � W2 = · · · = � Wk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In addition, SV approximation is preserved under multiplication on the left and right by ar- bitrary matrices of bounded spectral norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Indeed, it can be seen as the “closure” of standard approximation under this operation (up to a factor of 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let W and � W be doubly stochastic matrices such that � W sv≈εW.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for all complex matrices U and V of spectral norm at most 1, we have U� WV sv≈ε UWV, and hence U� WV ≈ε UWV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let W and � W be doubly stochastic matrices such that for all complex matrices U and V of spectral norm at most 1, we have U� WV ≈ε UWV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then � W sv≈2ε W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 8 Since UWV and U� WV need not be doubly stochastic matrices, Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 uses the gener- alization of SV approximation to more general matrices, which can be found in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Recall that standard spectral sparsifiers [ST04] are also cut sparsifiers [BK00].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' That is, if �G is an ε-approximation of G, then for every set S of vertices, the weight of the cut S in �G is within a (1 ± ε) factor of the weight of S in G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Indeed, if we take the test vector x to be the characteristic vector of the set S in inequality (4), we obtain ���� Pr (i,j)∼µedge [i ∈ S, j /∈ S] − Pr (i,j)∼�µedge [i ∈ S, j /∈ S] ���� ≤ ε · Pr (i,j)∼µedge [i ∈ S, j /∈ S], (10) where µedge def = µedge(W) (resp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �µedge def = �µedge(� W)) denotes a random edge according to the stationary distribution of W (resp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' � W), namely pick i ∈ [n] uniformly at random and obtain j as a random neighbor of i using W (resp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' � W).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that Pr(i,j)∼µedge[i ∈ S, j /∈ S] is simply a normalized version of the weight of the cut (S, Sc);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' namely it equals the sum of the edge weights leaving S divided the sum of all edge weights in the graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Similarly, we can obtain a combinatorial consequence of SV approximation, by taking x to be a characteristic vector of a set S of vertices and taking y to be a characteristic vector of a set T of vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This yields: Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let � W and W be doubly stochastic n × n random-walk matrices and suppose that � W sv≈ε W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for every two subsets S, T ⊆ [n], we have ���� Pr (i,j)∼µedge [i ∈ S, j ∈ T] − Pr (i,j)∼�µedge [i ∈ S, j ∈ T] ���� ≤ ε 2· � Pr (i,j,k)∼µfb [i ∈ S, k /∈ S] · Pr (i,j,k)∼µfb [i ∈ T, k /∈ T], where: µedge def = µedge(W) (resp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �µedge def = �µedge(� W)) denotes a random edge according to the station- ary distribution of W (resp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' � W), namely pick i ∈ [n] uniformly at random and obtain j as a random neighbor of i using W (resp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' � W), and µfb def = µfb(W) denotes a random forward-backward walk, namely pick i ∈ [n] uniformly at random, obtain j as a random neighbor of i using W, and k as a random neighbor of j using W⊤.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let us interpret Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, consider the case that W = J, the matrix with every entry equal to 1/n (the random-walk matrix for the complete graph with self-loops).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then (i, j) ∼ µedge and (i, j, k) ∼ µfb are all uniform and independent vertices, so Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 says: ����µ(S) · µ(T) − Pr (i,j)∼�µedge [i ∈ S, j ∈ T] ���� ≤ ε 2 · � µ(S) · (1 − µ(S)) · µ(T) · (1 − µ(T)), where µ(S) = |S|/n and µ(T) = |T|/n are the stationary probabilities of S and T, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This amounts to a restatement of the Expander Mixing Lemma ([Vad12, Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='15]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' indeed � W sv≈ε J if and only if � W is a spectral expander with all nontrivial singular values at most ε/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next, let’s consider the case that T = Sc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then, Pr (i,j,k)∼µfb [i ∈ S, k /∈ S] = Pr (i,j,k)∼µfb [i ∈ T, k /∈ T].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 9 (In an Eulerian graph, the weight of edges crossing any cut in one direction must equal the weight of edges crossing it in the other direction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=') Thus, SV approximation implies that: ���� Pr (i,j)∼µedge [i ∈ S, j /∈ S] − Pr (i,j)∼�µedge [i ∈ S, j /∈ S] ���� ≤ ε 2 · Pr (i,j,k)∼µfb [i ∈ S, k /∈ S].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (11) We claim that (11) is stronger than the standard notion of a cut approximator (10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Indeed, Pr (i,j,k)∼µfb [i ∈ S, k /∈ S] ≤ Pr (i,j,k)∼µfb [i ∈ S, j /∈ S] + Pr (i,j,k)∼µfb [j ∈ S, k /∈ S] = 2 Pr (i,j)∼µedge [i ∈ S, j /∈ S].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (12) Similarly, we have Pr (i,j,k)∼µfb [i ∈ S, k /∈ S] ≤ Pr (i,j)∼µedge [i ∈ S, j ∈ S] + Pr (i,j)∼µedge [i /∈ S, j /∈ S] (13) Thus, we conclude that an SV-approximator not only approximates every cut to within a small additive error that is scaled by the weight of the cut edges (as in (10)), but also scaled by the weight of the uncut edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 Algorithmic Results Even though SV approximation is stronger than previously considered notions of spectral approx- imation, we show that it still admits sparsification: Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given a regular directed graph G with n vertices and m edges, integer edge weights in [0, U], and random-walk matrix W, and ε > 0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There exists a weighted graph �G with at most O(nε−2 log(nU) log11 n) edges such that its random-walk matrix � W satisfies � W sv≈ε W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There is a randomized nearly-linear time algorithm that whp outputs a weighted graph �G with at most O(nε−2 log(nU) log19 n) edges such that its random-walk matrix � W satisfies � W sv≈εW.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A more general theorem that also applies to Eulerian digraphs is stated in the main body of the paper (Corollaries 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12 and 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='13).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Prior to this work, it was open whether or not even UC-sparsifiers with O(n · poly(log n, ǫ−1)) edges existed for all unweighted regular digraphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Instead, it was only known how to UC-sparsify powers of a random walk matrix to bound the sparsity increase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 By Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4, it suffices to prove Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8 for undirected bipartite graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We obtain the latter via an undirected sparsification algorithms based on expander partitioning [ST04].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It remains an open question whether algorithms based on edge sampling can yield SV approximation or unit circle approximation, even in undirected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The standard approach to spectral sparsification of undirected graphs via sampling, namely keeping each edge independently with probability pro- portional to its effective resistance [SS08], does not work for SV or UC approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For example, this method does not exactly preserve degrees, which we show is necessary for SV sparsification (Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 However, we remark that the work of Chu, Gao, Peng, Sawlani, and Wang [CGP+18] does yield something closer to sparsification via degree preserving sampling for standard approxima- tion [CKP+17] but not unit circle approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' They show that if one has a directed graph and 3We extend this result to SV approximation (Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='15).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4Unlike standard spectral approximation, degrees cannot be fixed just by adding self loops;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' indeed, self-loops ruin bipartiteness and periodicity, which are properties that UC and SV approximation retain (as they are captured by eigenvalues like -1 or other roots of unity).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 10 decomposes it into short “cycles” without regard for the direction of the edges on the cycle, then one can sparsify by randomly eliminating either the clockwise or counterclockwise edges on each such cycle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We build on their procedure and use it to obtain SV sparsification (and hence, unit circle) by showing that this technique obtains SV approximation, even if the cycles are not short, as long as (a) all the cycles are within expanding subgraphs, and (b) the cycles alternate between forward and backward edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (Note that such alternating cycles in a directed graph correspond to ordinary cycles in the undirected lift given by Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=') Sparsity Approximation Time Subgraph?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Citation O(nε−2 logc n) Standard O(m logc n) No [CKP+17] O(nε−2 logc n) Standard O(m1+o(1)) Yes [CGP+18,PY19] O(nε−2 log4 n) Standard O(nm) Yes [CGP+18] O(nε−2 log12 n) SV Existential Yes Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='13 O(nε−2 log20 n) SV O(m log7 n) Yes Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12 Our results also enable us to give the first nearly linear-time algorithm for sparsifying powers of random walk matrices of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' More precisely, we show the following.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9 (Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17, informally stated).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There is a randomized nearly-linear time al- gorithm that, given a regular digraph G with random walk matrix W and any positive integer k ≤ poly(n), outputs a nearly linear sized SV-sparsifer of Wk, provided that the second largest singular value of W is at most 1 − 1/poly(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This theorem is stated more generally for Eulerian digraphs and proven as Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A regular (or even Eulerian) digraph whose weak mixing time (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' the mixing time of the 1/2- lazy random walk) is at most poly(n) can be made to have second largest singular value at most 1 − 1/poly(n) by making it 1/poly(n)-lazy (Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus we obtain the following corollary that allows us to unconditionally sparsify lazy random walks with even very small amounts of laziness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Corollary 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10 (Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19, informally stated).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G be a regular digraph with poly(n) weak mixing time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For any positive integer k ≤ poly(n) and γ ≥ 1/poly(n), we can, in nearly linear time, output a nearly linear sized SV-sparsifer of ((1 − γ)W + γI)k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Importantly we can take γ ≪ 1/k, so a γ-lazy walk of length k is very close (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' in total variation distance) to an ordinary random walk of length k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Our algorithm is relatively simple: we show that we can SV-sparsify the square of random walk matrices of regular graphs, which gives all powers of two, then multiply different such powers together to get the precise exponent we want.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For example, to get a sparsifier for W7, we multiply sparsifiers for W4, W2, and W1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The use of SV approximation plays an important role in the analysis of this algorithm, because it has the property that the product of the approximations of the powers still approximates the product of the true powers (Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Finally for Challenge #3, we make progress towards simplifying the recursion and analysis of solving directed Laplacian linear system in the following way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We show that a simpler recursion, a variant of the Peng-Spielman repeated squaring based recrusion (Equation (15)) and analysis suffice if the directed Laplacian is normal (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' unitarily diagonalizable).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A solver of this form was given in the undirected case by Peng and Spielman [PS14], but the existing solvers for directed graphs are all significantly more involved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that this result on the squaring recursion is the only result in this paper that relies on a normality assumption, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' aforementioned sparsification results hold for all Eulerian directed graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 11 Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a doubly stochastic normal matrix W ∈ Rn×n with ∥W∥ ≤ 1, let W = W0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Wk−1 be a sequence of matrices such that for ε ≤ 1/4k we have Wi sv≈ǫ W2 i−1 ∀0 < i < k, (14) and Pi = 1 2 [I + (I + Wi)Pi+1(I + Wi)] ∀0 ≤ i < k (15) defining the Peng-Spielman squaring recursion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then, for a matrix Pk, such that ���(I − W2k) 1 2 � Pk − (I − W2k)+� (I − W2k) 1 2 ��� ≤ O(kǫ), we have ∥P0(I − W) − I∥B = ���(I − W) 1 2 � P0 − (I − W)+� (I − W) 1 2 ��� ≤ O(k2ǫ) (16) where B = ((I − W)1/2)∗(I − W)1/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11 says that that we can compute a good preconditioner P0 for the Laplacian I−W (eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 16) by repeatedly computing SV-approximate squares (eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 14) and use the simple recurrence (eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 15, starting with a preconditioner Pk for a sufficiently large power of W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Generally, Pk is easy to obtain for a large enough k = O(log n) since W2k is well-approximated by a complete graph (assuming the original graph is connected and aperiodic).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 Open Problems One open problem is to determine whether or not it is possible to obtain linear-sized sparsifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Recall that undirected graphs have sparsifiers with respect to standard spectral approximation that have only O(n/ε2) nonzero edge weights [BSS12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If this result could be extended to obtain linear- sized SV-sparsifiers of undirected graphs, we would also have linear-sized sparsifiers for directed graphs by Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4, which would be a new result even for standard approximation [CKP+17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 Roadmap The rest of the paper is organized as follows: Section 2 gives definitions we use throughout the paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Section 3 defines SV-approximation and provides several equivalent characterizations of it which are useful when analyzing algorithms and also interesting in their own right.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' These results immediately imply a reduction from the directed to undirected case for SV approxi- mation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Section 4 gives a direct SV sparsification algorithm based on sparsifying alternating cycles within expanders, and shows that the derandomized square gives SV square sparsifiers, and shows how to sparsify powers of random walk matrices with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Section 5 gives our simpler algorithm and analysis for solving Laplacian linear systems when the matrix is unitarily diagonalizable (normal).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Appendix A, we prove equivalences, properties, and separations, including several deferred proofs from the body of the paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 12 2 Preliminaries We begin by providing notation and definitions that we will use throughout the paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Complex Numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let C denote the set of complex numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For z def = a+bi ∈ C, let z∗ def = a−bi and ℜ[z] def = a and ℑ[z] def = b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Define the magnitude of z as |z| def = √ z∗z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Matrix Notation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For x ∈ Cn we let diag(x) ∈ Cn×n denote the diagonal matrix with [diag(x)]i,i = xi for all i ∈ [n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For complex matrix M that is not necessarily square, we write M∗ to denote the conjugate transpose of M and M+ to denote the Moore-Penrose pseudoinverse of M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We let lker(M) and rker(M) be the left and right kernels of M respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We say M ∈ Cn×n is Hermi- tian if M∗ = M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' M is positive semidefinite (PSD) if it is Hermitian and for every vector x ∈ Cn, we have x∗Mx ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' M is normal if M∗M = MM∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a square matrix M, let SM = (M+M∗)/2 denote its symmeterization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that SM is Hermitian, and if M is a scalar then SM = ℜ[M].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given any M ∈ Cn×n and subsets S, T ⊂ [n], we let MS,T ∈ CS×T be the submatrix of M with rows specified by S and columns specified by T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We let Sc def = [n] \\ S be the complement of S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Schur Complements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given an n × n matrix M and S ⊂ [n], we define the Schur Complement of M onto S as SCS(M) = MSc,Sc − MSc,SM+ S,SMS,Sc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Eigenvalues and Singular Values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a symmetric matrix A with nonnegative entries, let λi(A) denote the ith largest eigenvalue of A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a matrix A, let σi(A) def = λi(A∗A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given an undirected graph G with adjacency matrix A and degree matrix D def = diag(A⃗1), let λ(G) def = λ2(D+/2AD+/2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Norms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Throughout we use ∥ · ∥ to denote the spectral norm, where for any A ∈ Cn×m, ∥A∥ def = sup x∈Cm\\{⃗0} ∥Ax∥ ∥x∥ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' L¨owner Order.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given Hermitian A, B ∈ Cn×n we write A ⪯ B if B − A is PSD, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' for every x ∈ Cn we have x∗Ax ≤ x∗Bx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 Singular Value Approximation In this section we formally define the notions of approximation we work with throughout the paper (Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1), provide a number of equivalent definitions of SV approximation (Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2), and give key properties of SV approximation (Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' All proofs in this section are deferred to their analogous subsection in Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Matrix Approximation First we define a general notion of matrix approximation with respect to arbitrary PSD matrices (Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This definition and the equivalences are a generalization of those established in [CKP+17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 13 Intuitively, we say that a matrix �A is an ǫ-approximation of A with respect to error matrices E and F if we can bound the bilinear form of their difference, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' x∗( �A−A)y, by the quadratic forms of x with E and y with F, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' x∗Ex and y∗Fy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We use the term error matrix here to distinguish this notion of approximation from standard approximation (Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3) which we define later.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 (Matrix approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n, and let E ∈ Cm×m, F ∈ Cn×n be PSD matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ε ≥ 0, we say that �A is an ε-approximation of A with respect to error matrices E and F if any of the following equivalent conditions hold: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ���x∗( �A − A)y ��� ≤ ε 2 (x∗Ex + y∗Fy) for all x ∈ Cm, y ∈ Cn 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ���x∗( �A − A)y ��� ≤ ε · √ x∗Ex · √y∗Fy for all x ∈ Cm, y ∈ Cn 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ���E+/2( �A − A)F+/2��� ≤ ε, lker( �A − A) ⊇ ker(E), and rker( �A − A) ⊇ ker(F).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If E = F, we say that �A is an ε-approximation of A with respect to error matrix E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the following Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 we specialize and simplify the equivalences presented in Defini- tion 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 to Hermitian matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 (Hermitian matrix approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×m be Hermitian matrices and let E ∈ Cm×m be a PSD matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the following are equivalent conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �A is an ε-approximation of A with respect to error matrix E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For all x ∈ Cm we have ���x∗( �A − A)x ��� ≤ ε · (x∗Ex).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' −εE ⪯ �A − A ⪯ εE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next, for matrices A, �A, D ∈ Cm×m we define a natural notion of approximation between D − A and D − �A which we call standard approximation (Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The choice of this name is because when A and �A are adjacency matrices of undirected graphs with the same degrees, standard approximation coincides with spectral approximation of the Laplacian matrices of the associated graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Further, when A and �A are adjacency matrices of directed Eulerian digraphs with the same degrees then standard approximation coincides with ǫ-approximation from [CKP+17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Standard approximation generalizes these two cases in a natural way even when D, which we call a degree matrix, is not diagonal (as it was in these two cases and as we will often choose it to be).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We use the term degree matrix to emphasize when we are using standard approximation (Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3) and its variants rather than (Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We now present a series of definitions and equivalences and properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For clarity, the reader can initially think of the case that D = I and A = W is a doubly stochastic matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 (Standard approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×m and D ∈ Cm×m be a PSD matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ε ≥ 0, we say �A is a standard ε-approximation of A with respect to degree matrix D if E = D − SA is PSD and �A is an ε-approximation of A with respect to error matrix E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 5Recall by Section 2 that SA def = (A + A∗)/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 14 With matrix approximation and standard approximation established, we can now define unit- circle (UC) approximation and singular-value (SV) approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' UC approximation as we present it here, is a generalization of UC approximation as it was introduced for random walk matrices in [AKM+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' As discussed in the introduction, SV approximation is a new notion of approximation introduced in this paper;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' it has a number of natural desirable properties and facilitates our results on sparsification (Section 4) and linear system solving (Section 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 (UC approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×m and D ∈ Cm×m be PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ε ≥ 0, we say �A is a unit-circle (UC) ε-approximation of A with respect to degree matrix D if for every z ∈ C with |z| = 1, z �A is a standard ε-approximation of zA with respect to degree matrix D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If this holds, we write �A ≈ε A with respect to D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We omit “respect to D” when D = I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that for doubly stochastic matrices � W, W, the above definition (with D = I) corresponds to requiring for every z ∈ C and x, y ∈ Cn, ���x∗(z� W − zW)y ��� ≤ ε 2 � ∥x∥2 + ∥y∥2 − x∗zWx − y∗zWy � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (17) Optimizing over z in (17) implies Equation (7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 (SV approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ε ≥ 0, we say �A is a ε-singular-value (SV) approximation of A with respect to degree matrices Din and Dout, if 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ker(Din) ⊆ lker(A), and ker(Dout) ⊆ rker(A), 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' E = Din − AD+ outA∗ and F = Dout − A∗D+ inA are PSD, and 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �A is an ε/2-approximation of A with respect to error matrices E and F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If this holds, we write �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If A ∈ Rn×m ≥0 and Din = diag(A⃗1n), Dout = diag(⃗1⊤ mA), then we write �A sv≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If Din = Im and Dout = In, then we write �A svn ≈ ε A, which we call normalized SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If m = n and Din = Dout = D, then we write �A D≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that normalized SV approximation does not require the relevant matrices to be non- negative, whereas �A sv≈ε A is only defined for real non-negative matrices (such as the adjacency matrices of graphs).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The maximally general definition captures both cases, so we will prove prop- erties with respect to this notion and note their implications for the specialized notions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Equivalent definitions of SV approximation Here we give several equivalent formulations of SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first give conditions under which the error matrices E and F in the definition of SV approximation are PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 (Conditions for SV approximation to be defined).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n, and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices such that ker(Din) ⊆ lker(A) and ker(Dout) ⊆ rker(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the following are equivalent: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' σmax(D+/2 in AD+/2 out ) ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Din − AD+ outA∗ is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 15 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Dout − A∗D+ inA is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For some scalar z ∈ C with |z| = 1, � Din zA z∗A∗ Dout � is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every scalar z ∈ C with |z| ≤ 1, � Din zA z∗A∗ Dout � is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Suppose further that Din = diag(din) and Din = diag(dout) for din ∈ Rm ≥0, dout ∈ Rn ≥0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then Condition 1 below implies Condition 2 below, which implies Conditions 1–5 above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A is nonnegative, dout = A⃗1n, din = ⃗1⊤ mA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For all i ∈ [n], (dout)i ≥ � j |Ai,j|, and for all j ∈ [m], (din)j ≥ � i |Ai,j|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next we give several equivalent definitions of SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 (Equivalent formulations of SV approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the following are equivalent 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' slift � �A � D≈ε slift � �B � , where slift � �A � = �0m×m A A∗ 0n×n � , slift (A) = � 0m×m �A �A∗ 0n×n � , D = � Din 0m×n 0n×m Dout � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For some scalar z ∈ C with |z| = 1, ˜C is ε/2-approximation of C with respect to error matrix E, where C = �0m×m zA 0n×m 0n×n � , ˜C = � 0m×m z �A 0n×m 0n×n � , E = � Din zA z∗A∗ Dout � 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Item 3 holds for every z ∈ C such that |z| ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Each formulation of SV approximation in Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 has useful properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Item 2 implies that SV approximation between directed graphs is equivalent to a natural related statement between undirected graphs, which we use for sparsification (See Section 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A version of Item 2 is not known to hold for prior definitions of approximation between directed graphs, such as standard approximation and unit circle approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Item 3 and Item 4 have an error matrix E that is linear in A, Din, and Dout, which enables short proofs of properties such as summability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In addition, Item 3 and Item 4 characterize SV approximation of A in terms of ε-approximation of the “asymmetric lift” of A with respect to its symmetrization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This enables us to leverage results developed for ε-approximation, such as preservation under Schur complements (Theorem A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We next note the relation between SV and normalized SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n, and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices such that ker(Din) ⊆ lker(A), and ker(Dout) ⊆ rker(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let N = D+/2 in AD+/2 out and ˜N = D+/2 in �AD+/2 out Then �A Din,Dout ≈ε A if and only if ˜N svn ≈ ε N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 16 Note that if A is nonnegative and Din = diag(A⃗1n), Dout = diag(⃗1⊤ mA), then N and ˜N are the normalized adjacency matrices of A and �A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, when A, �A are the adjacency matrices of regular digraphs, then N and ˜N are the random-walk matrices of A and �A respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If instead A and �A are the adjacency matrices of Eulerian digraphs, we obtain that N and ˜N are similar to the random-walk matrices of A and �A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Normalized SV approximation is implied by standard ǫ/2-approximation with respect to the original degree matrix holding for all unitary multiples of the adjacency matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9 (Unitary transformation characterization of SV-approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ˜N, N ∈ Cn×n, we have that ˜N svn ≈ ε N if for every pair of unitary matrices U, V, U ˜NV is a standard ε/2- approximation of UNV with respect to degree matrix I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, if ˜N svn ≈ ε N then for every pair of matrices U, V satisfying ∥U∥ ≤ 1, ∥V∥ ≤ 1, we have that U ˜NV svn ≈ ε UNV, and hence U ˜NV is a standard ε-approximation of UNV with respect to degree matrix I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An important corollary is that SV approximation between adjacency matrices is preserved under multiplication on each side of the adjacency matrix by (possibly different) permutations: Corollary 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10 (SV preservation under multiplication by permutation matrices).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n ≥0 and suppose �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let U, V be arbitrary permutation matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then U �AV D′ in,D′ out ≈ε UAV where D′ in = UDinU∗ and D′ out = V∗DoutV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if �A sv≈ε A, then U�AV sv≈ε UAV and if �A svn ≈ ε A then U �AV svn ≈ ε UAV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An equivalent property was not known and is in fact not true for prior notions such as unit circle, as otherwise it would show SV and UC approximation are equivalent via Item 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Properties of Singular Value Approximation Here we provide several useful properties of SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, we show that SV approximation implies unit-circle approximation: Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If A, �A ∈ Cn×n with �A D≈ε A then �A ≈ε A with respect to degree matrix D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Interestingly, even for 2×2 symmetric matrices, SV approximation can be separated arbitrarily from UC approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every α, ε ∈ (0, 1), there are symmetric matrices W, � W with ∥W∥, ∥� W∥ ≤ 1 such that � W ≈ε W but � W is not an ε SV-approximation of W with respect to I − W for any ε′ < ε √ 1−α2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In prior work, ([AKM+20, Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1]), it was also shown that UC approximation can be arbitrary separated from standard approximation, even for undirected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='13 ([AKM+20]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every ǫ ∈ (0, 1), there exist undirected regular graphs with random walk matrices � W, W such that � W is an ε-approximation of W with respect to I − W but � W ≈c W does not hold for every c ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Further, since UC approximation trivially implies standard approximation (as was also argued in [AKM+20]) we see that SV approximation can be viewed of a strengthening of both UC and standard approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 17 A corollary of Corollary 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10 is that SV approximation is preserved under embedding in a block matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In particular, it shows the stronger fact that approximation is preserved even if we use a different block structure for rows and columns (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' we are not embedding into principal submatrices).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In contrast, unit circle approximation is only known to be preserved when the embedding pattern is a block pattern given by the directed cycle;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=', tensoring the adjacency matrix with that of the directed cycle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In fact, the separations we prove imply unit circle approximation cannot be preserved under all block embeddings, as otherwise it would imply SV approximation by Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='14 (SV preservation under arbitrary lifting).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n be matrices such that �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for all integers i, j, k, ℓ ≥ 0 \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j �A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb D′ in,D′ out ≈ε \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Where D′ in = \uf8ee \uf8f0 0i×i 0i×n 0i×ℓ 0n×i Din 0n×ℓ 0ℓ×i 0ℓ×n 0ℓ×ℓ \uf8f9 \uf8fb , D′ out = \uf8ee \uf8f0 0j×j 0j×m 0j×k 0m×j Din 0m×k 0k×j 0k×m 0k×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if �A sv≈ε A then \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j �A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb sv≈ε \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By taking sums of different such liftings, one can obtain approximation for arbitrary tensorings: if �A sv≈ε A, for any M ∈ {0, 1}i×j, we have �A⊗M sv≈ε A⊗M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A special case of this is concatenation: Corollary 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='15 (SV preservation under concatenation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A1, A2 �A1, �A2 ∈ Cm×n be matrices such that �A1 sv≈ε A1 and �A2 sv≈ε A2 then � �A1 �A2 � sv≈ε � A1 A2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Another corollary is that SV approximation is preserved under products of different walk ma- trices with essentially no loss in the approximation quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In contrast, the closest property known to be achieved by definitions of approximation considered in prior work is that unit circle approxi- mation is preserved with no loss under only powers of the same walk matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='16 (SV preservation under products).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let (Ni)i∈[ℓ], ( ˜Ni)i∈[ℓ] ∈ Cn×n be such that for every i, ˜Ni svn ≈ ε Ni.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then ˜Nℓ · · · ˜N2 ˜N1 svn ≈ ε+O(ǫ2) Nℓ · · · N2N1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A useful application of this result is to sparsifying powers of a walk matrix of an Eulerian digraph (Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SV approximation between digraphs requires exact preservation of the degrees: Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If �A sv≈ε A, then �A⃗1 = A⃗1 and �A⊤⃗1 = A⊤⃗1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SV sparsification is additive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We note that the analgous statement for normalized SV approxi- mation is that it is preserved under convex combinations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 18 Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If �Ai (Din)i,(Dout)i ≈ε Ai for all i ∈ [k], letting Din def = � i∈[k](Din)i and Dout def = � i∈[k](Dout)i, then � i∈[k] �Ai Din,Dout ≈ε � i∈[k] Ai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if �Ai sv≈ε Ai for every i, then � i∈[k] �Ai sv≈ε � i∈[k] Ai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SV approximation satisfies an approximate triangle inequality: Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If A3 Din,Dout ≈δ A2 and A2 Din,Dout ≈ε A1 then A3 Din,Dout ≈ε+δ+εδ A1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if A3 sv≈δ A2 and A2 sv≈εA1 then A3 sv≈ε+δ+εδA1, and if A3 svn ≈ δ A2 and A2 svn ≈ εA1 then A3 svn ≈ ε+δ+εδA1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, if for δ ∈ (0, 1/2) and A0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Aℓ we have Ai sv≈δ/2ℓ Ai−1 for every i, then Aℓ sv≈δ A0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, the equivalent claim holds for normalized SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A regular undirected graph SV approximates the complete graph with error equal to its expan- sion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G be a strongly connected, d-regular directed multigraph on n vertices with adjacency matrix A and let J ∈ Rn×n be a matrix with 1/n in every entry (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' J is the walk matrix of the complete graph with a self loop on every vertex).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then λ(G) ≤ 1 − λ/2 if and only if A/d svn ≈ λ J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4 Sparsification Given the Laplacian of an Eulerian directed graph, we wish to compute a sparsifier with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In this section, we show how to solve a more general problem of sparsi- fiying a non-negative rectangular matrix A ∈ Rm×n ≥0 with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Further, in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 we show how to construct square sparsifiers via the derandomized square, and in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 we show how to sparsify powers of walk matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first give an informal overview of our approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We give a series of reductions that reduce this problem to degree- and bipartition-preserving sparsification of expander graphs with respect to ε-approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 (Item 2), to SV-sparsify (possibly directed) A, it suffices to SV-sparsity the undirected bipartite lift �0 A⊤ A 0 � with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In particular, given a symmetric matrix satisfying slift � �A � sv≈ε slift (A), we have by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 that �A sv≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, any symmetric bipartition-preserving SV sparsifier of undirected graphs immediately gives an SV sparsifier for Eulerian digraphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In fact, every SV sparsifier of undirected graphs must be bipartition-preserving, and an asymmetric spar- sifier can be used to read off sparsifiers of the underlying directed graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a formal statement of this stronger claim, see Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We show (Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3) that for undirected bipartite expanders, degree- and bipartition-preserving approximation with respect to D implies SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' From this, we can construct SV sparsifiers (Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11) by computing an expander decom- position of the bipartite lift, then sparsifying the relevant subgraphs in a suitable way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' While we can use existing techniques to obtain this (with worse log factors), we also develop a new sparsifica- tion approach based on cycle decompositions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We show that in a bipartite expander, decomposing edges indicent to high-degree vertices into edge-disjoint cycles, then taking the odd or even edges 19 of each cycle with probability 1/2 produces a sparsifier with high probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This sparsification procedure automatically preserves degrees and bipartitions, which we require to lift approximation with respect to D to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Prior work [CGP+18] showed an equivalent result in general Eulerian digraphs, except they required the cycles to be short, resulting in almost-linear runtime [CGP+18,PY19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For convenience, we define notation for the adjacency and degree matrices of graphs: Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For the remainder of the section, for a weighted undirected graph G we let AG denote the adjacency matrix of G and DG def = AG⃗1 denote the degree matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' As such, an ε-SV sparsifier of G is a graph H such that AH sv≈ε AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, we show that for undirected expanders, degree- and bipartition-preserving sparsification with respect to the degree matrix implies sparsification with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 (Bipartiteness).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We say A ∈ Rn×n ≥0 is bipartite with bipartition S, T ⊆ [n] if S and T partition V and AS,S = 0S×S and AT,T = 0T×T .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given a graph G with adjacency matrix A, we say that a sparsifier H is bipartition-preserving if for every bipartition S, T in G, we have HS,S = 0S×S and HT,T = 0T×T (where the statement is vacuously satisfied if G is not bipartite).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the following lemma we show that for non-negative matrices that are bipartite with the same bipartition, SV approximation is implied by approximation with respect to the error matrix D (Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1), up to a loss in the condition number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 (From diagonal approximation to SV approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Suppose that symmetric �A ∈ Rn×n ≥0 is a degree and bipartition-preserving ǫ/2-approximation of symmetric bipartite A ∈ Rn×n ≥0 with respect to error matrix D def = diag(A⃗1) = diag(A⊤⃗1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then �A sv≈γ A for γ def = ε λ2 where λ def = 1 − λ2(D+/2AD+/2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Recall by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 that it suffices to show that �A is an ε/2-approximation of A with respect to E def = D − AD+A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, we have E ⪰ (D − A)D+(D + A) ⪰ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 and the assumption that �A is an ε/2-approximation of A with respect to D we have −ε/2 · D ⪯ A − �A ⪯ ε/2 · D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By assumption that A has expansion λ and is bipartite (and hence its spectrum is symmetric), we have (D − A)+ ⪯ 1 λD+, (D + A)+ ⪯ 1 λD+.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, as A is the adjacency matrix of an undirected graph the only singular values of A of magnitude 1 can be −1 and 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus since �A matches degrees exactly and preserves bipartitions we have that ker(A − �A) ⊇ ker(E).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus we have ���E+/2(A − �A)E+/2��� ≤ ���(D + A)+/2D1/2(D − A)+/2(A − �A)(D − A)+/2D1/2(D + A)+/2��� ≤ 1 λ ���D+/2D1/2(D − A)+/2(A − �A)(D − A)+/2D1/2D+/2��� ≤ 1 λ ���(D − A)+/2(A − �A)(D − A)+/2��� ≤ 1 λ2 ���D+/2(A − �A)D+/2��� ≤ ε λ2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 20 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Cycle and Expander Decompositions We give a new analysis of degree-preserving expander sparsification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Chu, Gao, Peng, Sachdeva, Sawlani, and Wang [CGP+18] constructed sparsifiers for Eulerian digraphs, which we now discuss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' They decomposed the low-importance edges of a graph into a series of edge-disjoint cycles C1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Ct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for each cycle, they chose a random orientation (clockwise or counterclockwise) and kept every edge in this orientation, and upweighted these by a factor of two.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' One can see that this procedure gives a subgraph and maintains Eulerianness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' They bounded the impact of a single such sample in terms of the length of the cycle, and by giving a sophisticated algorithm for decomposing a graph into many short cycles concluded an almost-linear time sparsifier for standard approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We follow this approach but make two modifications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, we require the cycles to be Forward- Backward (FB), where we always alternate between clockwise and counterclockwise edges (and all cycles are of even length).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This ensures that we exactly preserve degrees, not just Eulerianness, which is essential for SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We take advantage of the reduction from directed to undirected SV sparsification to perform this decomposition in the undirected bipartite lift, where undirected cycles correspond to alternating cycles in the original graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Our second modification is to prove a bound on the importance of the cycles that is independent of their length, as long as the cycles lie inside an expander.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This enables us to use a simple linear time algorithm for finding the decomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first recall the standard procedure that enables us to efficiently decompose an undirected graph into a union of (potentially long) FB cycles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given an unweighted undirected graph G = (V, E), the algorithm CycleDecomp re- turns a collection of edge-disjoint cycles C1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , CT ⊆ E in time O(|E| + |V |) such that at most n edges are not contained in some cycle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Algorithm 1: CycleDecomp (G = (V, E)), G undirected 1 Initialize Eex, C = {}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2 for v ∈ V do 3 while Greedily construct a non-backtracking path v = u0, u1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , uk in an arbitrary fashion and mark visited vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' do 4 if uk has no neighbors other than uk−1 then 5 Add Eex ← Eex ∪ {(uk−1, uk)} and remove (uk−1, uk) from E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 6 Continue search from uk−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 7 end 8 if uk = uk′ for some k′ < k − 1 then 9 Let C ← C ∪ (uk′, uk′+1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , uk−1) and let E ← E \\ {(uk′, uk′+1), (uk′+1, uk′+2), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , (uk−1, uk)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 10 Continue search from uk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 11 end 12 end 13 end 14 return C, Eex Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The runtime and the fact that the cycles are edge disjoint is direct from the description.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 21 Then note that for every edge (u, v) that we add to Eex (and remove from E) we must create an isolated vertex v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' But this can happen at most n times, so we are done.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, we recall the definition of an expander partition: Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given φ, γ and an undirected graph G, a (φ, γ)-expander partition of G is a partition V1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Vt such that at most γ · m edges cross between partitions and for every i ∈ [t], λ(G[Vi]) ≤ 1 − φ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We will use algorithms for expander decomposition as a key subroutine in our sparsifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We re- quire a small constant fraction of edges to lie between clusters, and thus state the algorithms fixing γ = 1/16 such that m/16 edges cross subgraphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In particular, given an algorithm ExpanderDe- comp computing a (Φ, m/16) expander decomposition, let T(ExpanderDecomp, m) be its run- time (on graphs with m edges) and Φ(ExpanderDecomp) = Φ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We recall the best known nearly- linear expander partitioning algorithm, as well as the (optimal) existential result: Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 ([ADK22]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There is an algorithm ExpanderDecomp(G) running in randomized time T(ExpanderDecomp, m) = O(m log7 m) that computes a (Φ = log−4 m, 1/16) expander decomposition of G with high probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 ([ST04]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There exists a (Φ = log−2 m, 1/16) expander decomposition of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Sparsifying Bipartite Expanders The core of our algorithm is a way to sparsify bipartite, unweighted expanders in a degree- and bipartition- preserving way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The algorithm decomposes the graph into a series of edge-disjoint cycles, then takes either twice the odd or twice the even edges with probability 1/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the case that the undirected graph corresponds to the bipartite lift of an Eulerian digraph G, this precisely corresponds to decomposing G into a union of Forward-Backward cycles and taking the forward or backward edges for each cycle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We bound the error in terms of the degree matrix D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This is tolerable as we apply this procedure inside expanders, so by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 we can transform this into an SV approximation guarantee with polylog(n) loss in approximation quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There is a constant c > 0 such that given δ ∈ (0, 1) and an unweighted, undirected bipartite graph G with m edges, SparsifyCycleUnweighted returns in time O(m) a graph H with edge weights in {1, 2} such that with high probability AH is a δ-approximation of AG with re- spect to error matrix DG, and H has at most max{cnδ−2 log n, (7/8)m} edges and exactly preserves degrees and bipartiteness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To prove correctness, we recall a result on matrix concentration of Tropp: Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9 ([Tro12]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let {Xi} be d × d random PSD matrices such that E[� i Xi] ⪯ X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For δ ≤ 1, Pr �� i Xi − E �� i Xi � ⪰ δX � ≤ 1−d·eδ2/3R, Pr �� i Xi − E �� i Xi � ⪯ −δX � ≤ 1−d·eδ2/3R, provided that R ≥ ��X+/2XiX+/2�� almost surely.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Using this matrix concentration result, we can prove the theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 22 Algorithm 2: SparsifyCycleUnweighted (δ, G = (V, E)) 1 if δ < � cn log n/m then 2 return G 3 end 4 Let dlow def = m 2n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 Let Elow be the edges adjacent to vertices v with d(v) ≤ dlow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 6 Let Eex, (C1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Ct) =CycleDecomp (V, E \\ Elow).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 7 Initialize H = (V, Eex ∪ E \\ Elow).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 8 for i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , t do 9 Let (a1, b1), (b1, a2), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , (bk, a1) = Ci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 10 Choose b ← 0, 1 uniformly at random.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 11 if b = 0 then 12 H ← H ∪ {2 · (ai, bi)}i∈[k] // Add with weight 2 13 end 14 else H ← H ∪ {2 · (bi, ai+1)}i∈[k] ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 15 end 16 return H Proof of Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, note that if we do not sparsify G at all we have cnδ−2 log n < m and hence the sparsity requirement is already satisfied.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Subsequently we assume this is not the case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Recall Elow is the set of edges adjacent to a vertex with degree at most dlow = (m/2n), and observe that there are at most dlow · n ≤ m/2 such edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Further recall C1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Ct are the cycles found by CycleDecomp applied to E \\ Elow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By assumption that G is bipartite, all such cycles are of even length.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let Xi def = �Di − �Ai be the random variable of the Laplacian of the ith sampled cycle and observe 0 ⪯ Xi ⪯ 2ICi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, let X def = DG ⪰ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We have E �� i Xi � = ⊤ � i=1 LCi ⪯ X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, for every i ∈ [t] we have ���X+/2XiX+/2��� ≤ 2 ���D+/2ICiD+/2��� ≤ 4n m where the final step follows from all cycles being exclusively incident to vertices with degree at least m/2n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, we apply Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9 with {Xi}i∈[t], R = 4n/m, X = X and δ = δ/c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Letting DC − AC be the random graph obtained by the cycle sampling procedure, with high probability we have −δDG ⪯ (DC − AC) − t � i=1 (DCi − ACi) ⪯ δDG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By adding the non-random components to we obtain −δDG ⪯ AH − AG ⪯ δDG and thus AH is an δ-approximation of AG with respect to DG with high probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 at least m/2−n ≥ m/4 edges are included in the cycles C1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Ct, and hence H has at most 7m/8 edges under the assumption that m ≥ cn, and H preserves degrees and bipartiteness by construction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 23 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Sparsifying a Constant Fraction of Edges We then construct our routine for sparsifying a graph with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' At each iteration, we decrease the number of edges by a constant factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first bucket edges into powers of two, then apply SparsifyCycleUnweighted .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ε > 0 and b ∈ N and an undirected bipartite graph G with m edges, each with weight in {1, 2, 22, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 2b}, SparsifyCycle returns a graph H with O � b · nε−2Φ−4 log n � + 15 16m edges where Φ = Φ(ExpanderDecomp), each with weight in {1, 2, 22, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 2b+1}, such that with high probability AH sv≈ε AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, SparsifyCycle runs in randomized time O(m + bn + T(ExpanderDecomp, m))).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Algorithm 3: SparsifyCycle (ε, b, G = (V, E)) 1 Let δ = ε−1Φ2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2 Initialize H = (V, ∅).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 for i = 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , b do 4 Initialize (unweighted) Gi as G restricted to edges with weight 2i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 Let Vi,1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Vi,t, Eex def =ExpanderDecomp(Gi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 6 Let H ← H ∪ 2i · Eex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 7 for j ∈ 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , t do 8 Let Hi,j def =SparsifyCycleUnweighted (δ, Gi[Vi,j]) 9 Let H ← H ∪ 2i · Hi,j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 10 end 11 end 12 return H Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Gb be the unweighted bipartite graphs where Gi contains an edge (u, v) if there is an equivalent edge with weight 2i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, observe that b � i=0 2i · (DGi − AGi) = DG − AG and it suffices to sparsify unweighted bipartite graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore let mi def = |E(Gi)| and observe � i mi = m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next, for every i observe that ExpanderDecomp computes a decomposition Vi,1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Vi,t such that letting Gi,j def = Gi[Vi,j] we have λ(Gi,j) ≤ 1 − Φ, and at most mi/16 edges do not lie in one of these subgraphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let ni,j def = |Gi,j| and observe � j ni,j ≤ n and let mi,j def = |E(Gi,j)| and observe � j mi,j ≤ mi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8, we have that for every i, j, AHi,j is a degree and bipartition-preserving δ-approximation of AGi,j with respect to DGi,j with high probability, and thus by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 AHi,j sv≈δ·Φ−2 AGi,j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 24 Thus by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='18 we have AH sv≈δ·Φ−2 AG and so by our choice of δ = εΦ2 we obtain the desired approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We now analyze the number of edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The number of edges not placed into a subgraph Gi,j is at most b � i=1 mi 16 = m 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8 the number of edges in the sparsified graphs is at most b � i=1 t � j=1 max{cni,jδ−2 log n, 7 8mi,j} ≤ b � i=1 t � j=1 � cni,jδ−2 log n + 7 8mi,j � = O � bnε−2Φ−4 log n � + 7 8 b � i=1 t � j=1 mi,j = O � bnε−2Φ−4 log n � + 7 8m and hence the total number of edges is bounded as desired.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Finally, we analyze the time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We can compute every Gi with a single pass through the edge list, so the total work is bounded by � i � j O(mi,j + ni,j) + �b i=1 T(ExpanderDecomp, mi) = O(m + bn + T(ExpanderDecomp, m)) as claimed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 Nearly Linear-Sized Sparsifers We now apply our routine for sparsifying a constant fraction of edges in the natural recursive way.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ε ∈ (0, 1) and an undirected bipartite graph G with m edges with integer edge weights in [1, U], SparsifyGraph returns a graph H with O � log(nU) · nε−2Φ−4 log3 n � edges where Φ = Φ(ExpanderDecomp), such that with high probability AH sv≈ε AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, SparsifyGraph runs in time O(m + nb + T(ExpanderDecomp, m log(U))).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First observe that in Phase 1 of the algorithm, we obtain H0 such that H0 has at most mb edges (recalling b = ⌈log(U)⌉) and each edge has weight {1, 2, 22, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 2b}, and AH0 = AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then in each phase we obtain Hj such that AHj sv≈ε/2 log n AHj−1 with high probability by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then applying Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19 we obtain AHt sv≈ε AG with high probability as desired.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We next analyze the sparsity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Observe that the final edge weights are bounded by 2b+t = poly(nU) and hence the final sparsity is O � (b + t)nε−2Φ−4 log3 n � = O � log(nU)nε−2Φ−4 log3 n � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 25 Algorithm 4: SparsifyGraph (ε, G = (V, E)) 1 Initialize H0 = (V, ∅).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2 Let b def = ⌈log(U)⌉.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 for i = 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , b do 4 for (u, v) ∈ E do 5 Let ⟨wG(u, v)⟩ be the binary expansion of wG(u, v).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 6 if ⟨wG(u, v)⟩i = 1 then Add (u, v) to H0 with weight 2i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 7 end 8 end 9 for j = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , t def = O(log n) do 10 Let Hj ←SparsifyCycle (ε/2 log n, b + j, Hj−1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 11 end 12 return Ht Finally, we analyze the time complexity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Our initial call has mb edges and thus runs in time O(mb + bn + T(ExpanderDecomp, mb)) by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10, and all subsequent calls are to sparsify graphs with a constant factor fewer edges, so the time is dominated by the initial term.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By plugging in the algorithmic (Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6) and existential (Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6) results for expander partitions, we obtain our sparsifier routines: Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ε ∈ (0, 1) and an undirected bipartite graph G with m edges with integer edge weights in [1, U], there is a randomized algorithm that returns a graph H with O � log(nU) · nε−2 log19 n � edges such that with high probability AH sv≈εAG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, the algorithm runs in time O(log(U)m log7 m).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ε ∈ (0, 1) and an undirected bipartite graph G with m edges with integer edge weights in [1, U], there is a graph H with O � log(nU) · nε−2 log11 n � edges such that AH sv≈ε AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 Derandomized Square Sparsification Here we show that the derandomized square of Rozenman and Vadhan [RV05] gives an SV sparsi- fication of the square graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first explain the derandomized square.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given a d-regular graph G, the squaring operation can be though of as placing a d × d complete graph between the in- and out-neighbors of every vertex v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The derandomized square replaces this complete graph with an expander.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' As an expander approximates the complete graph, this sparsification operation produces an approximate square.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first formally define the derandomized square.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given a d-regular graph G = ([n], E) with neighbor function ΓG : [n] × [d] → [n] and c-regular graph H = ([d], E′) with neighbor function ΓH : [d] × [c] → [d], the derandomized square of G with respect to H, denoted G s○H, is the dc-regular graph on n vertices with neighbor function ΓG s○H(v, (i, j)) = ΓG(ΓG(v, i), ΓH(i, j)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 26 Prior analyses of the derandomized square have shown that it improves expansion approximately as well as the true square [RV05] and that it produces unit-circle square sparsifiers [AKM+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We strengthen the latter conclusion to SV square sparsifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G = (V, E) be a d-regular directed multigraph with random walk matrix W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let H be a c-regular expander on d vertices with λ(H) ≤ 1 − ǫ and let � W be the adjacency matrix of G s○H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then � W svn ≈ ε W2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let WH be the random walk matrix corresponding to H, and J ∈ Rd×d be the random walk matrix of a complete graph with self loops.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemmas 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='14 and 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20 we have In ⊗ WH sv≈ε In ⊗ J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (18) Let F be an nd×nd edge-rotation permutation matrix such that F((v, j), (u, i)) = 1 if the ith edge leaving u is the jth edge entering v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It is straight-forward to verify that W2 = Q⊤F(I ⊗ J)FQ and � W = Q⊤F(I ⊗ WH)FQ where Q def = I ⊗ 1d √ d is an nd × n matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Now note that ∥F∥ ≤ 1 and ∥Q∥ ≤ 1 and thus ∥FQ∥, ∥Q⊤F∥ ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus by Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9, and Equation (18) we have � W sv≈ǫ W2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 Sparsification of Random Walk Polynomials We give the first nearly-linear time algorithm for sparsifying random walk polynomials of Eulerian digraphs with second singular value bounded away from 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In particular, given an Eulerian digraph W ∈ Rn×n ≥0 with second singular value bounded away from 1, we compute a sparse graph such that the walk matrix � W approximates Wℓ with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For this result, we crucially require that SV approximation is preserved under products, a feature that is not obtained by prior notions of approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given a strongly connected Eulerian digraph G, we let σ(G) = σ2(D−1/2 G AGD−1/2) and refer to this as the second normalized singular value of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ℓ ∈ N and ε ∈ (1/poly(n), 1) and an Eulerian digraph G where σ(G) ≤ 1 − 1/poly(n) with m edges with integer edge weights in [1, poly(n)], SparsifyPower returns in time ˜O(m + nε−2 log3(ℓ)) an Eulerian digraph H with ˜O(nε−2 log2(ℓ)) edges with integer edge weights in [1, poly(n, ℓ, 1/ε)] such that D+/2 H AHD+/2 H svn ≈ ε (D+/2 G AD+/2 G )ℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This implies the following corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A be the adjacency matrix of an Eulerian digraph with diagonal matrix of degrees D and walk matrix W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The kth random walk power of A is defined as D1/2(D−1/2AD−1/2)kD1/2 = WkD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 27 Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G be an Eulerian digraph with poly(n) weak mixing time, adjacency matrix A, diagonal degree matrix D, and walk matrix W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for any positive integer k ≤ poly(n) and γ ≥ 1/poly(n), we can, in nearly linear time, output an ˜O(n/ε2) sized ε-SV-sparsifer of kth random walk power of A′ = (1 − γ)A + γD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, the walk matrix of the true kth random walk power of A′ is given by ((1 − γ)W + γI)k so it is the k step γ-lazy random walk on G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Immediate from Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 and Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To prove these, we first recall the square sparsification algorithm implicit in the work of Cohen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CKP+17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20 (Implicit in [CKP+17]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given an Eulerian weighted digraph G with Laplacian D − AG, let G2 be the graph with Laplacian D − AGD+AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let U denote the Laplacian of the undirected graph corresponding to (AG2 + AT G2)/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then SparsifySquare of [CKP+17] produces in time ˜O(m), a weighted graph H with the following properties: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' H has the same weighted in- and out- degree sequences as those of G2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' H has ˜O(m/ε2) nonzero edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' AH is an ǫ/2-approximation of AGD−1AG with respect to SD−AGD+AG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We can use a block embedding and Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20 to obtain a nearly-linear time algorithm for sparsifying products with respect to SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ε > 0 and Eulerian digraphs G1, G2 each with at most m edges with matching degree matrix D, where each edge has integer weight in [1, U], SparsifyProduct returns an Eulerian digraph H with ˜O(m/ε2) edges such that with high probability AH sv≈ε AG2D+AG1 Moreover, SparsifyProduct runs in time ˜O(m/ε2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Algorithm 5: SparsifyProduct(ε, G1, G2) 1 Let G be the graph with AG = \uf8ee \uf8f0 0 0 0 AG1 0 0 0 AG2 0 \uf8f9 \uf8fb 2 Let H ← [SparsifySquare(ε/3, G)]3,1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 return H Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first analyze correctness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note by Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20 we have that \uf8ee \uf8f0 0 0 0 0 0 0 AH0 0 0 \uf8f9 \uf8fb is an ε/6-approximation of \uf8ee \uf8f0 0 0 0 0 0 0 AG2D+AG1 0 0 \uf8f9 \uf8fb with respect to R where R = \uf8ee \uf8f0 D 0 −(AG2D+AG1)⊤ 0 0 0 −AG2D+AG1 0 D \uf8f9 \uf8fb This is precisely Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7, so we conclude AH sv≈ε AG2D+AG1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 28 Finally, we require an algorithm that removes very small edge weights.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This is not trivial, as we must do so in a way that exactly preserves degrees for SV approximation to hold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' There is an algorithm FixEdgeWeights such that the following holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G be an Eulerian digraph with m edges where the degree matrix DG can be represented as a subset sum of {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 2b}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for t ∈ N, FixEdgeWeights(t, G) outputs in time O(mb) a graph H such that: Ht has m + 2n edges and is Eulerian and DH = DG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Edge weights in H can be represented as a subset sum of {2−t, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 2b}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ∥AH − AG∥ ≤ 2n · 2−t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, let ⌊G⌋ be the graph where all edge weights are rounded down to a multiple of 2−t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Observe that this satisfies ∥A⌊G⌋ − AG∥ ≤ n · 2−t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, note that each vertex v has an in-degree shortfall equal to sin(v) = Dv,v−� u(A⌊G⌋)v,u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that sin(v) ≥ 0 since we only remove weight from edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This can be expressed as a sub- set sum of {2−t, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , 2b} as all numbers in the summand can be.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An equivalent statement holds for the out-degree shortfall.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, the total in- and out- degree shortfalls must be equal, since every truncation contributes to both sums.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, by greedily matching in- shortfalls to out- shortfalls, we obtain the desired graph H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It is easy to see this satisfies the approximation claim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We are now prepared to state our algorithm for sparsifying an arbitrary power of an Eulerian digraph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Algorithm 6: SparsifyPower(ε, G, ℓ) 1 Let t def = ⌈log ℓ⌉.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2 Let ⟨ℓ⟩ be the binary expansion of ℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 Let l def = O(log(nℓ/ε)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4 Let G1 def = SparsifyGraph (ε/2, G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 for i = 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , t do 6 Let Pi def = SparsifyProduct(ε/t, Gi−1, Gi−1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 7 Let Hi def = FixEdgeWeights(l, Pi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 8 Let Gi def = SparsifyGraph (ε/t, Hi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 9 if ⟨ℓ⟩t−i = 1 then 10 Let Pi ← SparsifyProduct(ε/t, Gi−1, G1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 11 Let Hi ← FixEdgeWeights(l, Pi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 12 Let Gi ← SparsifyGraph (ε/t, Hi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 13 end 14 end 15 return Gt We are now prepared to prove the result: 29 Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given ℓ ∈ N and ε ∈ (1/poly(n), 1) and an Eulerian digraph G where σ(G) ≤ 1 − 1/poly(n) with m edges with integer edge weights in [1, poly(n)], SparsifyPower returns in time ˜O(m + nε−2 log3(ℓ)) an Eulerian digraph H with ˜O(nε−2 log2(ℓ)) edges with integer edge weights in [1, poly(n, ℓ, 1/ε)] such that D+/2 H AHD+/2 H svn ≈ ε (D+/2 G AD+/2 G )ℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first show the approximation guarantee.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that by the guarantees of Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='21 and Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12 and Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='22, for every i, D def = DPi = DHi = DGi = DG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let NPi def = D+/2A+/2 Pi D+/2 and NHi def = D+/2A+/2 Hi D+/2 and NGi def = D+/2A+/2 Gi D+/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The proof of correctness does not change if we assume ℓ = 2t for some t (and hence the if condition in the main loop never occurs), so we analyze it assuming this for simplicity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The fact that this does not matter follows from the fact that SV approximation is preserved under arbitrary products, not merely powers (Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first argue that the singular value of all graphs we sparsify remain bounded below 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Claim 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every i, σ(Hi), σ(Pi), σ(Gi) ≤ 1 − 1/poly(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We show the claim via induction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By assumption, we have σ(G) = 1 − 1/nc for some constant c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 we have σ(G1) ≤ 1−(1+ε/t)/nc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Subsequently, assume that σ(Gi−1) ≤ 1 − (1 + 2ε/t)2(i−1)/nc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 we have σ(G2 i−1) ≤ σ(Gi−1) ≤ 1 − (1 + 2ε/t)2(i−1)/nc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 we have σ(Pi) ≤ 1 − (1 + ε/t)(1 + 2ε/t)2(i−1)/nc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 we have σ(Hi) ≤ 1 − (1 + ε/t)(1 + 2ε/t)2(i−1)/nc + 1/n10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 we have σ(Gi) ≤ 1 − (1 + 2ε/t)2i/nc So the claim follows assuming ε ≤ 1/c′ for constant c′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We now show that SV approximation is preserved at every step of the loop.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Fixing i and Gi−1: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='21 we have that NPi svn ≈ ε/3t N2 Gi−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='22 we have ∥AHi −APi∥ ≤ 2n2−l, so AHi is a 2n2−l ≤ ε/poly(tn) approximation of APi with respect to D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Taking the bipartite lift, � 0 A⊤ Hi AHi 0 � is a degree- and bipartition- preserving ε/poly(tn)-approximation of � 0 A⊤ Pi APi 0 � with respect to �D 0 0 D � (where we use that Pi is not itself bipartite to ensure that the lift of Hi preserves all bipartitions).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 (using that the bipartite lift has normalized second eigenvalue at least 1/poly(n) by Claim 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='23) and Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7, we have that NHi sv≈ε/3t NPi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12, NGi svn ≈ ε/3t NHi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 30 Applying Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19 we obtain NGi svn ≈ ε/t N2 Gi−1 for every i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Therefore by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='16, NGt svn ≈ ε/t � NGi−1 �2 svn ≈ ε/t .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (NG2)2t−1 svn ≈ ε/t (NG1)2t and so we conclude by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19 (and taking ε ← ε/10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To analyze the time, note that by Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='22 we always call SparsifyGraph on graphs with (up to rescaling) integer edge weights in [1, poly(nℓ/ε)], and by Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20 Pi (and thus Hi) has ˜O(nε−2 log2 ℓ) edges, so we obtain the desired runtime and sparsity by Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 Squaring algorithm for solving normal directed Laplacians In [PS14], Peng and Speilman proposed a squaring-based parallel algorithm for computing an approximate pseudo-inverse (preconditioner) of a Laplacian matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A key ingredient of their algorithm is the following recursion, which we refer to as the Peng-Spielman squaring recursion (PS-recursion), for inverting I − W when ∥W∥ < 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 (I − W)−1 = 1 2 � I + (I + W)(I − W2)−1(I + W) � (19) Leveraging the PS-recursion a natural appproach to find an approximate inverse of I − W, is to compute a sparse matrix � W such that I − � W ≈ I − W2, and then compute the inverse of I− � W recursively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [PS14] showed that for symmetric Laplacians, using spectral approximation this recursion can be leveraged to produce a constant preconditioner with logarithmic depth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Here we prove that the same algorithm gives approximate pre-conditioner for I − W if W is normal but not necessarily symmetric and the square approximations are normalized SV approx- imations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We think Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 gives hope that a similar approach can expand the proof to non-normal matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If proved, this will result in simple but space and time efficient algorithms for solving directed Laplacian systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For example, this could simplify the space efficient solver introduced in [AKM+20] which in turn is used to estimate random walk probabilities to a high precision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Below we give the main theorem which formally defines the PS-recursion and states the main result of this section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This theorem uses a generalization of the spectral norm to a (semi-)norm induced by a PSD matrix, which we first define.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Definition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given PSD F ∈ Cn×n, we let ∥ · ∥F be the (semi)norm on vectors where ∥x∥F def = √ x∗Fx, and define the matrix norm ∥A∥F def = supx∈Cn\\{⃗0} ∥Ax∥F ∥x∥F .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a normal matrix W ∈ Cn×n with ∥W∥ ≤ 1, let W = W0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Wk−1 be a sequence of matrices such that for all i ∈ [k − 1] we have Wi svn ≈ ǫ W2 i−1 for ε ≤ 1/(4k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Define Pi def = 1 2 [I + (I + Wi)Pi+1(I + Wi)] for all 0 ≤ i < k where Pk is a matrix such that ���(I − W2k) 1 2 � Pk − (I − W2k)+� (I − W2k) 1 2 ��� ≤ O(kǫ) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for B def = ((I − W)1/2)∗(I − W)1/2 ∥P0(I − W) − I∥B = ���(I − W) 1 2 � P0 − (I − W)+� (I − W) 1 2 ��� ≤ O(k2ǫ) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (20) 6Technically the recursion was introduced for symmetric W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 31 Note that in the above statement, the square root of I − W is well defined by the assumption that W is normal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the case of non-normal matrices, which is outside the scope of this paper, one can consider the Maclaurin series for (1 − z)1/2 to work with (I − W)1/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We do not give a full algorithm for solving systems defined by Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 as the main purpose is to show that the error analysis of the squaring algorithm can be expanded to normal asymmetric matrices, and we already know space and time efficient algorithms [CKK+18,AKM+20] for solving a larger class of matrices which includes non-normal W’s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' However, one can still get an actual squaring based solver from Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For example, for a normal W corresponding to a random walk matrix of an Eulerian digraph one can carry the same steps in [AKM+20] by setting k = O(log n) and Pk = I − ⃗1⃗1⊤ n to get an actual squaring based solver using Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 and the derandomized square sparsification described in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We prove the statement of the theorem, following a few helper lemmas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' First, we show the error of Wi’s approximating W2i’s grows additively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let Wi be defined as in Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then Wi svn ≈ 2iǫ W2i for all for i ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Recall that Wj svn ≈ ε W2 j−1 for every j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='16 we have for every j ≤ i, (Wj)2i−j svn ≈ ε+O(ε2) (Wj−1)2i−j+1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The result then follows from Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next we give a general technical tool regarding normal matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For a normal matrix V with ∥V∥ ≤ 1 ∥(I + V) 1 2 ∥ ≤ √ 2 (21) and if �V svn ≈ δ V then ���(I − V) 1 2( �V − V)((I − V2)+) 1 2 ��� ≤ O(δ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (22) Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that since V is normal, we can write it as V = UDU∗ where U is unitary and D is diagonal and ∥D∥ ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then it is easy to see ∥(I + V) 1 2 ∥ = ∥U(I + D) 1 2 U∗∥ = ∥(I + D) 1 2 ∥ ≤ √ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (23) To prove (22), given the SV approximation definition, for unit vectors x, y we have, ���x∗(I − V) 1 2( �V − V)((I − V2)+) 1 2 y ��� ≤ δ 2 · x∗(I − V) 1 2(I − VV∗)(I − V∗) 1 2 x + δ 2 · y∗((I − (V∗)2)+) 1 2(I − V∗V)((I − V2)+) 1 2 y We bound each term on the right hand side separately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Since V is normal we can prove bounds on the norm of (I − V) 1 2 (I − VV∗)(I − V∗) 1 2 and ((I − (V∗)2)+) 1 2 (I − V∗V)((I − V2)+) 1 2 by looking at complex numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that for z ∈ C and |z| ≤ 1, |(1 − z) 1 2(1 − z · z∗)2(1 − z∗) 1 2| = |1 − z| · (1 − |z|2) ≤ 2, thus ∥(I − V) 1 2 (I − VV∗)(I − V∗) 1 2∥ ≤ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next, ���(1 − z∗2)− 1 2(1 − z · z∗)2(1 − z2)− 1 2 ��� = 1 − |z|2 |1 − z2| ≤ 1 32 because 1 − |z|2 ≤ |1 − z2| by triangle inequality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, ∥((I − (V∗)2)+) 1 2(I − V∗V)((I − V2)+) 1 2 ∥ ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Completing the proof that, ���x∗(I − V) 1 2 ( �V − V)((I − V2)+) 1 2y ��� ≤ 3 2 · δ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Leveraging these tools we give the central lemma towards proving Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 and conclude by proving Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ǫ and Pi’s defined as in Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2, ���(I − W2i) 1 2 � Pi − (I − W2i)+� (I − W2i) 1 2 ��� ≤ ǫi (24) where ǫi = (1 + O(i · ǫ))ǫi+1 + O(i · ǫ) for 0 ≤ i < k and ǫk = O(k · ǫ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We prove the statement by backward induction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that the statement is true for i = k by the assumption about Pk in Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let Qi = 1 2 � I + (I + Wi)(I − W2i+1)+(I + Wi) � , Ri = 1 2 � I + (I + Wi)(I − W2i+1)+(I + W2i) � , Si = 1 2 � I + (I + W2i)(I − W2i+1)+(I + W2i) � = (I − W2i)+.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then by triangle inequality we have, ∥(I − W2i) 1 2 � Pi − (I − W2i)+� (I − W2i) 1 2 ∥ ≤ ∥(I − W2i) 1 2 (Pi − Qi)(I − W2i) 1 2∥ (25) + ∥(I − W2i) 1 2 (Qi − Ri)(I − W2i) 1 2∥ (26) + ∥(I − W2i) 1 2 (Ri − Si)(I − W2i) 1 2 ∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (27) Next, we bound each term separately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For (25), we have ���(I − W2i) 1 2(Pi − Qi)(I − W2i) 1 2 ��� = 1 2 ���(I − W2i) 1 2(I + Wi) � Pi+1 − (I − W2i+1)+� (I + Wi)(I − W2i) 1 2 ��� ≤ 1 2 ���(I − W2i) 1 2(I + Wi)((I − W2i+1) 1 2 )+��� ���(I − W2i+1) 1 2 � Pi+1 − (I − W2i+1)+� (I − W2i+1) 1 2 ��� ���((I − W2i+1) 1 2 )+(I + Wi)(I − W2i) 1 2 ��� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that the middle term above satisfies ���(I − W2i+1) 1 2 � Pi+1 − (I − W2i+1)+� (I − W2i+1) 1 2 ��� ≤ ǫi+1 33 by the induction hypothesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next, we bound the first and third terms similarly: ����� (I − W2i) 1 2 (I + Wi)((I − W2i+1) 1 2)+ √ 2 ����� ≤ ����� (I − W2i) 1 2 (I + W2i)((I − W2i+1) 1 2 )+ √ 2 ����� + (28) ����� (I − W2i) 1 2 (Wi − W2i)((I − W2i+1) 1 2)+ √ 2 ����� ≤ 1 + O(i · ǫ) as ���� (I−W2i) 1 2 (I+W2i)((I−W2i+1 ) 1 2 )+ √ 2 ���� ≤ 1 by Equation (21) in Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 and ���� (I−W2i) 1 2 (Wi−W2i)((I−W2i+1 ) 1 2 )+ √ 2 ���� ≤ O(i · ǫ) by Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 and Equation (22) in Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Combining all the above, ���(I − W2i) 1 2(Pi − Qi)(I − W2i) 1 2 ��� ≤ (1 + O(i · ǫ))2ǫi+1, giving the bound for (25).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Next, for (26), we have ���(I − W2i) 1 2 (Qi − Ri)(I − W2i) 1 2 ��� = 1 2 ���(I − W2i) 1 2(I + Wi)(I − W2i+1)+(Wi − W2i)(I − W2i) 1 2 ��� ≤ ����� (I − W2i) 1 2 (I + Wi)((I − W2i+1) 1 2 )+ √ 2 ����� · ����� ((I − W2i+1) 1 2 )+(Wi − W2i)(I − W2i) 1 2 √ 2 ����� We can get bounds for the two terms above similar to (28).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus we get, ���(I − W2i) 1 2 (Qi − Ri)(I − W2i) 1 2 ��� ≤ ����� (I − W2i) 1 2 (I + Wi)((I − W2i+1) 1 2)+ √ 2 ����� · ����� ((I − W2i+1) 1 2 )+(Wi − W2i)(I − W2i) 1 2 √ 2 ����� ≤ (1 + O(i · ǫ)) · O(i · ǫ) ≤ O(i · ǫ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Finally, for (27), we have ∥(I − W2i) 1 2 (Ri − Si)(I − W2i) 1 2 ∥ = 1 2 ���(I − W2i) 1 2 (Wi − W2i)(I − W2i+1)+(I + W2i)(I − W2i) 1 2 ��� ≤ ����� (I − W2i) 1 2(Wi − W2i)((I − W2i+1) 1 2)+ √ 2 ����� · ����� ((I − W2i+1) 1 2 )+(I + W2i)(I − W2i) 1 2 √ 2 ����� Further, similar to the earlier upper bounds, we have ���� (I−W2i) 1 2 (Wi−W2i)((I−W2i+1 ) 1 2 )+ √ 2 ���� ≤ O(i · ǫ) and ���� ((I−W2i+1) 1 2 )+(I+W2i)(I−W2i ) 1 2 √ 2 ���� ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus ∥(I − W2i) 1 2(Ri − Si)(I − W2i) 1 2∥ ≤ O(i · ǫ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 34 Proof of Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ǫ < 1 k2 , we can solve for ǫ0 from ǫi = (1 + O(i · ǫ))ǫi+1 + O(i · ǫ) in Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5 and get ǫ0 = O(ǫ · k2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To complete the proof, we show ∥P0(I − W) − I∥B = ∥(I − W) 1 2 � P0 − (I − W)+� (I − W) 1 2∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that, ∥(I − W) 1 2 � P0 − (I − W)+� (I − W) 1 2∥ = ∥(I − W)1/2 [P0(I − W) − I] (I − W)+/2∥ = ���� � (I − W)∗/2(I − W)1/2�1/2 [P0(I − W) − I] � (I − W)+/2((I − W)∗)+/2�1/2���� = ���� � (I − W)∗/2(I − W)1/2�1/2 [P0(I − W) − I] � (I − W)∗/2(I − W)1/2�+/2���� = ∥P0(I − W) − I∥B where we used Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Acknowledgements We thank Jack Murtagh for his helpful collaboration at the early stage of research on this research and we thank Jonathan Kelner for helpful conversations at various stages of work on this project.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Aaron Sidford was supported by a Microsoft Research Faculty Fellowship, NSF CAREER Award CCF-1844855, NSF Grant CCF-1955039, a PayPal research award, and a Sloan Research Fellow- ship.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' References [ABKS21] Deeksha Adil, Brian Bullins, Rasmus Kyng, and Sushant Sachdeva.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Almost-linear- time weighted lp-norm solvers in slightly dense graphs via sparsification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Nikhil Bansal, Emanuela Merelli, and James Worrell, editors, 48th International Colloquium on Automata, Languages, and Programming, ICALP 2021, July 12-16, 2021, Glasgow, Scotland (Virtual Conference), volume 198 of LIPIcs, pages 9:1–9:15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Schloss Dagstuhl Leibniz-Zentrum f¨ur Informatik, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [ACL06] Reid Andersen, Fan R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Chung, and Kevin J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Local graph partitioning using pagerank vectors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In 47th Annual IEEE Symposium on Foundations of Computer Science (FOCS 2006), 21-24 October 2006, Berkeley, California, USA, Proceedings, pages 475–486.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [ADK22] Daniel Agassy, Dani Dorfman, and Haim Kaplan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Expander decomposition with fewer inter-cluster edges using a spectral cut player.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' CoRR, abs/2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10301, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [AG15] Nima Anari and Shayan Oveis Gharan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Effective-resistance-reducing flows, spectrally thin trees, and asymmetric TSP.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Venkatesan Guruswami, editor, IEEE 56th Annual Symposium on Foundations of Computer Science, FOCS 2015, Berkeley, CA, USA, 17-20 October, 2015, pages 20–39.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 35 [AGM+10] Arash Asadpour, Michel X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Goemans, Aleksander Madry, Shayan Oveis Gharan, and Amin Saberi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An o(log n/ log log n)-approximation algorithm for the asymmetric traveling salesman problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Moses Charikar, editor, Proceedings of the Twenty- First Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2010, Austin, Texas, USA, January 17-19, 2010, pages 379–389.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [Ahm20] AmirMahdi Ahmadinejad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Computing Stationary Distributions: Perron Vectors, Ran- dom Walks, and Ride-Sharing Competition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Stanford University, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [AJSS19] AmirMahdi Ahmadinejad, Arun Jambulapati, Amin Saberi, and Aaron Sidford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Perron-frobenius theory in nearly linear time: Positive eigenvectors, m-matrices, graph kernels, and other applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2019, San Diego, California, USA, Jan- uary 6-9, 2019, pages 1387–1404.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [AKM+20] AmirMahdi Ahmadinejad, Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, Jack Murtagh, John Peebles, Aaron Sidford, and Salil P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vadhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' High-precision estimation of random walks in small space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Sandy Irani, editor, 61st IEEE Annual Symposium on Foundations of Computer Science, FOCS 2020, Durham, NC, USA, November 16-19, 2020, pages 1295–1306.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [AM85] Noga Alon and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Milman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' lambda1, isoperimetric inequalities for graphs, and superconcentrators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Comb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theory, Ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' B, 38(1):73–88, 1985.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [BK00] Andr´as A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Bencz´ur and David R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Karger.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Augmenting undirected edge connectivity in ˜o(n2) time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Algorithms, 37(1):2–36, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [BSS12] Joshua D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Batson, Daniel A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Spielman, and Nikhil Srivastava.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Twice-ramanujan sparsifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=', 41(6):1704–1721, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CGP+18] Timothy Chu, Yu Gao, Richard Peng, Sushant Sachdeva, Saurabh Sawlani, and Junx- ing Wang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Graph sparsification, spectral sketches, and faster resistance computation, via short cycle decompositions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Mikkel Thorup, editor, 59th IEEE Annual Sympo- sium on Foundations of Computer Science, FOCS 2018, Paris, France, October 7-9, 2018, pages 361–372.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CKK+18] Michael B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Cohen, Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, Rasmus Kyng, John Peebles, Richard Peng, Anup B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Rao, and Aaron Sidford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Solving directed laplacian systems in nearly-linear time through sparse LU factorizations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Mikkel Thorup, editor, 59th IEEE Annual Symposium on Foundations of Computer Science, FOCS 2018, Paris, France, October 7-9, 2018, pages 898–909.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CKM+11] Paul F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Christiano, Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, Aleksander Madry, Daniel A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Spielman, and Shang-Hua Teng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Electrical flows, laplacian systems, and faster approximation of maximum flow in undirected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Lance Fortnow and Salil P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vadhan, editors, Proceedings of the 43rd ACM Symposium on Theory of Computing, STOC 2011, San Jose, CA, USA, 6-8 June 2011, pages 273–282.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CKM+14] Michael B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Cohen, Rasmus Kyng, Gary L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Miller, Jakub W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Pachocki, Richard Peng, Anup B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Rao, and Shen Chen Xu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Solving SDD linear systems in nearly mlog1/2n time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In David B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Shmoys, editor, Symposium on Theory of Computing, STOC 2014, New York, NY, USA, May 31 - June 03, 2014, pages 343–352.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 36 [CKP+16] Michael B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Cohen, Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, John Peebles, Richard Peng, Aaron Sidford, and Adrian Vladu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Faster algorithms for computing the stationary distribution, sim- ulating random walks, and more.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Irit Dinur, editor, IEEE 57th Annual Symposium on Foundations of Computer Science, FOCS 2016, 9-11 October 2016, Hyatt Regency, New Brunswick, New Jersey, USA, pages 583–592.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CKP+17] Michael B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Cohen, Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, John Peebles, Richard Peng, Anup B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Rao, Aaron Sidford, and Adrian Vladu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Almost-linear-time algorithms for markov chains and new spectral primitives for directed graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Hamed Hatami, Pierre McKenzie, and Valerie King, editors, Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2017, Montreal, QC, Canada, June 19-23, 2017, pages 410–419.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [CLTW22] Lijie Chen, Xin Lyu, Avishay Tal, and Hongxun Wu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Personal communication, November 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [GV22] Louis Golowich and Salil Vadhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Pseudorandomness of expander random walks for symmetric functions and permutation branching programs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Anindya De, editor, Proceedings of the 37th Computational Complexity Conference (CCC ‘22), volume 234 of LIPIcs, pages 27:1–27:15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Schloss Dagstuhl - Leibniz-Zentrum f¨ur Informatik, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Full version posted as ECCC TR22-024.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [HPV21] William M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Hoza, Edward Pyne, and Salil P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vadhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Pseudorandom generators for unbounded-width permutation branching programs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In James R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lee, editor, 12th Innovations in Theoretical Computer Science Conference, ITCS 2021, January 6-8, 2021, Virtual Conference, volume 185 of LIPIcs, pages 7:1–7:20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Schloss Dagstuhl - Leibniz-Zentrum f¨ur Informatik, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [INW94] Russell Impagliazzo, Noam Nisan, and Avi Wigderson.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Pseudorandomness for network algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Frank Thomson Leighton and Michael T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Goodrich, editors, Proceedings of the Twenty-Sixth Annual ACM Symposium on Theory of Computing, 23-25 May 1994, Montr´eal, Qu´ebec, Canada, pages 356–364.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [JS21] Arun Jambulapati and Aaron Sidford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Ultrasparse ultrasparsifiers and faster lapla- cian system solvers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In D´aniel Marx, editor, Proceedings of the 2021 ACM-SIAM Symposium on Discrete Algorithms, SODA 2021, Virtual Conference, January 10 - 13, 2021, pages 540–559.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [Kel21] Zander Kelley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An alternative proof that inw fools permutation programs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Unpub- lished manuscript, February 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [KLOS14] Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, Yin Tat Lee, Lorenzo Orecchia, and Aaron Sidford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An almost- linear-time algorithm for approximate max flow in undirected graphs, and its multi- commodity generalizations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Chandra Chekuri, editor, Proceedings of the Twenty- Fifth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2014, Portland, Oregon, USA, January 5-7, 2014, pages 217–226.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [KM09] Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner and Aleksander Madry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Faster generation of random spanning trees.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In 50th Annual IEEE Symposium on Foundations of Computer Science, FOCS 2009, October 25-27, 2009, Atlanta, Georgia, USA, pages 13–21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 37 [KMG22] Rasmus Kyng, Simon Meierhans, and Maximilian Probst Gutenberg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Derandomizing directed random walks in almost-linear time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' CoRR, abs/2208.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10959, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [KMP12] Jonathan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Kelner, Gary L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Miller, and Richard Peng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Faster approximate multi- commodity flow using quadratically coupled flows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Howard J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Karloff and Toniann Pitassi, editors, Proceedings of the 44th Symposium on Theory of Computing Con- ference, STOC 2012, New York, NY, USA, May 19 - 22, 2012, pages 1–18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [KMP14] Ioannis Koutis, Gary L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Miller, and Richard Peng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Approaching optimality for solving SDD linear systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=', 43(1):337–354, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [KVV04] Ravi Kannan, Santosh S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vempala, and Adrian Vetta.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' On clusterings: Good, bad and spectral.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 51(3):497–515, 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [LRS13] Yin Tat Lee, Satish Rao, and Nikhil Srivastava.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A new approach to computing maxi- mum flows using electrical flows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Dan Boneh, Tim Roughgarden, and Joan Feigen- baum, editors, Symposium on Theory of Computing Conference, STOC’13, Palo Alto, CA, USA, June 1-4, 2013, pages 755–764.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [LS13] Yin Tat Lee and Aaron Sidford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Efficient accelerated coordinate descent methods and faster algorithms for solving linear systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In 54th Annual IEEE Symposium on Foundations of Computer Science, FOCS 2013, 26-29 October, 2013, Berkeley, CA, USA, pages 147–156.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' IEEE Computer Society, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [LS18] Yin Tat Lee and He Sun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Constructing linear-sized spectral sparsification in almost- linear time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=', 47(6):2315–2336, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [MST15] Aleksander Madry, Damian Straszak, and Jakub Tarnawski.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Fast generation of ran- dom spanning trees and the effective resistance metric.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Piotr Indyk, editor, Pro- ceedings of the Twenty-Sixth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2015, San Diego, CA, USA, January 4-6, 2015, pages 2019–2036.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' SIAM, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [OSV12] Lorenzo Orecchia, Sushant Sachdeva, and Nisheeth K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vishnoi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Approximating the exponential, the lanczos method and an ˜o(m)-time spectral algorithm for balanced separator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Howard J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Karloff and Toniann Pitassi, editors, Proceedings of the 44th Symposium on Theory of Computing Conference, STOC 2012, New York, NY, USA, May 19 - 22, 2012, pages 1141–1160.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [PS14] Richard Peng and Daniel A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Spielman.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An efficient parallel solver for SDD linear systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' STOC, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [PV21] Edward Pyne and Salil Vadhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Pseudodistributions That Beat All Pseudorandom Generators (Extended Abstract).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Proceedings of the 36th Annual Computational Complexity Conference (CCC), pages 33:1–33:15, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [PY19] Merav Parter and Eylon Yogev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Optimal short cycle decomposition in almost lin- ear time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Christel Baier, Ioannis Chatzigiannakis, Paola Flocchini, and Stefano Leonardi, editors, 46th International Colloquium on Automata, Languages, and Pro- gramming, ICALP 2019, July 9-12, 2019, Patras, Greece, volume 132 of LIPIcs, pages 89:1–89:14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Schloss Dagstuhl - Leibniz-Zentrum f¨ur Informatik, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 38 [RV05] Eyal Rozenman and Salil P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vadhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Derandomized squaring of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Chandra Chekuri, Klaus Jansen, Jos´e D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Rolim, and Luca Trevisan, editors, Approxima- tion, Randomization and Combinatorial Optimization, Algorithms and Techniques, 8th International Workshop on Approximation Algorithms for Combinatorial Opti- mization Problems, APPROX 2005 and 9th InternationalWorkshop on Randomization and Computation, RANDOM 2005, Berkeley, CA, USA, August 22-24, 2005, Proceed- ings, volume 3624 of Lecture Notes in Computer Science, pages 436–447.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Springer, 2005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [SS08] Daniel A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Spielman and Nikhil Srivastava.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Graph sparsification by effective resis- tances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Cynthia Dwork, editor, Proceedings of the 40th Annual ACM Symposium on Theory of Computing, Victoria, British Columbia, Canada, May 17-20, 2008, pages 563–568.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [ST04] Daniel A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Spielman and Shang-Hua Teng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Nearly-linear time algorithms for graph partitioning, graph sparsification, and solving linear systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In L´aszl´o Babai, editor, Proceedings of the 36th Annual ACM Symposium on Theory of Computing, Chicago, IL, USA, June 13-16, 2004, pages 81–90.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [Tro12] Joel A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Tropp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' User-friendly tail bounds for sums of random matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Found.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=', 12(4):389–434, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [Vad12] Salil P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Vadhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Pseudorandomness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Found.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Trends Theor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=', 7(1-3):1– 336, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [vdBGJ+22] Jan van den Brand, Yu Gao, Arun Jambulapati, Yin Tat Lee, Yang P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Liu, Richard Peng, and Aaron Sidford.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Faster maxflow via improved dynamic spectral vertex sparsi- fiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Stefano Leonardi and Anupam Gupta, editors, STOC ’22: 54th Annual ACM SIGACT Symposium on Theory of Computing, Rome, Italy, June 20 - 24, 2022, pages 543–556.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' [vdBLL+21] Jan van den Brand, Yin Tat Lee, Yang P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Liu, Thatchaphol Saranurak, Aaron Sidford, Zhao Song, and Di Wang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Minimum cost flows, mdps, and l1-regression in nearly linear time for dense instances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In Samir Khuller and Virginia Vassilevska Williams, editors, STOC ’21: 53rd Annual ACM SIGACT Symposium on Theory of Computing, Virtual Event, Italy, June 21-25, 2021, pages 859–869.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' ACM, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A SV Approximation Proofs In this section we provide proofs of equivalences for definitions related to SV-approximation (Ap- pendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1), properties of SV-approximation (Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2), and separations of SV-approximation from other notions of approximation (Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We recall results useful in the proofs: Fact A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ([CKP+17]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A ∈ Cn×n be a Hermitian matrix and I ⊂ [n] be arbitrary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then A is PSD if and only if AI,I is PSD and SCI(A) is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We require a result that matrix approximation is approximately preserved under Schur comple- ments: 39 Theorem A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 (Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8 [AKM+20]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given N, ˜N ∈ Cn×n such that ∥N∥ ≤ 1, suppose ˜N is an ε-approximation of N with respect to SN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for F ⊂ [n] such that IF,F − NF,F is invertible, we have that IF c − SCF(I − ˜N) is an ε + O(ε2) approximation of IF c − SCF(I − N) with respect to SIF c−SCF (I−N).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We also prove a useful property of matri approximation: Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 (Manipulating matrix approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n, let E ∈ Cm×m, F ∈ Cn×n be PSD and Hermitian, and assume that A is an ε-approximation of �A with respect to error matrices E and F, for some ε ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For U ∈ Cm′×m, V ∈ Cn×n′, U�AV is an ε-approximation of UAV with respect to error matrices E′ and F′, where E′ = UEU∗ and F′ = V∗FV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For PSD Hermitian E′ ∈ Cm×m and F′ ∈ Cn×n a with E ⪯ cE′ and F ⪯ cF′ for a constant c ≥ 0, then �A is a cε-approximation of A with respect to error matrices E′ and F′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For arbitrary x ∈ Cm and y ∈ Cn the definition of ε-approximation implies, ���x∗(U �AV − UAV)y ��� = ���x∗U( �A − A)Vy ��� ≤ ε 2 (x∗UEU∗x + y∗V∗FVy) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For arbitrary x ∈ Cm, y ∈ Cn, ���x∗( �A − A)y ��� ≤ ε 2 (x∗Ex + y∗Fy) ≤ cε 2 � x∗E′x + y∗F′y � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 Equivalences Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6 (Conditions for SV approximation to be defined).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n, and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices such that ker(Din) ⊆ lker(A) and ker(Dout) ⊆ rker(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the following are equivalent: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' σmax(D+/2 in AD+/2 out ) ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Din − AD+ outA∗ is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Dout − A∗D+ inA is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For some scalar z ∈ C with |z| = 1, � Din zA z∗A∗ Dout � is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every scalar z ∈ C with |z| ≤ 1, � Din zA z∗A∗ Dout � is PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Suppose further that Din = diag(din) and Din = diag(dout) for din ∈ Rm ≥0, dout ∈ Rn ≥0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then Condition 1 below implies Condition 2 below, which implies Conditions 1–5 above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A is nonnegative, dout = A⃗1n, din = ⃗1⊤ mA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For all i ∈ [n], (dout)i ≥ � j |Ai,j|, and for all j ∈ [m], (din)j ≥ � i |Ai,j|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We first show Conditions 1 − 5 are equivalent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 40 1→ 3 Condition 1 implies for all y, y∗D+/2 out A∗D+ inAD+/2 out y ≤ y∗y and letting y ← D1/2 outy and using that ker(Dout) ⊆ ker(A) implies Condition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 → 1 Condition 3 implies for all y we have y∗A∗D+ inAy ≤ y∗Douty and taking y ← D+/2 out y we have y∗D+/2 out A∗D+ inAD+/2 out y ≤ y∗ΠDouty ≤ y∗y which implies Condition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3→ 5 Letting z with |z| ≤ 1 be arbitrary, we have that by Condition 3 Din − z∗A∗D+ outzA = Din − A∗D+ outA ⪰ 0, and since Din ⪰ 0 we apply Fact A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 and conclude Condition 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An analogous argument shows 2→ 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The fact that Fact A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 is an equivalence then implies 5→2 and 5→3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 5 → 4 Immediate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4 → 2 By Fact A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 and the fact that z∗z = 1 we have Din − z∗A∗D+ outzA = Din − A∗D+ outA ⪰ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We next show the latter set of conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1→2 For all i ∈ [n] we have [dout]i = � j Ai,j = � j |Ai,j|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2 → 5 We use that a Hermitian matrix that is diagonally dominant is positive semidefinite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We have (Din)i,i ≥ � j |zAi,j| and (Dout)i,i ≥ � j |z∗A∗ j,i| so �Din A A∗ Dout � is diagonally dominant and thus PSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 (Equivalent formulations of SV approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the following are equivalent 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' slift � �A � D≈ε slift � �B � , where slift � �A � = �0m×m A A∗ 0n×n � , slift (A) = � 0m×m �A �A∗ 0n×n � , D = � Din 0m×n 0n×m Dout � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For some scalar z ∈ C with |z| = 1, ˜C is ε/2-approximation of C with respect to error matrix E, where C = �0m×m zA 0n×m 0n×n � , ˜C = � 0m×m z �A 0n×m 0n×n � , E = � Din zA z∗A∗ Dout � 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Item 3 holds for every z ∈ C such that |z| ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We prove 1 ⇐⇒ 2 and 1 → 4 → 3 → 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1 ⇐⇒ 2 Suppose Item 1 holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for arbitrary test vectors x = �x1 x2 � , y = �y1 y2 � where x1, y1 ∈ Cm and x2, y2 ∈ Cn we have ���x∗(B − �B)y ��� ≤ ���x∗ 1(A − �A)y2 ��� + ���x∗ 2(A∗ − �A∗)y1 ��� ≤ ε 4 � x∗ 1Dinx1 + y∗ 2Douty2 + x∗ 1AD+ outA∗x1 + y∗ 2A∗D+ inAy2 � + ε 4 � x∗ 2Doutx2 + y∗ 1Diny1 + x∗ 2A∗D+ inAx2 + y∗ 1AD+ outA∗y1 � = ε 4 � x∗Dx + y∗Dy + x∗BD−1B∗x + y∗B∗D−1By � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 41 Furthermore, ker(D) ⊆ lker(B), ker(D) ⊆ rker(B) and D − BD+B ⪰ 0 by Item 5 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the other direction, we obtain 2 → 1 by considering the set of test vectors �x1 0n � and �0m y2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1 → 4: Fix arbitrary z with |z| ≤ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We have that E ⪰ 0 by Item 5 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for arbitrary test vectors x = �x1 x2 � , y = �y1 y2 � , let x′ = �x1 x′ 2 � , y′ = �y1 y′ 2 � where x′ 2 = argminv∈Cn � x∗ 1 v∗� � Din zA z∗A∗ Dout � �x1 v � , y′ 1 = argminv∈Cm � y∗ 2 v∗� �Dout z∗A∗ zA Din � �y2 v � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus x′∗ � Din zA z∗A∗ Dout � x′ = x∗ 1SC �� Din zA z∗A∗ Dout �� x1, y′∗ � Din zA z∗A∗ Dout � y′ = y∗ 2SC ��Dout z∗A∗ zA Din �� y2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then we have ���x∗(˜C − C)y ��� = ���(x1z∗)∗( �A − A)(y2z) ��� ≤ ε 4 � x∗ 1(Din − zAD+ outA∗z∗)x1 + y∗ 2(Dout − z∗A∗D+ inAz)y2 � = ε 4 � x∗ 1SC ��Din z∗A∗ zA Dout �� x1 + y∗ 2SC ��Dout zA z∗A∗ Din �� y2 � = ε 4 � x′∗ � Din zA z∗A∗ Dout � x′ + y′∗ � Din zA z∗A∗ Dout � y′ � ≤ ε 4 (x∗Ex + y∗Ey) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 4 → 3 Immediate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 3 → 1 For arbitrary test vectors x1 ∈ Cm,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' y2 ∈ Cn,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' let x = �zx1 x2 � ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' y = � y1 z∗y2 � where x2 = argminv∈Cn � z∗x∗ 1 v∗� � Din zA z∗A∗ Dout � �zx1 v � ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' y1 = argminv∈Cm � zy∗ 2 v∗� �Dout z∗A∗ zA Din � �z∗y2 v � Then ���x∗ 1( �A − A)y2 ��� = ���x∗(˜C − C)x ��� ≤ ε 4 (x∗Ex + y∗Ey) = ε 4 � z∗x∗ 1SC ��Din z∗A∗ zA Dout �� x1z + zy∗ 2SC ��Dout zA z∗A∗ Din �� z∗y2 � = ε 4 � z∗x∗ 1(Din − zAD+ outA∗z∗)zx1 + zy∗ 2(Dout − z∗A∗D+ inAz)z∗y2 � = ε 4 � x∗ 1(Din − AD+ outA∗)x1 + y∗ 2(Dout − A∗D+ inA)y2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 42 Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n, and let Din ∈ Cm×m and Dout ∈ Cn×n be PSD matrices such that ker(Din) ⊆ lker(A), and ker(Dout) ⊆ rker(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let N = D+/2 in AD+/2 out and ˜N = D+/2 in �AD+/2 out Then �A Din,Dout ≈ε A if and only if ˜N svn ≈ ε N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Suppose �A Din,Dout ≈ε A, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �A is an ε/2-approximation of A with respect to E = Din − AD+ outA∗, F = Dout − AD+ inA∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Applying Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 with U = D+/2 in , V = D+/2 out we obtain that ˜N is an ε/2-approximation of N with respect to error matrices D+/2 in (Din − AD+ outA∗)D+/2 in ⪯ I − (D+/2 in AD+/2 out )(D+/2 out A∗D+/2 in ) = I − NN∗ and D+/2 out (Dout − A∗D+ inA)D+/2 out ⪯ I − (D+/2 out A∗D+/2 in )(D+/2 in AD+/2 out ) = I − N∗N and hence we obtain that ˜N svn ≈ ε N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The other direction is analogous, and the kernel properties follow as ker(Din) ⊆ lker(D1/2 in D+/2 in AD+/2 out D1/2 out) and ker(Dout) ⊆ rker(D1/2 in D+/2 in AD+/2 out D1/2 out).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='9 (Unitary transformation characterization of SV-approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For ˜N, N ∈ Cn×n, we have that ˜N svn ≈ ε N if for every pair of unitary matrices U, V, U ˜NV is a standard ε/2- approximation of UNV with respect to degree matrix I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, if ˜N svn ≈ ε N then for every pair of matrices U, V satisfying ∥U∥ ≤ 1, ∥V∥ ≤ 1, we have that U ˜NV svn ≈ ε UNV, and hence U ˜NV is a standard ε-approximation of UNV with respect to degree matrix I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Suppose for every pair of unitary matrices U, V, we have for all x, y ∈ Cn that |x∗(U�AV − UAV)y| ≤ ε 2 · � x∗(I − SUAV)x · � y∗(I − SUAV)y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We will pick u, v, U, V to depend on x, y and the result will follow immediately from invoking the above equation with x, y ← u, v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Set U = I, u = x/∥x∥, and v = Ay/∥Ay∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that u∗v = (x∗Ay)/(∥x∥ · ∥y∥), so u and v have the same angle between them as A∗x and y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Hence, there exists a unitary matrix V such that Vu = A∗x/∥A∗x∥ and Vv = y/∥y∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Substituting gives ���x∗( �A − A)y ��� ∥x∥∥y∥ ≤ ε 2 · � 1 − x∗(AA∗)x ∥x∥∥A∗x∥ · � 1 − y∗(A∗A)y ∥y∥∥Ay∥ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Applying ∥A∗∥ = ∥A∥ ≤ 1 gives the desired result.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the other direction, we start with Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 with z = 1 applied to the approxi- mation statement ˜N svn ≈ ε N and apply Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 with left and right hand side matrices U′ ← �U 0 0 V∗ � and V′ ← �U∗ 0 0 V � 43 This results in the approximation statement that for the matrices given below, R, �R ǫ/2- approximate each other with respect to E where R = �0 UNV 0 0 � �R = � 0 U ˜NV 0 0 � , and E = � UIU∗ UNV (UNV)∗ V∗IV � ⪯ � I UNV (UNV)∗ I � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Hence, by Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 we conclude UNV svn ≈ ε U ˜NV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Corollary 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10 (SV preservation under multiplication by permutation matrices).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n ≥0 and suppose �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let U, V be arbitrary permutation matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then U �AV D′ in,D′ out ≈ε UAV where D′ in = UDinU∗ and D′ out = V∗DoutV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if �A sv≈ε A, then U�AV sv≈ε UAV and if �A svn ≈ ε A then U �AV svn ≈ ε UAV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We start with Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 with z = 1 applied to the approximation statement we are given in the lemma statement and apply Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 with left and right hand side matrices U′ ← �U 0 0 V∗ � and V′ ← �U∗ 0 0 V � This results in the approximation statement that for the matrices given below, R, �R ǫ/2-approximate each other with respect to E where R = �0 UAV 0 0 � �R = � 0 U �AV 0 0 � , and E = �UDinU∗ UAV (UAV)∗ V∗DoutV � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Hence, Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 is satisfied for SV approximation of UAV and U �AV with respect to UDinU∗ and V∗DoutV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For the first “consequently” claim, observe that UAV⃗1 = UA⃗1 since V is a permutation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Hence, diag(UAV⃗1) = UDinU∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A similar analysis shows diag(⃗1⊤UAV) = V∗DinV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' the second statement follows from UU∗ = V∗V = I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 Properties We prove that SV approximation implies standard approximation: Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cn×n and suppose �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then, �A is an ε-approximation of A with respect to E = Din − A, F = Dout − A∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This follows immediately from specializing Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 with z = −1 to test vectors of the form �x x � , �y y � It likewise implies UC approximation: Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If A, �A ∈ Cn×n with �A D≈ε A then �A ≈ε A with respect to degree matrix D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 44 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Item 4 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 we have that for every pair of test vectors �x x � , �y y � and every unit magnitude z, we have ���x∗(zA − z �A)y ��� ≤ ε 4 · 2 (x∗Dx + y∗Dy + ℜ(zx∗Ax + zyAy∗)) Then choosing z to minimize ℜ(zx∗Ax + zyAy∗), we obtain ���x∗(A − �A)y ��� ≤ 2ε 4 (x∗Dx + y∗Dy − |x∗Ax + yAy∗|) Which is precisely the condition for �A ≈ε A with respect to D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='14 (SV preservation under arbitrary lifting).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Cm×n be matrices such that �A Din,Dout ≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then for all integers i, j, k, ℓ ≥ 0 \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j �A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb D′ in,D′ out ≈ε \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Where D′ in = \uf8ee \uf8f0 0i×i 0i×n 0i×ℓ 0n×i Din 0n×ℓ 0ℓ×i 0ℓ×n 0ℓ×ℓ \uf8f9 \uf8fb , D′ out = \uf8ee \uf8f0 0j×j 0j×m 0j×k 0m×j Din 0m×k 0k×j 0k×m 0k×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if �A sv≈ε A then \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j �A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb sv≈ε \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Observe by expanding the definition of SV approximation that it is preserved under em- bedding into the first principal diagonal block of any larger matrix that is zeros elsewhere.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' That is, � �A 0m×(j+k) 0(i+ℓ)×n 0(i+ℓ)×(j+k) � D′ in,D′ out ≈ε � A 0m×(j+k) 0(i+ℓ)×n 0(i+ℓ)×(j+k) � (29) where D′ in, D′ out are the respective principle embeddings of the degree matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let U, V be permutation matrices such that U � A 0m×(j+k) 0(i+ℓ)×n 0(i+ℓ)×(j+k) � V = \uf8ee \uf8f0 0i×j 0i×n 0i×k 0m×j A 0m×k 0ℓ×j 0ℓ×n 0ℓ×k \uf8f9 \uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It suffices to argue that we can apply this transformation to both sides of Equation (29) and have the approximation still hold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This is implied by Corollary 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='10 with D′ in ← UD′ inU∗ and D′ out ← V∗DoutV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The “consequently” claim holds from inspecting the resulting block structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='16 (SV preservation under products).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let (Ni)i∈[ℓ], ( ˜Ni)i∈[ℓ] ∈ Cn×n be such that for every i, ˜Ni svn ≈ ε Ni.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then ˜Nℓ · · · ˜N2 ˜N1 svn ≈ ε+O(ǫ2) Nℓ · · · N2N1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 45 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Using preservation under lifting and sums (Lemmas 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='14 and 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='18) and that SV approx- imation implies standard approximation for square matrices (Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4), we can obtain the approximation statement that the matrices R, �R given below ǫ/2-standard-approximate each other with respect to the symmetrization SR of the first one: R = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 I/2 Nℓ 0 0 I Nℓ−1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' I N1 0n×n 0 I/2 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fb and �R = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 I/2 ˜Nℓ 0 0 I ˜Nℓ−1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' I ˜N1 0n×n 0 I/2 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fb .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Taking the Schur complement by eliminating the center n(l −1)×n(l −1) block gives the following two matrices R = � I/2 Nℓ · · · N2N1 0n×n I/2 � and �R = � I/2 ˜Nℓ · · · ˜N2 ˜N1 0n×n I/2 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' And we obtain by Theorem A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 that �R is an ǫ/2 + O(ǫ2)-approximation of R with respect to SR.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Hence, by Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7, we have ˜Nℓ · · · ˜N2 ˜N1 svn ≈ ε+O(ǫ2) Nℓ · · · N2N1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If �A sv≈ε A, then �A⃗1 = A⃗1 and �A⊤⃗1 = A⊤⃗1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let x ← δx be an arbitrary test vector and let y = ⃗1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then the approximation condition implies ���δx∗( �A − A)⃗1 ��� ≤ ε 4 � δ2x∗Doutx − δ2x∗AD+ inA∗x +⃗1⊤Din⃗1 −⃗1⊤A⊤D+ outA⃗1 � = ε 4 � δ2x∗Doutx − δ2x∗AD+ inA∗x � and so taking δ → 0 we obtain that A⃗1 = �A⃗1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' An analogous argument setting x = ⃗1 shows that ⃗1⊤A = ⃗1⊤ �A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If �Ai (Din)i,(Dout)i ≈ε Ai for all i ∈ [k], letting Din def = � i∈[k](Din)i and Dout def = � i∈[k](Dout)i, then � i∈[k] �Ai Din,Dout ≈ε � i∈[k] Ai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if �Ai sv≈ε Ai for every i, then � i∈[k] �Ai sv≈ε � i∈[k] Ai.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This follows immediately from Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 being linear over A and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The “consequently” claim follows immediately from noting \uf8eb \uf8ed� i∈[k] A \uf8f6 \uf8f8⃗1 = � i∈[k] (A⃗1) and \uf8eb \uf8ed� i∈[k] A⊤ \uf8f6 \uf8f8⃗1 = � i∈[k] (A⊤⃗1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' If A3 Din,Dout ≈δ A2 and A2 Din,Dout ≈ε A1 then A3 Din,Dout ≈ε+δ+εδ A1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Consequently, if A3 sv≈δ A2 and A2 sv≈εA1 then A3 sv≈ε+δ+εδA1, and if A3 svn ≈ δ A2 and A2 svn ≈ εA1 then A3 svn ≈ ε+δ+εδA1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, if for δ ∈ (0, 1/2) and A0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' , Aℓ we have Ai sv≈δ/2ℓ Ai−1 for every i, then Aℓ sv≈δ A0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Moreover, the equivalent claim holds for normalized SV approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 46 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 we have �Din A∗ 2 A2 Dout � ⪯ (1 + ε) �Din A∗ 1 A1 Dout � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus for arbitrary x ∈ Cm,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' y ∈ Cn we have ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='����x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='��0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='− ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='≤ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='����x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='��0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='− ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='���� + ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='����x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='��0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='− ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='���� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='≤ δ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x + y∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='+ ε ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x + y∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='≤ δ(1 + ε) ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x + y∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='+ ε ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x + y∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='= δ + ε + εδ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='x + y∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='�Din ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A∗ ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='A1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='Dout ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='y ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='� ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='and so we conclude by Item 3 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The first “consequently” claim follows from the fact that by Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='17 we have A2⃗1 = A⃗1 and A⊤ 2 ⃗1 = A⊤ 1 ⃗1, and the second follows immediately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For the “moreover” claim, we prove by induction on i that Ai sv≈iδ/2ℓ+iδ2/ℓ A0, which suffices as ℓδ/2ℓ + ℓδ2/ℓ ≤ δ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Assuming this holds for i, we have Ai+1 sv≈δ/2ℓ Ai sv≈iδ/2ℓ+iδ2/ℓ A0 and thus we have Ai+1 sv≈(i+1)δ/2ℓ+γ A0 for γ = iδ2 ℓ + δ 2ℓ � iδ 2ℓ + iδ2 ℓ � ≤ iδ2 ℓ + δ2/2ℓ + δ3/2ℓ2 ≤ (i + 1)δ2 ℓ as desired.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The proof for normalized SV approximation is essentially identical, so we omit it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G be a strongly connected, d-regular directed multigraph on n vertices with adjacency matrix A and let J ∈ Rn×n be a matrix with 1/n in every entry (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' J is the walk matrix of the complete graph with a self loop on every vertex).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then λ(G) ≤ 1 − λ/2 if and only if A/d svn ≈ λ J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that by definition of A svn ≈ λ J, |x∗(A/d − J)y| ≤ λ 4 (x∗(I − JJ∗)x + y∗(I − J∗J)y) = λ 4 (x∗(I − J)x + y∗(I − J)y) Therefore, A/d svn ≈ λ J if and only if A/d ≈λ/2 J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By [AKM+20, Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2] A/d ≈λ/2 J if and only if λ(G) ≤ 1 − λ/2, so the result follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 47 We show a slightly stronger version of Item 2 of Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 for non-negative matrices, in that we show SV sparsification of the bipartite lift of a graph implies SV sparsification of the undirlying directed graph, even if the sparsifier is not promised to be bipartite-preserving, and even if the sparsifier is not itself undirected.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given non-negative A ∈ Rm×n ≥0 , suppose there is nonnegative �A = � �AF,F �AF,F c �AF c,F �AF c,F c � such that �A sv≈ε �0 A⊤ A 0 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �AF,F = 0 and �AF c,F c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' �AF c,F sv≈ε A and �AF,F c sv≈ε A⊤.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let Dout = diag(A⃗1) and Din = diag(⃗1⊤A) and observe ⃗1⊤Dout⃗1−⃗1⊤AD+ inA⃗1 = ⃗1⊤Din⃗1− ⃗1⊤A⊤D+ outA⊤⃗1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let x ∈ Cn be arbitrary, and consider the test vectors x′ = δ � x 0m � , y′ = �⃗1 ⃗1 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then from the definition of SV approximation we obtain ���δx⊤ �AF,F⃗1 + δx∗ �AF,F c⃗1 − δxA⊤⃗1 ��� = ����x′∗ ��0 A⊤ A 0 � − �A � y ���� ≤ ε 2 · δ2x∗(Dout − AD+ inA)x and hence �AF,F⃗1 + x �AF,F c⃗1 = xA⊤⃗1 by taking δ → 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Considering the test vectors x′ = δ � x 0m � , y′′ = � ⃗1 −⃗1 � we likewise obtain x �AF,F⃗1 − x �AF,F c⃗1 = −xA⊤⃗1 and hence x �AF,F⃗1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By an analogous argument we obtain �AF c,F c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For arbitrary x ∈ Cm, y ∈ Cn, apply the test vectors [0n, x], [y, 0m].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then we obtain ���x∗(A − �AF c,F )y ��� ≤ ε 2 � x∗(D − AD+AT )x + y∗(D − AT D+A)y � thus �AF c,F sv≈ε A and an analogous argument shows �AF,F c sv≈ε AT .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3 Separations Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every α, ε ∈ (0, 1), there are symmetric matrices W, � W with ∥W∥, ∥� W∥ ≤ 1 such that � W ≈ε W but � W is not an ε SV-approximation of W with respect to I − W for any ε′ < ε √ 1−α2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Given α, we define W = �α 0 0 −α � and � W = · � α ε √ 1 − α2 ε √ 1 − α2 −α � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 48 We first show that � W ≈ε W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Since both matrices are symmetric, suffices [AKM+20, Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7] to show the equivalent statement for PSD approximation between I− � W, I−W and I+ � W, I+W, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' (1 − ε)(I − W) ⪯ I − � W ⪯ (1 + ε)(I − W) (1 − ε)(I + W) ⪯ I + � W ⪯ (1 + ε)(I + W).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' All four inequalities are implied by � ε(1 − α) ε √ 1 − α2 ε √ 1 − α2 ε(1 + α) � ⪰ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Using Fact A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1 this is equivalent to ε(1 − α) − ε(1 − α2)(1 + α)−1 ≥ 0 which holds with equality, so � W ≈ε W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Now assume that � W svn ≈ ε′ W for some ε′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' From the definition of SV approximation applied to test vector z = �x x � , we have ±(� W − W) ⪯ ε′ � I − W2� = ε′(1 − α2)I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' This is equivalent to ǫ � 1 − α2 ≤ ε′(1 − α2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus for � W svn ≈ ε W, a necessary condition is for ε′ ≥ ε √ 1−α2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' B Facts about Singular Values Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let σi(·) denote the ith largest singular value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For any matrix A ∈ Cn×n and any positive integer k, σ2(Ak) ≤ σ2(A) · σ1(A)k−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By the variational characterization of singular values, we have σ2(Ak) = min v∈Cn max x⊥v ∥Akx∥ ∥x∥ ≤ min v∈Cn max x⊥v ∥Ak−1∥ · ∥Ax∥ ∥x∥ ≤ ∥A∥k−1·min v∈Cn max x⊥v ∥Ax∥ ∥x∥ = σ2(A)·σ1(A)k−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let σi(·) denote the ith largest singular value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For any matrix A, B ∈ Cn×n, σ2(A + B) ≤ σ2(A) + σ1(B).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Furthermore, if A, B share a common right singular vector that achieves the maximum singular value in each of them respectively, then σ2(A + B) ≤ σ2(A) + σ2(B) Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For the first part of the statement, by the variational characterization of singular values, we have σ2(A + B) = min v∈Cn max x⊥v ∥(A + B)x∥ ∥x∥ ≤ min v∈Cn max x⊥v ∥Ax∥ + ∥Bx∥ ∥x∥ ≤ min v∈Cn max x⊥v ∥Ax∥ ∥x∥ + ∥B∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 49 For the second part of the statement, let v denote a common right singular vector of A, B that respectively achieves the maximum singular value in each of them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then we have σ2(A + B) = min v′∈Cn max x⊥v′ ∥(A + B)x∥ ∥x∥ ≤ max x⊥v ∥(A + B)x∥ ∥x∥ ≤ max x⊥v ∥Ax∥ + ∥Bx∥ ∥x∥ ≤ max x⊥v ∥Ax∥ ∥x∥ + max x⊥v ∥Bx∥ ∥x∥ = σ2(A) + σ2(B).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The third part of the statement has essentially the same proof as the second.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We now use these properties to show that small perturbations preserve the smallest singular value: Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let σ2(·) denote the second largest singular value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let G be an Eulerian graph such that σ2(D−1/2 G AGD−1/2) ≤ 1 − 1/γ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For every H such that DH = DG and ∥AH − AG∥ ≤ δ, we have σ2(D−1/2 H AHD−1/2 H ) ≤ 1 − 1/γ + δ/dmin where dmin is the minimum diagonal entry of DG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' By lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2, σ2(D−1/2 H AHD−1/2 H ) ≤ σ2(D−1/2 G AGD−1/2) + ∥D−1/2 G (AH − AG)D−1/2∥ ≤ 1 − 1/γ + ∥D−1/2 G ∥2 · ∥AH − AG∥ = 1 − 1/γ + δ/dmin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A, �A ∈ Rn×n be the adjacency matrices of Eulerian graphs with no isolated vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Suppose �A sv≈ε A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let D be their diagonal matrix of degrees.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Then 1 − σ2(D−1/2 �AD−1/2) ≤ (1 + 2ε) · (1 − σ2(D−1/2AD−1/2)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Define N = D−1/2AD−1/2 and ˜N = D−1/2 �AD−1/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It suffices to show ±(σ2( ˜N) − σ2(N)) ≤ σ2( ˜N − N) ≤ 2ε · (1 − σ2(N)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The first inequality follows from lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For the second inequality, note that the definition of SV approximation can be rewritten as for all x ∈ Cn, ∥( ˜N − N)x∥ ≤ ε · (1 − ∥Nx∥2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that ˜N, N have a common left and right eigenvector given by v = D1/2⃗1, and this achieves their respective maximum singular values of 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='7 Hence, if we wish to maximize the LHS over all unit vectors x, it suffices to maximize over all unit vectors x ⊥ v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Doing so, we obtain ∥ ˜N − N∥ ≤ ε · (1 − σ2(N)2) ≤ 2ε · (1 − σ2(N)) where the last inequality is because ∥N∥ ≤ 1 ([CKP+17] lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 7We know it corresponds to a singular value of 1 and this its not possible to have a singular value larger than 1 for this matrix ([CKP+17] lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 50 Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let σi(·) denote the ith largest singular value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let A ∈ Rn×n be the adjacency matrix of a strongly connected Eulerian graph G with diagonal degree matrix D and no isolated vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let N = D−1/2AD−1/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let λ denote the second largest eigenvalue of SN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='8 Then we have σ2((1 − γ)N + γI) ≤ 1 − (1 − λ)γ + O(γ2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In particular, if γ ≥ 1/poly(n) and A has poly(n) weak mixing time and all weights are specified using O(log n) bits of precision, then σ2((1 − γ)N + γI) ≤ 1 − 1/poly(n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Let v denote a singular vector of (1 − γ)N + γI corresponding to its second-largest singular value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We then have [σ2((1 − γ)N + γI)]2 = (1 − γ)2v∗N∗Nv + 2γ(1 − γ)v∗SNv + γ2 ≤ (1 − γ)2 + 2γ(1 − γ)λ + γ2 ≤ 1 − 2(1 − λ)γ + O(γ2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' In the above, we used the fact that ∥N∥ ≤ 1 ([CKP+17] lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' To show the second part of the statement, let τ(M) denote the weak mixing time of the random walk on the graph corresponding to any Eulerian adjacency matrix M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Define L = I − N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Note that L is related to I − AD−1 by a change of basis with polynomial condition number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We may assume the graph is strongly connected as this is necessary for the weak mixing time to be finite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' It suffices to prove 1 1 − λ ≤ poly(n) · τ(N).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We do so by showing 1 1 − λ ≤ poly(n) · τ(SN) ≤ poly(n) · ∥S+ L∥ ≤ poly(n) · ∥L+∥ ≤ poly(n) · τ(N).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The first inequality is just the folklore result relating spectral gap to weak mixing time with an O(log n)-factor loss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The second and fourth inequalities are from ([CKP+16] Theorem 21).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We now show the third inequality, that ∥S+ L∥ ≤ poly(n) · ∥L+∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' We have ∥S+ L∥ ≤ poly(n) · ∥(LT S+ LL)+∥ = poly(n) · ∥SL+∥ ≤ poly(n) · (∥L+∥ + ∥LT+∥)/2 = poly(n) · ∥L+∥ where the first inequality in the line immediately above is a corollary of ([CKP+16] Lemma 13).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Lemma B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For matrices A, B ∈ Cn×n, ∥AB∥ = ∥(A∗A)1/2B∥, ∥BA∥ = ∥B(AA∗)1/2∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' For any matrix M, ∥M∥ = � λmax(M∗M) = max x∈Cm−{⃗0} x∗M∗Mx x∗x .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' Thus, ∥BA∥ = max x∈Cm−{⃗0} x∗B∗A∗ABx x∗x = max x∈Cm−{⃗0} x∗B∗(A∗A)1/2(A∗A)1/2Bx x∗x = ∥(A∗A)1/2B∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' The other equality is proved similarly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 8Here we mean largest according to the actual eigenvalues, not their magnitudes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} +page_content=' 51' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/i9FRT4oBgHgl3EQfVze1/content/2301.13541v1.pdf'} diff --git a/ktAyT4oBgHgl3EQfYPen/content/tmp_files/2301.00200v1.pdf.txt b/ktAyT4oBgHgl3EQfYPen/content/tmp_files/2301.00200v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..37d3b19bcda7b071122900ea0c7e40ec08eac56f --- /dev/null +++ b/ktAyT4oBgHgl3EQfYPen/content/tmp_files/2301.00200v1.pdf.txt @@ -0,0 +1,1042 @@ +Logic Mill – A Knowledge Navigation System +Sebastian Erhardt, Mainak Ghosh, Erik Buunk, Michael E. Rose, Dietmar Harhoff +Max Planck Institute for Innovation and Competition +Munich, Germany +Email: {sebastian.erhardt, mainak.ghosh, erik.buunk, michael.rose, dietmar.harhoff}@ip.mpg.de +Abstract—Logic Mill is a scalable and openly accessible soft- +ware system that identifies semantically similar documents within +either one domain-specific corpus or multi-domain corpora. It +uses advanced Natural Language Processing (NLP) techniques +to generate numerical representations of documents. Currently +it leverages a large pre-trained language model to generate +these document representations. The system focuses on scientific +publications and patent documents and contains more than 200 +million documents. It is easily accessible via a simple Application +Programming Interface (API) or via a web interface. Moreover, +it is continuously being updated and can be extended to text +corpora from other domains. We see this system as a general- +purpose tool for future research applications in the social sciences +and other domains. +I. INTRODUCTION +There is a growing need for tools that allow researchers to +identify related documents within the same, but also across +different domains. With the ever-growing volume of scien- +tific publications and patents, scholars find it burdensome +to manage relevant documents and search important prior +contributions efficiently. Finding relevant documents plays a +significant role in building coherent scientific arguments, but +is also important in assessing the use of scientific research +outside academia [1, 2]. Patent examination is another field +in which finding related documents and identifying prior art +is essential. In ex post analyses, researchers often rely on +citation data to identify relations between documents. While +citations are helpful in tracing citation networks and in un- +covering important patterns in the production and diffusion of +knowledge within the same corpus, they are typically limited +when searching for relations across different corpora. Even +within the same corpus, citations can be selective or even +systematically biased (see [3, 4]). Finally, references may not +exist for texts that are in the process of being created, so that +authors are faced with the challenge of identifying relevant +references in the first place. Therefore, tools are needed that +allow for processing and analyzing the textual contents of +high-volume text corpora and of establishing measures of +relatedness (similarity) between them. +We refer to the system described here as a knowledge navi- +gation system, since it allows for tracing knowledge elements +within and across text corpora (e.g., the corpus of all patents +For helpful comments, we thank Matt Marx, as well as participants at the +2022 Munich Summer Institute and the 2022 Summer School on Data and +Algorithms for Science, Technology & Innovation Studies at KU Leuven. We +also thank the team at the computing and IT competence center of GWDG +for their continuous support. +or of all scientific publications). Previous attempts include +[5] for scientific publications and [6] for patent documents, +both of make use of bag-of-words approaches. While such +systems are sometimes referred to as recommender systems +[7], recommendation of related documents is only one possible +application of knowledge navigation (see below for a non- +exhaustive list of use cases). However, these systems often use +proprietary algorithms, usually focus on one domain corpus, +are not openly accessible, or are not continuously being up- +dated. A knowledge navigation system of the kind envisioned +here should be capable of efficiently retrieving, storing, and +processing hundreds of millions of documents. Moreover, it +requires capabilities for fast detection of particularly similar +documents within and across corpora. +An important problem that needs to be addressed is how to +implement the concept of document similarity. This requires +representing documents in a numerical form that computers +can process, with the goal of generating similar representations +for similar documents at large scale. In the field of natural +language processing (NLP) this process is known as document +encoding. +There are various methods for representing documents +numerically. Traditional NLP approaches like TF-IDF [8] +are used widely in the literature. However, these traditional +methods are not scalable, since an extension of the vocabulary +requires a re-computation of all representations of the corpus. +In addition, these approaches do not capture the semantics +of the documents; that is the meaning of words, or the +interpretation of sentences in context, is lost. +Therefore we propose Logic Mill, a software system aiming +to satisfy the shortcomings of existing systems and approaches. +The system creates document representations using modern +NLP techniques and contains large document sets with pre- +calculated encodings. It is easy to use and allows users to +access and compare texts from different text corpora. Further- +more, it is scalable and built on non-proprietary algorithms. We +regularly update the datasets based on their release schedule. +Our objective is to provide a fast system of high accuracy that +is openly accessible. +In the release version, the system encodes text documents +using the SPECTER document encoder [9] which leverages +state-of-the-art transformer architecture [10]. This model, in +combination with the database containing numerical represen- +tations of documents from different corpora (analogous to a +vector search database), is the backbone for the Logic Mill +system. +arXiv:2301.00200v1 [cs.CL] 31 Dec 2022 + +At present, we provide the numerical representations for +the scientific articles of Semantic Scholar and for the patent +publications issued by the United States Patent and Trademark +Office (USPTO), the European Patent Office (EPO), and the +World Intellectual Property Organization (WIPO). In addition, +users can also feed their own text data to Logic Mill for +encoding and for obtaining similarity measures, both within +their data and between own data and standard text data of +the system. They can thus link their own curated documents +to patents and scientific publications according to textual +similarity. +Logic Mill can be used in a number of different research +applications, such as: +• Explore literature: Search for research papers, and find +the best matches based on textual similarity to a paper in +the database or to own text documents. +• Prior art search in patent examination: Look for previ- +ously granted patents or (not yet granted) patent applica- +tions, but are similar to the focal one. +• Link patents to related scientific publications: Search for +patents that the scientific publication might be based on +or have a strong similarity to. +• Recommend citations and readings for new documents: +Find documents that are very similar to a focal one and +may be useful as a reference or reading. +• Assess the novelty of patents and publications: Check if +a patent or publication is new or not by comparing it +to prior texts. Documents that have few highly similar +documents may be new or even unique. +• Trace concepts across domains and over time: Identify +documents across domains (e.g. publications and patents) +that are highly similar and possibly related. +II. DOCUMENT ENCODING +Document encoding in natural language processing (NLP) +is a process for representing textual data in a numerical form. +There are various approaches to encode documents. +a) Bag of words: Simple and fast procedures construct +a vector whose binary elements indicate the presence of a +word in a document (”binary term encoding”), the number of +occurrences of a word in a document (”count matrix”) or the +weighted number of occurrences of a word in a document +(”Term frequency-inverse document frequency (TF-IDF)”). +These approaches have the drawback of not capturing the +meaning or context of the words in the document – hence +the common term bag-of-words. Furthermore, they are not +scalable since the whole model must be retrained if a new word +is added to the vocabulary and the length of each vector equals +with the vocabulary size. From a computational perspective, +they are inefficient since they generate sparse matrices where +most elements equal 0. +b) Word embeddings: Word embeddings are dense, fixed- +size, and continuous-valued vector representations of words +that capture the meaning of the words in the document. +These word representations are learned via training over large +corpora of textual data using methods such as Word2Vec +[11], GloVe [12], or FastText [13]. The advent of these word +embedding methods was a leap towards memory-efficient +dense numerical representation for words and documents from +the bag of words models’ sparse representation. A baseline +approach to representing a document is to average or sum the +learned word embeddings of the words in a document. +c) Sentence/Paragraph Embeddings: Sentence embed- +dings can be understood as an extension of the basic idea of +word embeddings. Word embeddings are static representation +of words and do not change even in the presence of multiple +contexts in the document collection. However, the same word +can have different meanings in different contexts. For example, +the word ”bank” can be a financial institute or can relate +to a river bank. Hence, the neural network architecture such +as recurrent neural networks (RNNs) and Long Short-Term +Memory (LSTM) [14] are used for representing the word +to its true context dynamically with a notion that words +appearing either before or after the focal word reveal the +context around the focal word. RNN and LSTM also aid +the vector representation of the context of the sentences or +paragraph [15]. +However, RNN and LSTM architectures cannot appropri- +ately capture the meaning of the words at the beginning of +a very long sentence or paragraph in its numerical repre- +sentation due to their sequential structure. The rise of deep +neural networks and recent advancements in the field of NLP +introduced transformer architecture [10] that looks at each +word of a sentence together, unlike RNN and LSTM, and +learns the degree to which the words reflect the context of that +sentence using the so-called attention mechanism. Therefore, +the transformer architecture can translate the meaning of each +word in a sentence through its network into the numerical +representation of the sentence, also called sentence embedding. +As stated before, a baseline approach to representing a doc- +ument is to average or sum the learned sentence embeddings +of that document. +d) BERT Language Model: Bidirectional Encoder Rep- +resentations from Transformer, BERT [16] is another recent +development that uses transformer architecture. It exhibits all +the traits of transformer architecture, meaning it learns the +sense of the words of an input sentence, the context of that +sentence, and the semantic relation between the words and the +context. BERT being a ”bidirectional” model, considers the +context of a word from both sides (left and right) at the same +time in a sentence, which makes this model effectively process +long contiguous text sequences, such as entire paragraph, not +limited to short phrase. +e) SciBERT: SciBERT is a BERT language model for +tasks involving scientific publications [17]. It was trained on +a large corpus of 1.14M scientific publications from computer +science and the broad biomedical field. This model also +outputs numerical representation like BERT. +Fine-tuning a general purpose BERT model on scientific +papers produces more accurate result in this domain [17, 18], +because it uses a domain-specific vocabulary. This also extends +to similar domains, in this case patents, where SciBERT +2 + +model outperforms the original BERT for tasks such as IPC +classification and similar patent finding [19]. SciBERT is +thus particularly well-suited for tasks such as information +extraction, document classification, and text representation in +the scientific domain. +f) SPECTER: SPECTER is an extension of SciBERT +to encode scientific publications also with the help of inter- +document relatedness [9]. In the scientific literature, citations +signal relatedness, but this information is not used by SciB- +ERT. SPECTER transfers the learned relatedness signal to the +representation of a scientific article. During the application +of this model, it generates similar embeddings for related +scientific documents without knowing citation information. +We use SPECTER model as our workhorse document en- +coder model.1 It generates embeddings for scientific publica- +tions and patents using all available information based on a +domain-specific pre-learned vocabulary, even without citation +information during the encoding process. +III. SYSTEM SPECIFICS +Logic Mill system is designed using a Microservice Archi- +tecture. This software design approach breaks down a large, +monolithic application into smaller, independent components +that can be developed, deployed, and maintained separately. +Each microservice is a self-contained unit that performs a +specific function and communicates with other microservices +through well-defined interfaces, typically using APIs. +Figure 1 shows the multiple services of our system. From +a high-level perspective, there are 6 distinct parts: +a) External Data Sources: where patent documents and +scientific publications are obtained from +b) Document Encoder: transforms the text of the document +into a numerical representation using a machine learning +model +c) Extract, Transform, Load (ETL): processes move docu- +ments from the external sources, process and store them +d) Vector Search Database: stores the computed numerical +representation along with metadata and the text +e) Backend with Web API: propagate the user requests to +the database +f) User-Interface: provides the users with a web application +where they can interact with the API via our website or +using their own scripts, for example in Python or R. +a) External Data Sources: The system can be connected +to various external data sources. The release version of the +system retrieves patent documents and scientific publications +from public and access-restricted sources. +Scientific publications are obtained from the Semantic +Scholar dataset, including abstracts and additional metadata +such as publication date, journal name, or Digital Object +Identifier (DOI) [21]. The bulk dataset of Semantic Scholar is +published as JSON files and contains more than 200 million +1There is one other BERT-based language model specifically trained on +patents, namely PatentBERT [20]. Its task is to classify patents, and therefore +not suitable for our purpose. +documents. We automatically retrieve the latest version every +month. +We obtain patent documents from multiple sources, namely +the European Patent Office (EPO), the United States Patent +and Trademark Office (USPTO), and the World Intellectual +Property Organization (WIPO). +The EPO offers various data sets and data feeds.2 We +use DocDB and the API of the European Publication Server +(EPS)3 to obtain full text and metadata. DocDB includes +bibliographic data from over 100 countries worldwide, and +for some patent authorities, the data goes back as far as the +1830s. The EPS API gives online access to all the European +patent documents published by the EPO, which Logic Mill +retrieves as XML feeds. Currently the system contains more +than 7 million full-text patent documents from the EPO EPS +and 145 million metadata records from the EPO DocDB. +We also retrieve XML files from the USPTO containing +the full text and all relevant metadata. Currently, there are +more than 10 million patent documents, which are obtained +through the USPTO Bulk Api. 4 Currently the system contains +more than 10 million full-text patent documents issued by the +USPTO. +WIPO likewise provides XML files for the full text of +international patents with metadata. Presently, there are around +3.8 million patent documents in the system, which were +obtained via the WIPO’s file server. +Logic Mill retrieves continuously the latest patent docu- +ments from the respective sources once per week and feeds +them into the system automatically. +b) Extract, Transform, Load: Through various Extract, +Transform, and Load (ETL) processes, the system obtains +the raw documents from external data sources and processes +them. In the first step, the textual content and the metadata are +extracted and stored in a global data structure. The structure +is independent of the document type (patent or scientific +publication). In the second step, the document encoder encodes +the text parts of the document and generates the numerical rep- +resentation. Finally, the search database stores the numerical +representation of a document along with the metadata and the +full text. +c) Document +Encoder: +The +document +encoder +is +machine-learning model that transforms text documents into +a numerical representation. We use SPECTER [9] to encode +documents. The output of this model is a dense vector with +768 dimensions. +Since the encoding of all documents requires significant +computing power, the encoding was conducted on desktop +workstations with Nvidia graphics processing units (GPU) and +in high-performance cloud computing facilities. To allow for +real-time inference, a CPU container was deployed in the cloud +an is connected to the system and accessible for end-users via +the API. +2See https://www.epo.org/searching-for-patents/data.html +3See https://data.epo.org/publication-server/. +4See https://developer.uspto.gov/api-catalog. +3 + +Fig. 1: Logic Mill Architecture Overview +d) Storage & Search: For the search and the storage of +documents with their numerical representation ElasticSearch is +used. This database is capable of full-text search and can be +used in a distributed context, which is essential for scalability +reasons. +Finally, ElasticSearch allows storing dense vectors that +nearest-neighbor search algorithms can use. Exact Nearest +Neighbor searchers are guaranteed to find a solution, but are +inefficient and not scalable.5 Therefore we use approximate +Nearest Neighbor searches (ANN), which trade-off precision +for lower computational and resource burden. ElasticSearch +uses the Hierarchical Navigable Small World graphs (HNSW) +[22] algorithm (as of version 8.0).6 HNSW organizes vectors +as a graph based on their similarity to each other. Together, +this setup finds the most similar documents with very high +probability for any query document within milliseconds.7 +In our current setting the cluster consists of 12 nodes with +the 8 vCPU cores, 128 GB of RAM, and 1 TB of SSD storage +each. This setting is needed to allow for fast and efficient +computation, because the RAM required by the ElasticSearch +database can be distributed over multiple nodes. The database +cluster as well as all other components are running on the +GWDG OpenStack Cloud IT infrastructure8 in G¨ottingen, +Germany. +e) Computation & API: The back-end extends the soft- +ware stack for more functionality and to handle the user +interactions. It is written in the language Go and provides +a plug-and-play Application Programming Interface (API) for +end-users using GraphQL. This query language for APIs is +5They would require to pre-compute all distance metrics between the query +vector and every vector in the database. In our setting this amounts to 220M +pairs. +6Compared to a wide spectrum of alternative ANN and according to various +distance measures, HNSW performs consistently well [23]; see also http:// +ann-benchmarks.com. +7We intend to provide further information on the time-accuracy trade-off +in future research. +8Gesellschaft f¨ur wissenschaftliche Datenverarbeiting mbH G¨ottingen, see +https://www.gwdg.de/. +a strongly typed interface that provides complete and under- +standable API documentation. Furthermore, it allows the users +to retrieve the data precisely that they have asked for. +The back-end also connects the document encoder with the +client-facing run-time environment. Doing that ensures that +end-users can send text, which can then be encoded to the +numerical representation, and finally, this representation can +be used to query similar documents from the database. +It can also be used to calculate distances between texts +provided by the user using their numerical representation +and distance metrics like the Cosine Similarity (Eq. (1)), +the Manhattan (L1) Distance (Eq. (2)) or the Euclidean (L2) +Distance (Eq. (3)). +cos(a, b) = +ab +∥a∥∥b∥ = +�n +i=1 aibi +��n +i=1 (ai)2��n +i=1 (bi)2 +(1) +l1(a, b) = ∥a − b∥ = +n +� +i=1 +|ai − bi| +(2) +l2(a, b) = ∥b − a∥2 = +� +� +� +� +n +� +i=1 +(bi − ai)2 +(3) +Furthermore, the back-end authenticates and authorizes the +end-users. +f) User Interfaces: The website https://logic-mill.net/ +features user registration, project presentation, and documen- +tation in a Single Page web Application (SPA). We built this +using the Vue JavaScript framework9. It also features a web +API for intended as the main point of interaction for registered +users. +IV. USAGE +The general idea behind the user interface is to enable a +simple and easy plug-and-play interaction that simplifies the +project setup. The machinery to use the document encoder +9See https://vuejs.org. +4 + +Extract, Transform, Load +Storage & Search +Computation & APl +External Data Sources +UserInterfaces +Backend +ETLProcesses +Vector Search +Web Application +GraphQL API +Patents +Database +User +Document Encoder +User Scripts +Scientific Publications +Machine +Learning Modelmodel is already available, and the embeddings of large +corpora (Semantic Scholar, EPO, USPTO, WIPO) are pre- +computed and ready to be used. Users can do their projects in +less time with fewer resources, since there is no need to down- +load the raw data (time, storage), set up the machine learning +pipeline (time, CPU/GPU, memory), encode the documents +(time), and search through results (time, CPU/GPU, memory, +internal storage). +a) User interfaces: Upon registration on the website, +users can access the system either through the web application +on https://logic-mill.net/ (Fig. 2) or the Application Program- +ming Interface (API). +The web application aims to help users familiarize them- +selves with the queries and data structure. It explains the +different functionalities of the system and interactively shows +the syntactically correct GraphQL queries. Thus users can +experiment, design, and adapt their queries. +The web app auto-generates these queries for various pro- +gramming languages and tools. The release version of Logic +Mill provides these examples in GraphQL for Curl, Python, +R, and Go. However, any programming language with the +ability to make HTTP requests and retrieve and process JSON +responses can interact with the API endpoint. +b) API Functionality: Logic Mill provides 9 API end- +points for retrieval, pairwise similarity metric computation, +and Nearest Neighbor search. Each functionality can be ex- +tended to multi-domain corpora (i.e., searching the most sim- +ilar scientific publications for a patent), and they can involve +available documents in the database as well as user-supplied +documents or texts. The endpoints are summarized in Table I. +TABLE I: Overview of Logic Mill’s API functionality. +Purely database +with own documents +Retrieval +Document, Documents +searchDocuments +encodeDocument, encodeDocuments +Calculation +similarityCalculation +encodeDocumentAndSimilarityCalculation +NN Search +SimilaritySearch +embedDocumentAndSimilaritySearch +A. Retrieval +a) Document Retrieval: In many cases, users can use the +system at the beginning of their research projects. One can +retrieve the pre-computed embeddings for a set of documents, +e.g., patent documents or scientific publications, that already +exist in the database via the Document and Documents +endpoints. Users would provide the IDs and corresponding +database of the documents along with an indication as to what +information they wish to access, e.g. title, abstract, claims, +description, authors, inventions, classifications, country, DOI, +journal name, and the embedding. The information the system +will return depends on the document type and the data source. +The online API documentation shows the fields that are +available in the current version. +b) Document Search: +To conduct a keyword-based +search, the searchDocuments endpoint can be used. Logic +Mill will retrieve the documents matching the given keywords +and metadata from the available corpora. It returns the same +information as a retrieval via Document and Documents. +c) Encode own documents: The embeddings for own +curated documents can be generated and retrieved via the +encodeDocument and encodeDocuments endpoints of +our API. Users provide the title and the abstract, and the +document encoder model returns the numerical representation. +1) Calculation: Users often will be interested in pairwise +similarities between documents. Although it is possible to +retrieve embeddings for a set of documents one-by-one and +compute similarity metrics, the system also provides an end- +point for doing precisely the same. +a) Calculate +Document +Similarities: +The +endpoint +similarityCalculation is used to retrieve the similar- +ity matrix of multiple documents in the database and compare +them. The input is a list of source documents and target +documents (by providing identifiers and indices) as well as +the type of distance calculation metric (cosine, l1, l2). +b) Calculate Similarities with Own Documents: To re- +trieve the similarities between a set of own curated doc- +uments and documents in the database, users use the +encodeDocumentAndSimilarityCalculation end- +point. The user provides, for instance, title and abstract of the +different documents, identifiers for later reference, and the type +of distance calculation metric (cosine, l1, l2). The system +encodes the documents with the help of the document encoder +and uses provided metric to compute the distances among the +encoded documents. The user could then transform the results +into a similarity matrix. +B. Nearest Neighbor search +a) Database Document Similarity Search: +Given a +known document, users can search for the approximate nearest +neighbor within the same or other corpora for the most +similar documents, based on the cosine similarity using the +SimilaritySearch query. For example, the five most +similar scientific publications for a specific patent can be +requested. The title and the similarity score will be returned +alongside their IDs within the respective index. +b) Own Document Similarity Search: Users can also +provide their own documents and search for similar ones +of Semantic Scholar, EPO, USPTO, WIPO, that have al- +ready been encoded and stored in the database using the +embedDocumentAndSimilaritySearch query. +V. CONCLUSION AND FUTURE DEVELOPMENTS +Logic Mill is a novel software system that helps navigating +knowledge embedded in scientific publications, patent doc- +uments and other text corpora. The system is scalable and +openly accessible. It is being updated regularly and easy to +use. We plan to expand its scope by adding more text corpora +such as the English-language Wikipedia and corpus-specific +encoders. +Users can leverage the system in the following contexts: +• retrieve numerical representations of existing documents +in Logic Mill’s database +• generate numerical representations for their own docu- +ments +5 + +Fig. 2: Logic Mill Website - Overview +• calculate similarities between users’ given documents, +or documents in the database, or between users’ given +document and the one in the database +• search for similar documents present in the database given +a query document that either exists in the database or +users can provide the query document +Researchers interested in innovation, science of science and +knowledge transfer may be particularly interested in these +capabilities. Its search capabilities may also be of interest to +patent examiners and inventors looking for prior art related to +their current inventions. Researchers in many fields may want +to use Logic Mill as a literature and citation recommender +system. +APPENDIX +Logic Mill provides an API endpoint that uses GraphQL. +The GraphQL query determines what should be executed and +what information should be returned. +Logic Mill web app provides a user-interface where dynamic +GraphQL queries are generated for cURL, Python, R and Go. +10 +Once a query is constructed in GraphQL, it can be im- +plemented and executed using any modern programming lan- +guage. +10Stata is another commonly used statistics tool, but cannot retrieve these +JSON responses. Recent versions of Stata can, however, embed Python code. +a) Basic structure of a request: A basic request writ- +ten in Python is displayed in source code .11 The variable +called query defines the GraphQL query. User specific +variables and requested information. By providing the +index and the ID (in this case an EPO patent with the number +of EP19164094B1) (line 7), the title of the document and +the vector (the numerical representation) of the encoded text +are requested. The return variables can be customized (lines +8-11) to include other information about the document. The +object called response (line 19) is a dictionary that can be +processed further. As one can see, the Python code includes +the GraphQL query directly. +b) Example code showing the structure of any interaction +with the web API in Python: +import requests +import json +ENDPOINT = ’https://lm/api/endpoint/url/here’ +query="""{ +Document(index: "epo_cos", id: "EP19164094B1") { +documentParts { +title +} +vector +} +}""" +11Our examples are written in Python using the requests package for +http requests. +6 + +MAX PLANCK +LOGICMILL +App +About +Profile +Logout + Overview +Functionality +All of the current available functionality of logic millis listed here. +Encode Document +Calculate Document Similarity +Encode Document +CalculateSimilaritybetween +DocumentSearch +DocumentRetrieval +Document Search +Documents +Document Retrieval +[[月] +Own Document Similarity Search +Database Document Similarity +Search +You have a text document and want +You want to search already encoded +You want to retrieve an already +to transform the text into a numerical +You have two or moretext +documents inour databases. +encodeddocumentfromour +representationfordownstreamtasks. +documents and want calculate the +databases. +document similarity among them. +Own Document Similarity +Database Document Similarity +Search +Search +You have a document and want to +You select a document in our +find the most similar documents in +database and search for the most +ourdatabases. +similar ones in our database.r = requests.post(ENDPOINT, +json={’query’: query}) +if r.status_code == 200: +response = r.json() +print(response) +else: +print(f"Error executing\n{query}\non {url}") +c) Parameters: While the previous example is the easiest +and most basic implementation, it is not always the most +suitable in practice. In many cases one has to do multiple +requests to retrieve all the data. For example, it is possible to +request the encoding for multiple documents in one request, +however, it is not possible to retrieve them, for example, +10,000 documents. To do this the code needs to include a loop +where, with each iteration, a query with different parameters +is executed. We will call this a parameterized query. The user +provides two parameters to interact with the web API. These +are the GraphQL query and the query parameters. Doing the +loop with parameters will make the code more readable. The +eode example 0d shows the example with the query and the +variables object. The code for looping is omitted as well as +the base code for handling the request. +d) Example code showing with a query with parameters: +# (...) +# Build GraphQL query +query=""" +query Documents($index: String!, $keyword: String!) { +Documents(index: $index, keyword: $keyword) { +id +documentParts { +title +} +vector +} +} +""" +# Build variables +variables = { +"keyword": "EP19164094B1", +"index": "epo_cos" +} +# Send request +r = requests.post(ENDPOINT, +json={’query’: query, ’variables’: variables}) +# Handle request +# (...) +Differences with the first example are: +• An extra header where the variables indicated with $- +signs and data types are defined. (line 4) These data types +can be found on the documentation website. +• The places where the variables are used in the query (line +5). +• The definition of the values of the variables (line 16). +• And the request with the additional variables object (line +22-23). +e) Other languages: Using the GraphQL queries in other +languages is very similar and the GraphQL structure stays the +same. In Code Sample 0f a query in the R-language is shown. +The ghql library is used to be able to use the GraphQL query +directly. +f) Example code showing with a query with parameters +in R: +library(jsonlite) +library(ghql) +URL <- +’https://lm/api/endpoint/url/here’ +variables <- +fromJSON(’{ +"data": { +"id": "ID", +"parts": [ +{ +"key": "title", +"value": "Airbags" +}, +{ +"key": "abstract", +"value": "Airbags are (...) crash." +} +] +} +}’) +conn <- GraphqlClient$new(url = URL) +query <- ’query encodeDocument($data: EncodeObject) { +encodeDocument(data: $data) +} +’ +new <- Query$new()$query(’link’, query) +res <- conn$exec(new$link, variables = variables) %>% +fromJSON(flatten = F) +res$data +7 + +REFERENCES +[1] M. +Marx +and +A. +Fuegi, +“Reliance +on +science: +Worldwide +front-page +patent +citations +to +scientific +articles,” +Strategic +Management +Journal, +vol. +41, +no. 9, pp. 1572–1594, Sep. 2020. [Online]. Available: +https://onlinelibrary.wiley.com/doi/10.1002/smj.3145 +[2] F. +Poege, +D. +Harhoff, +F. +Gaessler, +and +S. +Baruffaldi, +“Science +quality +and +the +value +of +inventions,” +Science +Advances, +vol. +5, +no. +12, +p. +eaay7323, +Dec. +2019. +[Online]. +Available: +https://www.science.org/doi/10.1126/sciadv.aay7323 +[3] A. B. Jaffe and G. de Rassenfosse, “Patent citation data +in social science research: Overview and best practices,” +Journal of the Association for Information Science and +Technology, vol. 68, no. 6, pp. 1360–1374, Jun. 2017. +[Online]. Available: https://onlinelibrary.wiley.com/doi/ +10.1002/asi.23731 +[4] A. Rubin and E. Rubin, “Systematic Bias in the Progress +of Research,” Journal of Political Economy, vol. 129, +no. 9, pp. 2666 – 2719, 2021. [Online]. Available: +https://doi.org/10.1086/715021 +[5] S. +L. +Woltmann +and +L. +Alkærsig, +“Tracing +university–industry +knowledge +transfer +through +a +text +mining +approach,” +Scientometrics, +vol. +117, +no. 1, pp. 449–472, Oct. 2018. [Online]. Available: +http://link.springer.com/10.1007/s11192-018-2849-9 +[6] B. Kelly, D. Papanikolaou, A. Seru, and M. Taddy, +“Measuring Technological Innovation over the Long +Run,” American Economic Review: Insights, vol. 3, +no. 3, pp. 303–320, Sep. 2021. [Online]. Available: +https://pubs.aeaweb.org/doi/10.1257/aeri.20190499 +[7] J. +Beel, +B. +Gipp, +S. +Langer, +and +C. +Breitinger, +“Research-paper +recommender +systems: +a +literature +survey,” International Journal on Digital Libraries, +vol. 17, no. 4, pp. 305–338, Nov. 2016. [Online]. +Available: https://doi.org/10.1007/s00799-015-0156-0 +[8] K. Sparck Jones, “A Statistical Interpretation of Term +Specificity and its Application in Retrieval,” Journal +of Documentation, vol. 28, no. 1, pp. 11–21, Jan. +1972, publisher: MCB UP Ltd. [Online]. Available: +https://doi.org/10.1108/eb026526 +[9] A. Cohan, S. Feldman, I. Beltagy, D. Downey, and +D. S. Weld, “SPECTER: Document-level Representation +Learning using Citation-informed Transformers,” May +2020, number: arXiv:2004.07180 arXiv:2004.07180 [cs]. +[Online]. Available: http://arxiv.org/abs/2004.07180 +[10] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, +L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, +“Attention Is All You Need,” in NIPS’17: Proceedings of +the 31st International Conference on Neural Information +Processing Systems, Dec. 2017. [Online]. Available: +https://dl.acm.org/doi/10.5555/3295222.3295349 +[11] T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient +Estimation of Word Representations in Vector Space,” +Sep. 2013, arXiv:1301.3781 [cs]. [Online]. Available: +http://arxiv.org/abs/1301.3781 +[12] J. Pennington, R. Socher, and C. Manning, “GloVe: +Global Vectors for Word Representation,” in Proceedings +of the 2014 Conference on Empirical Methods in +Natural Language Processing (EMNLP). +Doha, Qatar: +Association for Computational Linguistics, Oct. 2014, +pp. 1532–1543. [Online]. Available: https://aclanthology. +org/D14-1162 +[13] P. Bojanowski, E. Grave, A. Joulin, and T. Mikolov, +“Enriching Word Vectors with Subword Information,” +Jun. 2017, arXiv:1607.04606 [cs]. [Online]. Available: +http://arxiv.org/abs/1607.04606 +[14] S. Hochreiter and J. Schmidhuber, “Long Short-Term +Memory,” Neural Computation, vol. 9, no. 8, pp. +1735–1780, +Nov. +1997. +[Online]. +Available: +https: +//direct.mit.edu/neco/article/9/8/1735-1780/6109 +[15] O. Melamud, J. Goldberger, and I. Dagan, “context2vec: +Learning Generic Context Embedding with Bidirectional +LSTM,” in Proceedings of the 20th SIGNLL Conference +on Computational Natural Language Learning. +Berlin, +Germany: Association for Computational Linguistics, +Aug. +2016, +pp. +51–61. +[Online]. +Available: +https: +//aclanthology.org/K16-1006 +[16] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, +“BERT: Pre-training of Deep Bidirectional Transformers +for +Language +Understanding,” +in +Proceedings +of +the +2019 +Conference +of +the +North. +Minneapolis, +Minnesota: +Association +for +Computational +Linguis- +tics, +2019, +pp. +4171–4186. +[Online]. +Available: +http://aclweb.org/anthology/N19-1423 +[17] I. Beltagy, K. Lo, and A. Cohan, “SciBERT: A +Pretrained Language Model for Scientific Text,” in +Proceedings of the 2019 Conference on Empirical +Methods in Natural Language Processing and the 9th +International Joint Conference on Natural Language +Processing (EMNLP-IJCNLP). +Hong Kong, China: +Association for Computational Linguistics, 2019, pp. +3613–3618. [Online]. Available: https://www.aclweb.org/ +anthology/D19-1371 +[18] G. Aslanyan and I. Wetherbee, “Patents Phrase to Phrase +Semantic +Matching +Dataset,” +Aug. +2022, +number: +arXiv:2208.01171 +arXiv:2208.01171 +[cs]. +[Online]. +Available: http://arxiv.org/abs/2208.01171 +[19] S. +Althammer, +M. +Buckley, +S. +Hofst¨atter, +and +A. +Hanbury, +“Linguistically +Informed +Masking +for +Representation Learning in the Patent Domain,” Jun. +2021, number: arXiv:2106.05768 arXiv:2106.05768 [cs]. +[Online]. Available: http://arxiv.org/abs/2106.05768 +[20] J.-S. +Lee +and +J. +Hsiang, +“Patent +classification +by +fine-tuning +BERT +language +model,” +World +Patent +Information, +vol. +61, +p. +101965, +Jun. +2020. +[Online]. +Available: +https://linkinghub.elsevier. +com/retrieve/pii/S0172219019300742 +[21] W. Ammar, D. Groeneveld, C. Bhagavatula, I. Beltagy, +M. Crawford, D. Downey, J. Dunkelberger, A. Elgohary, +S. Feldman, V. Ha, R. Kinney, S. Kohlmeier, K. Lo, +8 + +T. Murray, H.-H. Ooi, M. Peters, J. Power, S. Skjonsberg, +L. L. Wang, C. Wilhelm, Z. Yuan, M. van Zuylen, +and O. Etzioni, “Construction of the Literature Graph +in Semantic Scholar,” in Proceedings of the 2018 +Conference of the North American Chapter of the +Association +for +Computational +Linguistics: +Human +Language Technologies, Volume 3 (Industry Papers). +New Orleans - Louisiana: Association for Computational +Linguistics, Jun. 2018, pp. 84–91. [Online]. Available: +https://aclanthology.org/N18-3011 +[22] Y. A. Malkov and D. A. Yashunin, “Efficient and +robust +approximate +nearest +neighbor +search +using +Hierarchical Navigable Small World graphs,” Aug. +2018, arXiv:1603.09320 [cs]. [Online]. Available: http: +//arxiv.org/abs/1603.09320 +[23] M. Aum¨uller, E. Bernhardsson, and A. Faithfull, “ANN- +Benchmarks: A benchmarking tool for approximate +nearest +neighbor +algorithms,” +Information +Systems, +vol. 87, p. 101374, Jan. 2020. [Online]. Available: https:// +linkinghub.elsevier.com/retrieve/pii/S0306437918303685 +9 + diff --git a/ktAyT4oBgHgl3EQfYPen/content/tmp_files/load_file.txt b/ktAyT4oBgHgl3EQfYPen/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..d077a4a273bb28578f481c61f16c97209de07edf --- /dev/null +++ b/ktAyT4oBgHgl3EQfYPen/content/tmp_files/load_file.txt @@ -0,0 +1,572 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf,len=571 +page_content='Logic Mill – A Knowledge Navigation System Sebastian Erhardt, Mainak Ghosh, Erik Buunk, Michael E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Rose, Dietmar Harhoff Max Planck Institute for Innovation and Competition Munich, Germany Email: {sebastian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='erhardt, mainak.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='ghosh, erik.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='buunk, michael.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='rose, dietmar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='harhoff}@ip.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='mpg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='de Abstract—Logic Mill is a scalable and openly accessible soft- ware system that identifies semantically similar documents within either one domain-specific corpus or multi-domain corpora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It uses advanced Natural Language Processing (NLP) techniques to generate numerical representations of documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Currently it leverages a large pre-trained language model to generate these document representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The system focuses on scientific publications and patent documents and contains more than 200 million documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It is easily accessible via a simple Application Programming Interface (API) or via a web interface.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Moreover, it is continuously being updated and can be extended to text corpora from other domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We see this system as a general- purpose tool for future research applications in the social sciences and other domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' INTRODUCTION There is a growing need for tools that allow researchers to identify related documents within the same, but also across different domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' With the ever-growing volume of scien- tific publications and patents, scholars find it burdensome to manage relevant documents and search important prior contributions efficiently.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Finding relevant documents plays a significant role in building coherent scientific arguments, but is also important in assessing the use of scientific research outside academia [1, 2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Patent examination is another field in which finding related documents and identifying prior art is essential.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In ex post analyses, researchers often rely on citation data to identify relations between documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' While citations are helpful in tracing citation networks and in un- covering important patterns in the production and diffusion of knowledge within the same corpus, they are typically limited when searching for relations across different corpora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Even within the same corpus, citations can be selective or even systematically biased (see [3, 4]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Finally, references may not exist for texts that are in the process of being created, so that authors are faced with the challenge of identifying relevant references in the first place.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Therefore, tools are needed that allow for processing and analyzing the textual contents of high-volume text corpora and of establishing measures of relatedness (similarity) between them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We refer to the system described here as a knowledge navi- gation system, since it allows for tracing knowledge elements within and across text corpora (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=', the corpus of all patents For helpful comments, we thank Matt Marx, as well as participants at the 2022 Munich Summer Institute and the 2022 Summer School on Data and Algorithms for Science, Technology & Innovation Studies at KU Leuven.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We also thank the team at the computing and IT competence center of GWDG for their continuous support.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' or of all scientific publications).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Previous attempts include [5] for scientific publications and [6] for patent documents, both of make use of bag-of-words approaches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' While such systems are sometimes referred to as recommender systems [7], recommendation of related documents is only one possible application of knowledge navigation (see below for a non- exhaustive list of use cases).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' However, these systems often use proprietary algorithms, usually focus on one domain corpus, are not openly accessible, or are not continuously being up- dated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' A knowledge navigation system of the kind envisioned here should be capable of efficiently retrieving, storing, and processing hundreds of millions of documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Moreover, it requires capabilities for fast detection of particularly similar documents within and across corpora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' An important problem that needs to be addressed is how to implement the concept of document similarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This requires representing documents in a numerical form that computers can process, with the goal of generating similar representations for similar documents at large scale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In the field of natural language processing (NLP) this process is known as document encoding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' There are various methods for representing documents numerically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Traditional NLP approaches like TF-IDF [8] are used widely in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' However, these traditional methods are not scalable, since an extension of the vocabulary requires a re-computation of all representations of the corpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In addition, these approaches do not capture the semantics of the documents;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' that is the meaning of words, or the interpretation of sentences in context, is lost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Therefore we propose Logic Mill, a software system aiming to satisfy the shortcomings of existing systems and approaches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The system creates document representations using modern NLP techniques and contains large document sets with pre- calculated encodings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It is easy to use and allows users to access and compare texts from different text corpora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Further- more, it is scalable and built on non-proprietary algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We regularly update the datasets based on their release schedule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Our objective is to provide a fast system of high accuracy that is openly accessible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In the release version, the system encodes text documents using the SPECTER document encoder [9] which leverages state-of-the-art transformer architecture [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This model, in combination with the database containing numerical represen- tations of documents from different corpora (analogous to a vector search database), is the backbone for the Logic Mill system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='00200v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='CL] 31 Dec 2022 At present, we provide the numerical representations for the scientific articles of Semantic Scholar and for the patent publications issued by the United States Patent and Trademark Office (USPTO), the European Patent Office (EPO), and the World Intellectual Property Organization (WIPO).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In addition, users can also feed their own text data to Logic Mill for encoding and for obtaining similarity measures, both within their data and between own data and standard text data of the system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' They can thus link their own curated documents to patents and scientific publications according to textual similarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Logic Mill can be used in a number of different research applications, such as: Explore literature: Search for research papers, and find the best matches based on textual similarity to a paper in the database or to own text documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Prior art search in patent examination: Look for previ- ously granted patents or (not yet granted) patent applica- tions, but are similar to the focal one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Link patents to related scientific publications: Search for patents that the scientific publication might be based on or have a strong similarity to.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Recommend citations and readings for new documents: Find documents that are very similar to a focal one and may be useful as a reference or reading.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Assess the novelty of patents and publications: Check if a patent or publication is new or not by comparing it to prior texts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Documents that have few highly similar documents may be new or even unique.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Trace concepts across domains and over time: Identify documents across domains (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' publications and patents) that are highly similar and possibly related.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' DOCUMENT ENCODING Document encoding in natural language processing (NLP) is a process for representing textual data in a numerical form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' There are various approaches to encode documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' a) Bag of words: Simple and fast procedures construct a vector whose binary elements indicate the presence of a word in a document (”binary term encoding”), the number of occurrences of a word in a document (”count matrix”) or the weighted number of occurrences of a word in a document (”Term frequency-inverse document frequency (TF-IDF)”).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' These approaches have the drawback of not capturing the meaning or context of the words in the document – hence the common term bag-of-words.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Furthermore, they are not scalable since the whole model must be retrained if a new word is added to the vocabulary and the length of each vector equals with the vocabulary size.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' From a computational perspective, they are inefficient since they generate sparse matrices where most elements equal 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) Word embeddings: Word embeddings are dense, fixed- size, and continuous-valued vector representations of words that capture the meaning of the words in the document.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' These word representations are learned via training over large corpora of textual data using methods such as Word2Vec [11], GloVe [12], or FastText [13].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The advent of these word embedding methods was a leap towards memory-efficient dense numerical representation for words and documents from the bag of words models’ sparse representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' A baseline approach to representing a document is to average or sum the learned word embeddings of the words in a document.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' c) Sentence/Paragraph Embeddings: Sentence embed- dings can be understood as an extension of the basic idea of word embeddings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Word embeddings are static representation of words and do not change even in the presence of multiple contexts in the document collection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' However, the same word can have different meanings in different contexts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' For example, the word ”bank” can be a financial institute or can relate to a river bank.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Hence, the neural network architecture such as recurrent neural networks (RNNs) and Long Short-Term Memory (LSTM) [14] are used for representing the word to its true context dynamically with a notion that words appearing either before or after the focal word reveal the context around the focal word.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' RNN and LSTM also aid the vector representation of the context of the sentences or paragraph [15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' However, RNN and LSTM architectures cannot appropri- ately capture the meaning of the words at the beginning of a very long sentence or paragraph in its numerical repre- sentation due to their sequential structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The rise of deep neural networks and recent advancements in the field of NLP introduced transformer architecture [10] that looks at each word of a sentence together, unlike RNN and LSTM, and learns the degree to which the words reflect the context of that sentence using the so-called attention mechanism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Therefore, the transformer architecture can translate the meaning of each word in a sentence through its network into the numerical representation of the sentence, also called sentence embedding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' As stated before, a baseline approach to representing a doc- ument is to average or sum the learned sentence embeddings of that document.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' d) BERT Language Model: Bidirectional Encoder Rep- resentations from Transformer, BERT [16] is another recent development that uses transformer architecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It exhibits all the traits of transformer architecture, meaning it learns the sense of the words of an input sentence, the context of that sentence, and the semantic relation between the words and the context.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' BERT being a ”bidirectional” model, considers the context of a word from both sides (left and right) at the same time in a sentence, which makes this model effectively process long contiguous text sequences, such as entire paragraph, not limited to short phrase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' e) SciBERT: SciBERT is a BERT language model for tasks involving scientific publications [17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It was trained on a large corpus of 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='14M scientific publications from computer science and the broad biomedical field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This model also outputs numerical representation like BERT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Fine-tuning a general purpose BERT model on scientific papers produces more accurate result in this domain [17, 18], because it uses a domain-specific vocabulary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This also extends to similar domains, in this case patents, where SciBERT 2 model outperforms the original BERT for tasks such as IPC classification and similar patent finding [19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' SciBERT is thus particularly well-suited for tasks such as information extraction, document classification, and text representation in the scientific domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' f) SPECTER: SPECTER is an extension of SciBERT to encode scientific publications also with the help of inter- document relatedness [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In the scientific literature, citations signal relatedness, but this information is not used by SciB- ERT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' SPECTER transfers the learned relatedness signal to the representation of a scientific article.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' During the application of this model, it generates similar embeddings for related scientific documents without knowing citation information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We use SPECTER model as our workhorse document en- coder model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1 It generates embeddings for scientific publica- tions and patents using all available information based on a domain-specific pre-learned vocabulary, even without citation information during the encoding process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' SYSTEM SPECIFICS Logic Mill system is designed using a Microservice Archi- tecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This software design approach breaks down a large, monolithic application into smaller, independent components that can be developed, deployed, and maintained separately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Each microservice is a self-contained unit that performs a specific function and communicates with other microservices through well-defined interfaces, typically using APIs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Figure 1 shows the multiple services of our system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' From a high-level perspective,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' there are 6 distinct parts: a) External Data Sources: where patent documents and scientific publications are obtained from b) Document Encoder: transforms the text of the document into a numerical representation using a machine learning model c) Extract,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Transform,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Load (ETL): processes move docu- ments from the external sources,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' process and store them d) Vector Search Database: stores the computed numerical representation along with metadata and the text e) Backend with Web API: propagate the user requests to the database f) User-Interface: provides the users with a web application where they can interact with the API via our website or using their own scripts,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' for example in Python or R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' a) External Data Sources: The system can be connected to various external data sources.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The release version of the system retrieves patent documents and scientific publications from public and access-restricted sources.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Scientific publications are obtained from the Semantic Scholar dataset, including abstracts and additional metadata such as publication date, journal name, or Digital Object Identifier (DOI) [21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The bulk dataset of Semantic Scholar is published as JSON files and contains more than 200 million 1There is one other BERT-based language model specifically trained on patents, namely PatentBERT [20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Its task is to classify patents, and therefore not suitable for our purpose.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We automatically retrieve the latest version every month.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We obtain patent documents from multiple sources, namely the European Patent Office (EPO), the United States Patent and Trademark Office (USPTO), and the World Intellectual Property Organization (WIPO).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The EPO offers various data sets and data feeds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='2 We use DocDB and the API of the European Publication Server (EPS)3 to obtain full text and metadata.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' DocDB includes bibliographic data from over 100 countries worldwide, and for some patent authorities, the data goes back as far as the 1830s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The EPS API gives online access to all the European patent documents published by the EPO, which Logic Mill retrieves as XML feeds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Currently the system contains more than 7 million full-text patent documents from the EPO EPS and 145 million metadata records from the EPO DocDB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We also retrieve XML files from the USPTO containing the full text and all relevant metadata.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Currently, there are more than 10 million patent documents, which are obtained through the USPTO Bulk Api.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 4 Currently the system contains more than 10 million full-text patent documents issued by the USPTO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' WIPO likewise provides XML files for the full text of international patents with metadata.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Presently, there are around 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='8 million patent documents in the system, which were obtained via the WIPO’s file server.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Logic Mill retrieves continuously the latest patent docu- ments from the respective sources once per week and feeds them into the system automatically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) Extract, Transform, Load: Through various Extract, Transform, and Load (ETL) processes, the system obtains the raw documents from external data sources and processes them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In the first step, the textual content and the metadata are extracted and stored in a global data structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The structure is independent of the document type (patent or scientific publication).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In the second step, the document encoder encodes the text parts of the document and generates the numerical rep- resentation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Finally, the search database stores the numerical representation of a document along with the metadata and the full text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' c) Document Encoder: The document encoder is machine-learning model that transforms text documents into a numerical representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We use SPECTER [9] to encode documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The output of this model is a dense vector with 768 dimensions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Since the encoding of all documents requires significant computing power, the encoding was conducted on desktop workstations with Nvidia graphics processing units (GPU) and in high-performance cloud computing facilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' To allow for real-time inference, a CPU container was deployed in the cloud an is connected to the system and accessible for end-users via the API.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2See https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='epo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/searching-for-patents/data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='html 3See https://data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='epo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/publication-server/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 4See https://developer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='uspto.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='gov/api-catalog.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 3 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1: Logic Mill Architecture Overview d) Storage & Search: For the search and the storage of documents with their numerical representation ElasticSearch is used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This database is capable of full-text search and can be used in a distributed context, which is essential for scalability reasons.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Finally, ElasticSearch allows storing dense vectors that nearest-neighbor search algorithms can use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Exact Nearest Neighbor searchers are guaranteed to find a solution, but are inefficient and not scalable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='5 Therefore we use approximate Nearest Neighbor searches (ANN), which trade-off precision for lower computational and resource burden.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' ElasticSearch uses the Hierarchical Navigable Small World graphs (HNSW) [22] algorithm (as of version 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='6 HNSW organizes vectors as a graph based on their similarity to each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Together, this setup finds the most similar documents with very high probability for any query document within milliseconds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='7 In our current setting the cluster consists of 12 nodes with the 8 vCPU cores, 128 GB of RAM, and 1 TB of SSD storage each.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This setting is needed to allow for fast and efficient computation, because the RAM required by the ElasticSearch database can be distributed over multiple nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The database cluster as well as all other components are running on the GWDG OpenStack Cloud IT infrastructure8 in G¨ottingen, Germany.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' e) Computation & API: The back-end extends the soft- ware stack for more functionality and to handle the user interactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It is written in the language Go and provides a plug-and-play Application Programming Interface (API) for end-users using GraphQL.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' This query language for APIs is 5They would require to pre-compute all distance metrics between the query vector and every vector in the database.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In our setting this amounts to 220M pairs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 6Compared to a wide spectrum of alternative ANN and according to various distance measures, HNSW performs consistently well [23];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' see also http:// ann-benchmarks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='com.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 7We intend to provide further information on the time-accuracy trade-off in future research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 8Gesellschaft f¨ur wissenschaftliche Datenverarbeiting mbH G¨ottingen, see https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='gwdg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='de/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' a strongly typed interface that provides complete and under- standable API documentation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Furthermore, it allows the users to retrieve the data precisely that they have asked for.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The back-end also connects the document encoder with the client-facing run-time environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Doing that ensures that end-users can send text, which can then be encoded to the numerical representation, and finally, this representation can be used to query similar documents from the database.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It can also be used to calculate distances between texts provided by the user using their numerical representation and distance metrics like the Cosine Similarity (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' (1)), the Manhattan (L1) Distance (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' (2)) or the Euclidean (L2) Distance (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' (3)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' cos(a, b) = ab ∥a∥∥b∥ = �n i=1 aibi ��n i=1 (ai)2��n i=1 (bi)2 (1) l1(a, b) = ∥a − b∥ = n � i=1 |ai − bi| (2) l2(a, b) = ∥b − a∥2 = � � � � n � i=1 (bi − ai)2 (3) Furthermore, the back-end authenticates and authorizes the end-users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' f) User Interfaces: The website https://logic-mill.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='net/ features user registration, project presentation, and documen- tation in a Single Page web Application (SPA).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We built this using the Vue JavaScript framework9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It also features a web API for intended as the main point of interaction for registered users.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' USAGE The general idea behind the user interface is to enable a simple and easy plug-and-play interaction that simplifies the project setup.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The machinery to use the document encoder 9See https://vuejs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 4 Extract, Transform, Load Storage & Search Computation & APl External Data Sources UserInterfaces Backend ETLProcesses Vector Search Web Application GraphQL API Patents Database User Document Encoder User Scripts Scientific Publications Machine Learning Modelmodel is already available, and the embeddings of large corpora (Semantic Scholar, EPO, USPTO, WIPO) are pre- computed and ready to be used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Users can do their projects in less time with fewer resources, since there is no need to down- load the raw data (time, storage), set up the machine learning pipeline (time, CPU/GPU, memory), encode the documents (time), and search through results (time, CPU/GPU, memory, internal storage).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' a) User interfaces: Upon registration on the website, users can access the system either through the web application on https://logic-mill.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='net/ (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2) or the Application Program- ming Interface (API).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The web application aims to help users familiarize them- selves with the queries and data structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It explains the different functionalities of the system and interactively shows the syntactically correct GraphQL queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Thus users can experiment, design, and adapt their queries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The web app auto-generates these queries for various pro- gramming languages and tools.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The release version of Logic Mill provides these examples in GraphQL for Curl, Python, R, and Go.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' However, any programming language with the ability to make HTTP requests and retrieve and process JSON responses can interact with the API endpoint.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) API Functionality: Logic Mill provides 9 API end- points for retrieval, pairwise similarity metric computation, and Nearest Neighbor search.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Each functionality can be ex- tended to multi-domain corpora (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=', searching the most sim- ilar scientific publications for a patent), and they can involve available documents in the database as well as user-supplied documents or texts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The endpoints are summarized in Table I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' TABLE I: Overview of Logic Mill’s API functionality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Purely database with own documents Retrieval Document, Documents searchDocuments encodeDocument, encodeDocuments Calculation similarityCalculation encodeDocumentAndSimilarityCalculation NN Search SimilaritySearch embedDocumentAndSimilaritySearch A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Retrieval a) Document Retrieval: In many cases, users can use the system at the beginning of their research projects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' One can retrieve the pre-computed embeddings for a set of documents, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=', patent documents or scientific publications, that already exist in the database via the Document and Documents endpoints.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Users would provide the IDs and corresponding database of the documents along with an indication as to what information they wish to access, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' title, abstract, claims, description, authors, inventions, classifications, country, DOI, journal name, and the embedding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The information the system will return depends on the document type and the data source.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The online API documentation shows the fields that are available in the current version.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) Document Search: To conduct a keyword-based search, the searchDocuments endpoint can be used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Logic Mill will retrieve the documents matching the given keywords and metadata from the available corpora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It returns the same information as a retrieval via Document and Documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' c) Encode own documents: The embeddings for own curated documents can be generated and retrieved via the encodeDocument and encodeDocuments endpoints of our API.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Users provide the title and the abstract, and the document encoder model returns the numerical representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1) Calculation: Users often will be interested in pairwise similarities between documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Although it is possible to retrieve embeddings for a set of documents one-by-one and compute similarity metrics, the system also provides an end- point for doing precisely the same.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' a) Calculate Document Similarities: The endpoint similarityCalculation is used to retrieve the similar- ity matrix of multiple documents in the database and compare them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The input is a list of source documents and target documents (by providing identifiers and indices) as well as the type of distance calculation metric (cosine, l1, l2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) Calculate Similarities with Own Documents: To re- trieve the similarities between a set of own curated doc- uments and documents in the database, users use the encodeDocumentAndSimilarityCalculation end- point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The user provides, for instance, title and abstract of the different documents, identifiers for later reference, and the type of distance calculation metric (cosine, l1, l2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The system encodes the documents with the help of the document encoder and uses provided metric to compute the distances among the encoded documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The user could then transform the results into a similarity matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Nearest Neighbor search a) Database Document Similarity Search: Given a known document, users can search for the approximate nearest neighbor within the same or other corpora for the most similar documents, based on the cosine similarity using the SimilaritySearch query.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' For example, the five most similar scientific publications for a specific patent can be requested.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The title and the similarity score will be returned alongside their IDs within the respective index.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) Own Document Similarity Search: Users can also provide their own documents and search for similar ones of Semantic Scholar, EPO, USPTO, WIPO, that have al- ready been encoded and stored in the database using the embedDocumentAndSimilaritySearch query.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' CONCLUSION AND FUTURE DEVELOPMENTS Logic Mill is a novel software system that helps navigating knowledge embedded in scientific publications, patent doc- uments and other text corpora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The system is scalable and openly accessible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' It is being updated regularly and easy to use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We plan to expand its scope by adding more text corpora such as the English-language Wikipedia and corpus-specific encoders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Users can leverage the system in the following contexts: retrieve numerical representations of existing documents in Logic Mill’s database generate numerical representations for their own docu- ments 5 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2: Logic Mill Website - Overview calculate similarities between users’ given documents, or documents in the database, or between users’ given document and the one in the database search for similar documents present in the database given a query document that either exists in the database or users can provide the query document Researchers interested in innovation, science of science and knowledge transfer may be particularly interested in these capabilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Its search capabilities may also be of interest to patent examiners and inventors looking for prior art related to their current inventions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Researchers in many fields may want to use Logic Mill as a literature and citation recommender system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' APPENDIX Logic Mill provides an API endpoint that uses GraphQL.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The GraphQL query determines what should be executed and what information should be returned.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Logic Mill web app provides a user-interface where dynamic GraphQL queries are generated for cURL, Python, R and Go.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 10 Once a query is constructed in GraphQL, it can be im- plemented and executed using any modern programming lan- guage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 10Stata is another commonly used statistics tool, but cannot retrieve these JSON responses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Recent versions of Stata can, however, embed Python code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' a) Basic structure of a request: A basic request writ- ten in Python is displayed in source code .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='11 The variable called query defines the GraphQL query.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' User specific variables and requested information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' By providing the index and the ID (in this case an EPO patent with the number of EP19164094B1) (line 7), the title of the document and the vector (the numerical representation) of the encoded text are requested.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The return variables can be customized (lines 8-11) to include other information about the document.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The object called response (line 19) is a dictionary that can be processed further.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' As one can see, the Python code includes the GraphQL query directly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' b) Example code showing the structure of any interaction with the web API in Python: import requests import json ENDPOINT = ’https://lm/api/endpoint/url/here’ query="""{ Document(index: "epo_cos", id: "EP19164094B1") { documentParts { title } vector } }""" 11Our examples are written in Python using the requests package for http requests.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 6 MAX PLANCK LOGICMILL App About Profile Logout Overview Functionality All of the current available functionality of logic millis listed here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Encode Document Calculate Document Similarity Encode Document CalculateSimilaritybetween DocumentSearch DocumentRetrieval Document Search Documents Document Retrieval [[月] Own Document Similarity Search Database Document Similarity Search You have a text document and want You want to search already encoded You want to retrieve an already to transform the text into a numerical You have two or moretext documents inour databases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' encodeddocumentfromour representationfordownstreamtasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' documents and want calculate the databases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' document similarity among them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Own Document Similarity Database Document Similarity Search Search You have a document and want to You select a document in our find the most similar documents in database and search for the most ourdatabases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' similar ones in our database.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='r = requests.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='post(ENDPOINT, json={’query’: query}) if r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='status_code == 200: response = r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='json() print(response) else: print(f"Error executing\\n{query}\\non {url}") c) Parameters: While the previous example is the easiest and most basic implementation, it is not always the most suitable in practice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In many cases one has to do multiple requests to retrieve all the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' For example, it is possible to request the encoding for multiple documents in one request, however, it is not possible to retrieve them, for example, 10,000 documents.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' To do this the code needs to include a loop where, with each iteration, a query with different parameters is executed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' We will call this a parameterized query.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The user provides two parameters to interact with the web API.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' These are the GraphQL query and the query parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Doing the loop with parameters will make the code more readable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The eode example 0d shows the example with the query and the variables object.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The code for looping is omitted as well as the base code for handling the request.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' d) Example code showing with a query with parameters: # (.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=') # Build GraphQL query query=""" query Documents($index: String!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=', $keyword: String!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=') { Documents(index: $index, keyword: $keyword) { id documentParts { title } vector } } """ # Build variables variables = { "keyword": "EP19164094B1", "index": "epo_cos" } # Send request r = requests.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='post(ENDPOINT, json={’query’: query, ’variables’: variables}) # Handle request # (.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=') Differences with the first example are: An extra header where the variables indicated with $- signs and data types are defined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' (line 4) These data types can be found on the documentation website.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The places where the variables are used in the query (line 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The definition of the values of the variables (line 16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' And the request with the additional variables object (line 22-23).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' e) Other languages: Using the GraphQL queries in other languages is very similar and the GraphQL structure stays the same.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' In Code Sample 0f a query in the R-language is shown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' The ghql library is used to be able to use the GraphQL query directly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' f) Example code showing with a query with parameters in R: library(jsonlite) library(ghql) URL <- ’https://lm/api/endpoint/url/here’ variables <- fromJSON(’{ "data": { "id": "ID", "parts": [ { "key": "title", "value": "Airbags" }, { "key": "abstract", "value": "Airbags are (.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=') crash.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='" } ] } }’) conn <- GraphqlClient$new(url = URL) query <- ’query encodeDocument($data: EncodeObject) { encodeDocument(data: $data) } ’ new <- Query$new()$query(’link’, query) res <- conn$exec(new$link, variables = variables) %>% fromJSON(flatten = F) res$data 7 REFERENCES [1] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Marx and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Fuegi, “Reliance on science: Worldwide front-page patent citations to scientific articles,” Strategic Management Journal, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 41, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 9, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1572–1594, Sep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://onlinelibrary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='wiley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='com/doi/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1002/smj.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='3145 [2] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Poege, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Harhoff, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Gaessler, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Baruffaldi, “Science quality and the value of inventions,” Science Advances, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 5, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 12, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' eaay7323, Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='science.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/doi/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1126/sciadv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='aay7323 [3] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Jaffe and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' de Rassenfosse, “Patent citation data in social science research: Overview and best practices,” Journal of the Association for Information Science and Technology, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 68, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 6, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1360–1374, Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://onlinelibrary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='wiley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='com/doi/ 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1002/asi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='23731 [4] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Rubin and E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Rubin, “Systematic Bias in the Progress of Research,” Journal of Political Economy, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 129, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 9, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2666 – 2719, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1086/715021 [5] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Woltmann and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Alkærsig, “Tracing university–industry knowledge transfer through a text mining approach,” Scientometrics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 117, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 449–472, Oct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://link.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='springer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='com/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1007/s11192-018-2849-9 [6] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Kelly, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Papanikolaou, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Seru, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Taddy, “Measuring Technological Innovation over the Long Run,” American Economic Review: Insights, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 3, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 303–320, Sep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://pubs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='aeaweb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/doi/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1257/aeri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='20190499 [7] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Beel, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Gipp, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Langer, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Breitinger, “Research-paper recommender systems: a literature survey,” International Journal on Digital Libraries, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 17, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 305–338, Nov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1007/s00799-015-0156-0 [8] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Sparck Jones, “A Statistical Interpretation of Term Specificity and its Application in Retrieval,” Journal of Documentation, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 28, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 11–21, Jan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1972, publisher: MCB UP Ltd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='1108/eb026526 [9] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Cohan, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Feldman, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Beltagy, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Downey, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Weld, “SPECTER: Document-level Representation Learning using Citation-informed Transformers,” May 2020, number: arXiv:2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='07180 arXiv:2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='07180 [cs].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/abs/2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='07180 [10] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Vaswani, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Shazeer, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Parmar, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Uszkoreit, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Jones, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Gomez, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Kaiser, and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Polosukhin, “Attention Is All You Need,” in NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Systems, Dec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://dl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='acm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/doi/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='5555/3295222.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='3295349 [11] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Mikolov, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Chen, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Corrado, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Dean, “Efficient Estimation of Word Representations in Vector Space,” Sep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2013, arXiv:1301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='3781 [cs].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/abs/1301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='3781 [12] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Pennington, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Socher, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Manning, “GloVe: Global Vectors for Word Representation,” in Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Doha, Qatar: Association for Computational Linguistics, Oct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2014, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1532–1543.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://aclanthology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' org/D14-1162 [13] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Bojanowski, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Grave, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Joulin, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Mikolov, “Enriching Word Vectors with Subword Information,” Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2017, arXiv:1607.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='04606 [cs].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/abs/1607.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='04606 [14] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Hochreiter and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 9, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 8, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1735–1780, Nov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 1997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https: //direct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='mit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='edu/neco/article/9/8/1735-1780/6109 [15] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Melamud, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Goldberger, and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Dagan, “context2vec: Learning Generic Context Embedding with Bidirectional LSTM,” in Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Berlin, Germany: Association for Computational Linguistics, Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2016, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 51–61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https: //aclanthology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/K16-1006 [16] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Devlin, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='-W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Chang, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Lee, and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in Proceedings of the 2019 Conference of the North.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Minneapolis, Minnesota: Association for Computational Linguis- tics, 2019, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 4171–4186.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://aclweb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/anthology/N19-1423 [17] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Beltagy, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Lo, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Cohan, “SciBERT: A Pretrained Language Model for Scientific Text,” in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Hong Kong, China: Association for Computational Linguistics, 2019, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 3613–3618.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='aclweb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/ anthology/D19-1371 [18] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Aslanyan and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Wetherbee, “Patents Phrase to Phrase Semantic Matching Dataset,” Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2022, number: arXiv:2208.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='01171 arXiv:2208.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='01171 [cs].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/abs/2208.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='01171 [19] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Althammer, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Buckley, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Hofst¨atter, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Hanbury, “Linguistically Informed Masking for Representation Learning in the Patent Domain,” Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2021, number: arXiv:2106.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='05768 arXiv:2106.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='05768 [cs].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/abs/2106.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='05768 [20] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='-S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Lee and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Hsiang, “Patent classification by fine-tuning BERT language model,” World Patent Information, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 61, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 101965, Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://linkinghub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='elsevier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' com/retrieve/pii/S0172219019300742 [21] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Ammar, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Groeneveld, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Bhagavatula, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Beltagy, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Crawford, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Downey, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Dunkelberger, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Elgohary, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Feldman, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Ha, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Kinney, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Kohlmeier, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Lo, 8 T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Murray, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Ooi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Peters, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Power, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Skjonsberg, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Wang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Wilhelm, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Yuan, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' van Zuylen, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Etzioni, “Construction of the Literature Graph in Semantic Scholar,” in Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 3 (Industry Papers).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' New Orleans - Louisiana: Association for Computational Linguistics, Jun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2018, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 84–91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https://aclanthology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/N18-3011 [22] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Malkov and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Yashunin, “Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs,” Aug.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2018, arXiv:1603.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='09320 [cs].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: http: //arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='org/abs/1603.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='09320 [23] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Aum¨uller, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Bernhardsson, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Faithfull, “ANN- Benchmarks: A benchmarking tool for approximate nearest neighbor algorithms,” Information Systems, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 87, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 101374, Jan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' [Online].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content=' Available: https:// linkinghub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='elsevier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} +page_content='com/retrieve/pii/S0306437918303685 9' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ktAyT4oBgHgl3EQfYPen/content/2301.00200v1.pdf'} diff --git a/ktAzT4oBgHgl3EQfpv1w/content/2301.01617v1.pdf b/ktAzT4oBgHgl3EQfpv1w/content/2301.01617v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9122ab06cdb698302d53f4ed300c181314db5ee1 --- /dev/null +++ b/ktAzT4oBgHgl3EQfpv1w/content/2301.01617v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42b4c95eeb87b0c8562c035f82626e92cff5114671fa9ce858c67cf1c0a631c4 +size 1357315 diff --git a/mtAyT4oBgHgl3EQfYvcj/content/tmp_files/2301.00208v1.pdf.txt b/mtAyT4oBgHgl3EQfYvcj/content/tmp_files/2301.00208v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..65d84b0e40725ef94c5729152d994831770ab23a --- /dev/null +++ b/mtAyT4oBgHgl3EQfYvcj/content/tmp_files/2301.00208v1.pdf.txt @@ -0,0 +1,362 @@ +arXiv:2301.00208v1 [math.OA] 31 Dec 2022 +INVERTIBILITY PRESERVING MAPPINGS ONTO +FINITE C*-ALGEBRAS +MARTIN MATHIEU AND FRANCOIS SCHULZ +Abstract. We prove that every surjective unital linear mapping which preserves +invertible elements from a Banach algebra onto a C*-algebra carrying a faithful tracial +state is a Jordan homomorphism thus generalising Aupetit’s 1998 result for finite von +Neumann algebras. +1. Introduction +A linear mapping T between two unital, complex Banach algebras is said to be spectrum- +preserving if, for every element a in the domain algebra, its spectrum σ(a) coincides with +σ(Ta). Provided the codomain is semisimple and T is surjective, T must be bounded (a +result belonging to Aupetit [1, Theorem 5.5.2]). Provided the domain is semisimple too, +T is injective; this follows from Zem´anek’s characterisation of the radical ([1, Theorem +5.3.1]) as +σ(a + x) = σ(Ta + Tx) = σ(Tx) = σ(x) +for each a such that Ta = 0 and every x +which implies that a belongs to the radical, which is zero in the semisimple case. More- +over, T1 = 1, that is, T is unital. As a result, a surjective spectrum-preserving mapping +between semisimple Banach algebras is a topological isomorphism and one naturally +wonders if it is also an isomorphism of (some of) the algebraic structure. +A Jordan homomorphism is a linear mapping T with the property T(a2) = (Ta)2 for +all a in the domain (which is equivalent to T(ab + ba) = TaTb + TbTa for all a and b). +A Jordan isomorphism turns out to be spectrum-preserving, and a lot of work has been +invested to explore to what extent the reverse implication holds. A pleasant survey on +the history of this topic is contained in [2]; see also [8], [13] and [14] for related questions. +In [3], Aupetit proved that every surjective spectrum-preserving linear mapping be- +tween von Neumann algebras is a Jordan isomorphism. It is not difficult to see that it +suffices that one of the algebras is a unital C*-algebra of real rank zero and the other +a unital semisimple Banach algebra. However, the problem remains open for general +C*-algebras. It is also known that the assumption on T can be relaxed to a surjec- +tive unital invertibility-preserving linear mapping (that is, σ(Ta) ⊆ σ(a) for all a); the +conclusion is then that T is a Jordan homomorphism. +2020 Mathematics Subject Classification. 47B48, 47A10, 46L05, 46L30, 16W10, 17C65. +Key words and phrases. C*-algebras, tracial states, Jordan homomorphisms, invertibility preserving +mappings. +1 + +2 +MARTIN MATHIEU AND FRANCOIS SCHULZ +In an earlier paper [2], Aupetit had already obtained the same result for finite von +Neumann algebras. The main tool in that result was the Fuglede–Kadison determinant +∆; see [9, pp. 105] for its definition and properties. Its relation to the finite trace τ +is given by ∆(a) = exp(τ(log |a|)), for every invertible element a. In our approach we +bypass the determinant and work exclusively with a (faithful) tracial state instead in +order to obtain the following generalisation. +Theorem. Let B be a unital complex Banach algebra and let A be a unital finite C*- +algebra. Let T : B → A be a surjective unital linear mapping which preserves invertible +elements. Then T is a Jordan homomorphism. +We largely follow Aupetit’s arguments but, to emphasise the differences, we split up +the proof into a series of lemmas in the next section. +2. Preliminaries +Let A be a unital C*-algebra. We say that A is finite if it comes equipped with a faithful +tracial state, that is, a linear functional τ such that τ(1) = 1 = ∥1∥, τ(ab) = τ(ba) for +all a, b ∈ A and τ(a∗a) = 0 implies a = 0. Such a functional is necessarily positive and +bounded. +We denote the set of all states of A (positive linear functionals of norm 1) by S and +by Sp the subset of all spectral states f of A, that is, f ∈ S and |f(x)| ≤ ρ(x) for every +x ∈ A, where ρ(x) denotes the spectral radius of x. It is known ([7, Theorem 4 in §13]) +that every f ∈ Sp has the trace property, that is, f(ab) = f(ba) for all a, b ∈ A, and +that f(a) ∈ co σ(a), the convex hull of the spectrum σ(a) of a, for each a ∈ A; see, [7, +Lemma 2 in §13] or [1, Lemma 4.1.15]. +Conversely, every tracial state τ belongs to Sp as follows from the subsequent argu- +ment. For a ∈ A, denote by V (a) = {f(a) | f ∈ S} its (algebra) numerical range [7]. As +is shown in [5, Lemma], and attributed to [11, §2], co σ(a) = � +b∈G(A) V (bab−1), where +G(A) stands for the group of invertible elements in A. Clearly, τ(a) belongs to the right +hand side of the above identity and hence, τ ∈ Sp. (Compare also [12].) +Lemma 2.1. Let A be a unital C*-algebra with faithful tracial state τ. Suppose that +g : C → A is an entire function with values in G(A). Then the mapping gτ : C → R, +gτ(λ) = τ(log(|g(λ)|) is harmonic. +Proof. The argument in the proof of Th´eor`eme 1.11 in [2], which is already entirely +formulated in terms of the trace, takes over verbatim. +□ +In the following, T will denote a surjective unital linear mapping defined on a (com- +plex, unital) Banach algebra B with values in a finite unital C*-algebra A. We will +assume that T preserves invertible elements so that T G(B) ⊆ G(A). It follows from [1, +Theorem 5.5.2] that T is bounded. +Fix a, b ∈ B and define +g : C × C −→ G(A), +g(λ, µ) = T(eλaeµb)e−λTae−µTb. + +INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS +3 +Then g is a separately entire function. Its series expansion reads as follows +g(λ, µ) = 1 + λ2 +2 +� +T(a2) − (Ta)2� ++ µ2 +2 +� +T(b2) − (Tb)2� ++ λµ +� +T(ab) − TbTa +� ++ λ3 +6 +� +T(a3) + 2(Ta)3 − 3T(a2)Ta +� ++ λ2µ +2 +� +T(a2b) + (Ta)2Tb + Tb(Ta)2 − T(a2)Tb − 2T(ab)Ta +� ++ λµ2 +2 +� +T(ab2) + 2TbTaTb − 2T(ab)Tb − T(b2)Ta +� ++ µ3 +6 +� +T(b3) + 2(Tb)3 − 3T(b2)Tb +� ++ remainder +(2.1) +where the remainder only contains terms of degree 4 or higher in λ and µ; we will put +it to good use in the proof of the main theorem. +By Lemma 2.1, the function +gτ : C × C −→ R, +gτ(λ, µ) = τ(log(|g(λ, µ)|)) +is separately harmonic in λ and µ and thus there exists a separately entire function +h(λ, µ) such that Re h(λ, µ) = gτ(λ, µ) for all λ, µ ∈ C. +The next step will be to establish the following three lemmas; for their proofs, see +Section 3. +Lemma 2.2. For all λ, µ ∈ C, we have egτ(λ,µ) ≤ ∥g(λ, µ)∥. +Lemma 2.3. With the above notation and caveats, let g∗(λ, µ) stand for (g(λ, µ))∗. +Then there exists r > 0 such that, for all λ, µ ∈ C with |λ|, |µ| < r, we have +2 Re h(λ, µ) = τ +� +log(g∗(λ, µ)g(λ, µ) +� += − +∞ +� +k=1 +1 +kτ +� +(1 − g∗(λ, µ)g(λ, µ))k� +. +(2.2) +Lemma 2.4. For all λ, µ in a neighbourhood of zero, τ +� +log(g∗(λ, µ)g(λ, µ) +� += 0. +3. Proofs of the Lemmas and the Main Theorem +The argument of the first lemma differs from [2] in that we cannot make use of the +determinant in order to locate the appropriate values in the convex hull of the spectrum. +Proof of Lemma 2.2. As we observed above, τ(log(|g(λ, µ)|)) ∈ co σ(log(|g(λ, µ)|)) for +all λ, µ ∈ C. Since the spectrum of log(|g(λ, µ)|) is contained in R, it follows that +co σ(log(|g(λ, µ)|)) = [s, t] for some s, t ∈ σ(log(|g(λ, µ)|)) with s ≤ t. Since the expo- +nential function is strictly increasing, the Spectral Mapping Theorem implies that +egτ (λ,µ) ∈ [es, et] = co σ(elog(|g(λ,µ)|)) = co σ(|g(λ, µ)|). + +4 +MARTIN MATHIEU AND FRANCOIS SCHULZ +As a result, 0 < egτ (λ,µ) ≤ ρ(|g(λ, µ)|) and therefore, +e2 gτ(λ,µ) ≤ ρ(|g(λ, µ)|)2 = ρ(|g(λ, µ)|2) += ρ(g∗(λ, µ)g(λ, µ)) = ∥g∗(λ, µ)g(λ, µ)∥ += ∥g(λ, µ)∥2 +as claimed. +□ +The next argument is rather straightforward. +Proof of Lemma 2.3. As g(0, 0) = T1 = 1, by continuity, there is r > 0 such that, for +all λ, µ with |λ|, |µ| < r, we have ∥1 − g∗(λ, µ)g(λ, µ)∥ < 1. The series expansion of the +logarithm thus yields +− log(g∗(λ, µ)g(λ, µ)) = +∞ +� +k=1 +1 +k(1 − g∗(λ, µ)g(λ, µ))k. +(3.1) +The definition of gτ entails that +2 Re h(λ, µ) = 2 τ(log(|g(λ, µ)|)) = τ(log(|g(λ, µ)|2)) = τ +� +log(g∗(λ, µ)g(λ, µ) +� +. +(3.2) +Combining these two identities gives the claim. +□ +The proof of the third lemma follows exactly Aupetit’s arguments. (There appears +to be some misprint at the bottom of page 61 and top of page 62 of [2].) +Proof of Lemma 2.4. For all λ, µ ∈ C, we have +��eh(λ,µ)�� = eRe h(λ,µ) = egτ (λ,µ) ≤ ∥g(λ, µ)∥ +by Lemma 2.2. Since +∥g(λ, µ)∥ ≤ ∥T∥ e|λ|(∥a∥+∥Ta∥)+|µ|(∥b∥+∥Tb∥) +it follows that eh(λ,µ) = eαλ+βµ+γ for suitable α, β, γ ∈ C ([4, Lemma 3.2]). As gτ(0, 0) = +0 we have |eγ| = 1, thus we may assume that γ = 0 (since we need only the real part +of γ). Therefore, 2 Re h(λ, µ) = αλ + βµ + ¯α¯λ + ¯β¯µ. From Lemma 2.3 we obtain +αλ + βµ + ¯α¯λ + ¯β¯µ = − +∞ +� +k=1 +1 +kτ +� +(1 − g∗(λ, µ)g(λ, µ))k� +for all λ, µ such that |λ|, |µ| < r for suitable r > 0. The series expansion of g(λ, µ) in +(2.1) does not contain any powers of λ or µ of first order, hence the series expansion +in (3.1) cannot either. This entails that both α and β are equal to zero. +It now follows from (2.2) that τ +� +log(g∗(λ, µ)g(λ, µ) +� += 0. +□ +We now have all the tools to prove our main theorem by adapting the arguments in +Theorem 1.12 of [2] to our situation. + +INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS +5 +Proof of the Theorem. Set f(λ, µ) = �∞ +k=1 +1 +kτ +� +(1 − g∗(λ, µ)g(λ, µ))k� +for all λ, µ such +that |λ|, |µ| < r for suitable r > 0 (given by Lemma 2.3). By Lemma 2.4, f = 0 and +thus +∂2 +∂λ∂µf(0, 0) = 0 = +∂3 +∂λ2∂µf(0, 0). +Using these identities after substituting in the series expansion (2.1) into the log-series +we find +τ +� +T(ab) − TaTb +� += 0 +(3.3) +and +τ +� +T(a2b) + (Ta)2Tb + Tb(Ta)2 − T(a2)Tb − 2 T(ab)Ta +� += 0 +(3.4) +for all a, b ∈ B. From (3.3) we obtain +τ(T(a2b)) = τ(T(a2)Tb) +and +τ(T(a2b)) = τ(T(a(ab))) = τ(TaT(ab)) +so that (3.4) reduces to +τ((Ta)2Tb) = τ(TaT(ab)), +using the trace property. It follows that +τ((Ta)2Tb) = τ(T(a2)Tb). +Since T is surjective we may choose b ∈ B such that Tb = ((Ta)2 − T(a2))∗ wherefore +the last identity yields, for each a ∈ B, that +τ +� +((Ta)2 − T(a2))((Ta)2 − T(a2))∗� += 0. +The faithfulness of τ implies that T is a Jordan homomorphism. +□ +4. Conclusions +In this section, we collect together some consequences and sharpening of our main +theorem. We also relate it to open problems of a similar nature. +Suppose T is a surjective linear mapping between two semisimple unital Banach +algebras which preserves the spectrum of each element. Then T is injective (as explained +in the Introduction) and T1 = 1. The latter follows, for example, from +σ((T1 − 1) + Tx) = σ(T1 + Tx) − 1 = σ(1 + x) − 1 = σ(x) = σ(Tx) +and the surjectivity of T which entails that σ((T1 − 1) + y) = σ(y) for all y in the +codomain. Thus, by Zem´anek’s characterisation of the radical, T1 − 1 = 0. +As a result, we have a symmetric situation and can apply the Theorem to either T +or its inverse to obtain the following consequence. +Corollary 4.1. Let T be a surjective spectrum-preserving linear mapping between two +semisimple unital Banach algebras. If either of them is a unital C*-algebra equipped +with a faithful tracial state then T is a Jordan isomorphism. + +6 +MARTIN MATHIEU AND FRANCOIS SCHULZ +This is another contribution to a longstanding, still open problem by Kaplansky who +asked in 1970 whether the above statement holds without any further assumptions on +the Banach algebras. For further references, see [2] and [10]. +All the steps in the proof of the Theorem but the very last one can be performed for +each individual tracial state on a unital C*-algebra. Therefore, the assumption can be +relaxed to the existence of a faithful family of tracial states, that is, a family {τi | i ∈ I} +of tracial states τi such that τi(a∗a) = 0 for all i ∈ I implies a = 0. +In particular, since any tracial state on a simple unital C*-algebra is faithful we obtain +the following result. +Corollary 4.2. Let T : B → A be a surjective unital invertibility-preserving linear +mapping into a simple unital C*-algebra A which carries a tracial state. Then T is a +Jordan homomorphism. +Remark 4.3. Our terminology of a “finite”C*-algebra is not quite standard. In [6, +III.1.3.1], a unital C*-algebra A is called finite if the identity of A is a finite projection; +that is, there is no proper subprojection which is Murray–von Neumann equivalent +to 1. Every unital C*-algebra with a faithful tracial state is finite in this sense but the +converse fails in general (though it holds for stably finite exact C*-algebras). We prefer +here a definition that does not make reference to any projections. +We can also strengthen our main theorem in a different direction. Let G1(B) denote +the principal component of G(B), where B is a unital Banach algebra. It is known, see, +e.g., [1, Theorem 3.3.7], that G1(B) = {ex1 · · · exn | xi ∈ B, n ∈ N}. The associated +exponential spectrum of x ∈ B is +σε(x) = {λ ∈ C | λ − x /∈ G1(B)}. +In certain situations it is more natural and expedient to consider the exponential spec- +trum instead of the smaller spectrum, see, e.g., [1, Theorem 3.3.8]. From the proof of +our main result we see that it suffices that the mapping T sends the product of any two +exponentials in B onto an invertible element in A. This gives the following corollary. +Corollary 4.4. Let B be a unital complex Banach algebra and let A be a unital finite C*- +algebra. Let T : B → A be a surjective unital linear mapping such that TG1(B) ⊆ G(A). +Then T is a Jordan homomorphism. +A spectral isometry between two Banach algebras A and B is a linear mapping S +such that ρ(Sx) = ρ(x) for all x ∈ A. Clearly, every spectrum-preserving mapping is +a spectral isometry and so is every Jordan isomorphism. A conjecture related to Ka- +plansky’s problem mentioned above states that every unital surjective spectral isometry +between two C*-algebras is a Jordan isomorphism. This conjecture has been confirmed +in many cases, see, e.g., [13, 14], but is open in all generality. Notably it was verified in +[15] if A is a unital C*-algebra of real rank zero and without tracial states. The above +Corollary 4.1 is thus a step forward in the direction of confirming the general conjecture. + +INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS +7 +Acknowledgements. The research for this paper was completed while the second- +named author was visiting Queen’s University Belfast. He would like to thank both +Professor Martin Mathieu and the Mathematical Sciences Research Centre at Queen’s +University Belfast for their hospitality, and the National Research Foundation of South +Africa for their financial support (NRF Grant Number: 129692). +References +[1] B. Aupetit, A Primer on Spectral Theory, Springer-Verlag, New York, 1991. +[2] B. Aupetit, Sur les transformations qui conservent le spectre, in: Banach Algebras ’97, Proc. 13th +Int. Conf. on Banach Algebras, Blaubeuren, July 20–August 3, 1997, M. Mathieu and E. Albrecht +(eds.), de Gruyter, Berlin, 1998, pp. 55–78. +[3] B. Aupetit, Spectrum-preserving linear mappings between Banach algebras or Jordan–Banach al- +gebras, J. London Math. Soc. 62 (2000), 917–924. +[4] B. Aupetit and H. du T. Mouton, Trace and determinant in Banach algebras, Studia Math. 121 +(1996), 115–136. +[5] S. K. Berberian, Trace and the convex hull of the spectrum in a von Neumann algebra of finite +class, Proc. Amer. Math. Soc. 23 (1969), 211–212. +[6] B. Blackadar, Operator Algebras. Theory of C*-Algebras and von Neumann Algebras, Encyclopae- +dia of Math. Sciences 122, Springer-Verlag, Berlin, 2006. +[7] F. F. Bonsall and J. Duncan, Numerical Ranges of Operators and of Elements of Normed Algebras, +London Math. Soc. Lecture Notes 2, Cambridge Univ. Press, Cambridge, 1971. +[8] R. Brits, F. Schulz and C. Tour´e, A spectral characterization of isomorphisms on C*-algebras, +Arch. Math. 113 (2019), 391–398. +[9] J. Dixmier, Les alg`ebres d’op´erateurs dans l’espace Hilbertien, Gauthier-Villars, Paris, 1957. +[10] L. W. Harris, Invertibility preserving linear maps of Banach algebras, in: Complex Analysis and +Dynamical Systems, Contemp. Math. 364, Israel Math. Conf. Proc., Amer. Math. Soc., Provi- +dence, RI, 2004; pp. 59–66. +[11] S. Hildebrandt, ¨Uber den numerischen Wertebereich eines Operators, Math. Ann. 163 (1966), +230–247. +[12] M. Mathieu, Spectrally bounded traces on C*-algebras, Bull. Austral. Math. Soc. 68 (2003), 169– +173. +[13] M. Mathieu, A collection of problems on spectrally bounded operators, Asian-Eur. J. Math. 2 +(2009), 487–501. +[14] M. Mathieu, Interplay between spectrally bounded operators and complex analysis, Irish Math. Soc. +Bull. 72 (2013), 1–13. +[15] M. Mathieu, Characterizing Jordan homomorphisms, Acta Sci. Math. (Szeged) 86 (2020), 697–701. +Mathematical Sciences Research Centre, Queen’s University Belfast, Belfast BT7 +1NN, Northern Ireland +Email address: m.m@qub.ac.uk +Department of Mathematics and Applied Mathematics, Faculty of Science, Univer- +sity of Johannesburg, P.O. Box 524, Auckland Park, 2006, South Africa +Email address: francoiss@uj.ac.za + diff --git a/mtAyT4oBgHgl3EQfYvcj/content/tmp_files/load_file.txt b/mtAyT4oBgHgl3EQfYvcj/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..c31a3672119dc6f0f6b9ae0e87f9e475fd6a7c68 --- /dev/null +++ b/mtAyT4oBgHgl3EQfYvcj/content/tmp_files/load_file.txt @@ -0,0 +1,294 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf,len=293 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='00208v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='OA] 31 Dec 2022 INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS MARTIN MATHIEU AND FRANCOIS SCHULZ Abstract.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We prove that every surjective unital linear mapping which preserves invertible elements from a Banach algebra onto a C*-algebra carrying a faithful tracial state is a Jordan homomorphism thus generalising Aupetit’s 1998 result for finite von Neumann algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Introduction A linear mapping T between two unital, complex Banach algebras is said to be spectrum- preserving if, for every element a in the domain algebra, its spectrum σ(a) coincides with σ(Ta).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Provided the codomain is semisimple and T is surjective, T must be bounded (a result belonging to Aupetit [1, Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Provided the domain is semisimple too, T is injective;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' this follows from Zem´anek’s characterisation of the radical ([1, Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1]) as σ(a + x) = σ(Ta + Tx) = σ(Tx) = σ(x) for each a such that Ta = 0 and every x which implies that a belongs to the radical, which is zero in the semisimple case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' More- over, T1 = 1, that is, T is unital.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' As a result, a surjective spectrum-preserving mapping between semisimple Banach algebras is a topological isomorphism and one naturally wonders if it is also an isomorphism of (some of) the algebraic structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' A Jordan homomorphism is a linear mapping T with the property T(a2) = (Ta)2 for all a in the domain (which is equivalent to T(ab + ba) = TaTb + TbTa for all a and b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' A Jordan isomorphism turns out to be spectrum-preserving, and a lot of work has been invested to explore to what extent the reverse implication holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' A pleasant survey on the history of this topic is contained in [2];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' see also [8], [13] and [14] for related questions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' In [3], Aupetit proved that every surjective spectrum-preserving linear mapping be- tween von Neumann algebras is a Jordan isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It is not difficult to see that it suffices that one of the algebras is a unital C*-algebra of real rank zero and the other a unital semisimple Banach algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' However, the problem remains open for general C*-algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It is also known that the assumption on T can be relaxed to a surjec- tive unital invertibility-preserving linear mapping (that is, σ(Ta) ⊆ σ(a) for all a);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' the conclusion is then that T is a Jordan homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 2020 Mathematics Subject Classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 47B48, 47A10, 46L05, 46L30, 16W10, 17C65.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Key words and phrases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' C*-algebras, tracial states, Jordan homomorphisms, invertibility preserving mappings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 1 2 MARTIN MATHIEU AND FRANCOIS SCHULZ In an earlier paper [2], Aupetit had already obtained the same result for finite von Neumann algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The main tool in that result was the Fuglede–Kadison determinant ∆;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' see [9, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 105] for its definition and properties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Its relation to the finite trace τ is given by ∆(a) = exp(τ(log |a|)), for every invertible element a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' In our approach we bypass the determinant and work exclusively with a (faithful) tracial state instead in order to obtain the following generalisation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let B be a unital complex Banach algebra and let A be a unital finite C*- algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let T : B → A be a surjective unital linear mapping which preserves invertible elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Then T is a Jordan homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We largely follow Aupetit’s arguments but, to emphasise the differences, we split up the proof into a series of lemmas in the next section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Preliminaries Let A be a unital C*-algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We say that A is finite if it comes equipped with a faithful tracial state, that is, a linear functional τ such that τ(1) = 1 = ∥1∥, τ(ab) = τ(ba) for all a, b ∈ A and τ(a∗a) = 0 implies a = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Such a functional is necessarily positive and bounded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We denote the set of all states of A (positive linear functionals of norm 1) by S and by Sp the subset of all spectral states f of A, that is, f ∈ S and |f(x)| ≤ ρ(x) for every x ∈ A, where ρ(x) denotes the spectral radius of x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It is known ([7, Theorem 4 in §13]) that every f ∈ Sp has the trace property, that is, f(ab) = f(ba) for all a, b ∈ A, and that f(a) ∈ co σ(a), the convex hull of the spectrum σ(a) of a, for each a ∈ A;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' see, [7, Lemma 2 in §13] or [1, Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Conversely, every tracial state τ belongs to Sp as follows from the subsequent argu- ment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' For a ∈ A, denote by V (a) = {f(a) | f ∈ S} its (algebra) numerical range [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' As is shown in [5, Lemma], and attributed to [11, §2], co σ(a) = � b∈G(A) V (bab−1), where G(A) stands for the group of invertible elements in A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Clearly, τ(a) belongs to the right hand side of the above identity and hence, τ ∈ Sp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' (Compare also [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=') Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let A be a unital C*-algebra with faithful tracial state τ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Suppose that g : C → A is an entire function with values in G(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Then the mapping gτ : C → R, gτ(λ) = τ(log(|g(λ)|) is harmonic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The argument in the proof of Th´eor`eme 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='11 in [2], which is already entirely formulated in terms of the trace, takes over verbatim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' □ In the following, T will denote a surjective unital linear mapping defined on a (com- plex, unital) Banach algebra B with values in a finite unital C*-algebra A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We will assume that T preserves invertible elements so that T G(B) ⊆ G(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It follows from [1, Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2] that T is bounded.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Fix a, b ∈ B and define g : C × C −→ G(A), g(λ, µ) = T(eλaeµb)e−λTae−µTb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS 3 Then g is a separately entire function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Its series expansion reads as follows g(λ, µ) = 1 + λ2 2 � T(a2) − (Ta)2� + µ2 2 � T(b2) − (Tb)2� + λµ � T(ab) − TbTa � + λ3 6 � T(a3) + 2(Ta)3 − 3T(a2)Ta � + λ2µ 2 � T(a2b) + (Ta)2Tb + Tb(Ta)2 − T(a2)Tb − 2T(ab)Ta � + λµ2 2 � T(ab2) + 2TbTaTb − 2T(ab)Tb − T(b2)Ta � + µ3 6 � T(b3) + 2(Tb)3 − 3T(b2)Tb � + remainder (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1) where the remainder only contains terms of degree 4 or higher in λ and µ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' we will put it to good use in the proof of the main theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' By Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1, the function gτ : C × C −→ R, gτ(λ, µ) = τ(log(|g(λ, µ)|)) is separately harmonic in λ and µ and thus there exists a separately entire function h(λ, µ) such that Re h(λ, µ) = gτ(λ, µ) for all λ, µ ∈ C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The next step will be to establish the following three lemmas;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' for their proofs, see Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' For all λ, µ ∈ C, we have egτ(λ,µ) ≤ ∥g(λ, µ)∥.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' With the above notation and caveats, let g∗(λ, µ) stand for (g(λ, µ))∗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Then there exists r > 0 such that, for all λ, µ ∈ C with |λ|, |µ| < r, we have 2 Re h(λ, µ) = τ � log(g∗(λ, µ)g(λ, µ) � = − ∞ � k=1 1 kτ � (1 − g∗(λ, µ)g(λ, µ))k� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2) Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' For all λ, µ in a neighbourhood of zero, τ � log(g∗(λ, µ)g(λ, µ) � = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Proofs of the Lemmas and the Main Theorem The argument of the first lemma differs from [2] in that we cannot make use of the determinant in order to locate the appropriate values in the convex hull of the spectrum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Proof of Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' As we observed above, τ(log(|g(λ, µ)|)) ∈ co σ(log(|g(λ, µ)|)) for all λ, µ ∈ C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Since the spectrum of log(|g(λ, µ)|) is contained in R, it follows that co σ(log(|g(λ, µ)|)) = [s, t] for some s, t ∈ σ(log(|g(λ, µ)|)) with s ≤ t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Since the expo- nential function is strictly increasing, the Spectral Mapping Theorem implies that egτ (λ,µ) ∈ [es, et] = co σ(elog(|g(λ,µ)|)) = co σ(|g(λ, µ)|).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 4 MARTIN MATHIEU AND FRANCOIS SCHULZ As a result, 0 < egτ (λ,µ) ≤ ρ(|g(λ, µ)|) and therefore, e2 gτ(λ,µ) ≤ ρ(|g(λ, µ)|)2 = ρ(|g(λ, µ)|2) = ρ(g∗(λ, µ)g(λ, µ)) = ∥g∗(λ, µ)g(λ, µ)∥ = ∥g(λ, µ)∥2 as claimed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' □ The next argument is rather straightforward.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Proof of Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' As g(0, 0) = T1 = 1, by continuity, there is r > 0 such that, for all λ, µ with |λ|, |µ| < r, we have ∥1 − g∗(λ, µ)g(λ, µ)∥ < 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The series expansion of the logarithm thus yields − log(g∗(λ, µ)g(λ, µ)) = ∞ � k=1 1 k(1 − g∗(λ, µ)g(λ, µ))k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1) The definition of gτ entails that 2 Re h(λ, µ) = 2 τ(log(|g(λ, µ)|)) = τ(log(|g(λ, µ)|2)) = τ � log(g∗(λ, µ)g(λ, µ) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2) Combining these two identities gives the claim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' □ The proof of the third lemma follows exactly Aupetit’s arguments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' (There appears to be some misprint at the bottom of page 61 and top of page 62 of [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=') Proof of Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' For all λ, µ ∈ C, we have ��eh(λ,µ)�� = eRe h(λ,µ) = egτ (λ,µ) ≤ ∥g(λ, µ)∥ by Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Since ∥g(λ, µ)∥ ≤ ∥T∥ e|λ|(∥a∥+∥Ta∥)+|µ|(∥b∥+∥Tb∥) it follows that eh(λ,µ) = eαλ+βµ+γ for suitable α, β, γ ∈ C ([4, Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' As gτ(0, 0) = 0 we have |eγ| = 1, thus we may assume that γ = 0 (since we need only the real part of γ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Therefore, 2 Re h(λ, µ) = αλ + βµ + ¯α¯λ + ¯β¯µ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' From Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3 we obtain αλ + βµ + ¯α¯λ + ¯β¯µ = − ∞ � k=1 1 kτ � (1 − g∗(λ, µ)g(λ, µ))k� for all λ, µ such that |λ|, |µ| < r for suitable r > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The series expansion of g(λ, µ) in (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1) does not contain any powers of λ or µ of first order, hence the series expansion in (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1) cannot either.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' This entails that both α and β are equal to zero.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It now follows from (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2) that τ � log(g∗(λ, µ)g(λ, µ) � = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' □ We now have all the tools to prove our main theorem by adapting the arguments in Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='12 of [2] to our situation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS 5 Proof of the Theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Set f(λ, µ) = �∞ k=1 1 kτ � (1 − g∗(λ, µ)g(λ, µ))k� for all λ, µ such that |λ|, |µ| < r for suitable r > 0 (given by Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' By Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='4, f = 0 and thus ∂2 ∂λ∂µf(0, 0) = 0 = ∂3 ∂λ2∂µf(0, 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Using these identities after substituting in the series expansion (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1) into the log-series we find τ � T(ab) − TaTb � = 0 (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3) and τ � T(a2b) + (Ta)2Tb + Tb(Ta)2 − T(a2)Tb − 2 T(ab)Ta � = 0 (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='4) for all a, b ∈ B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' From (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3) we obtain τ(T(a2b)) = τ(T(a2)Tb) and τ(T(a2b)) = τ(T(a(ab))) = τ(TaT(ab)) so that (3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='4) reduces to τ((Ta)2Tb) = τ(TaT(ab)), using the trace property.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It follows that τ((Ta)2Tb) = τ(T(a2)Tb).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Since T is surjective we may choose b ∈ B such that Tb = ((Ta)2 − T(a2))∗ wherefore the last identity yields, for each a ∈ B, that τ � ((Ta)2 − T(a2))((Ta)2 − T(a2))∗� = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The faithfulness of τ implies that T is a Jordan homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' □ 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Conclusions In this section, we collect together some consequences and sharpening of our main theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We also relate it to open problems of a similar nature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Suppose T is a surjective linear mapping between two semisimple unital Banach algebras which preserves the spectrum of each element.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Then T is injective (as explained in the Introduction) and T1 = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The latter follows, for example, from σ((T1 − 1) + Tx) = σ(T1 + Tx) − 1 = σ(1 + x) − 1 = σ(x) = σ(Tx) and the surjectivity of T which entails that σ((T1 − 1) + y) = σ(y) for all y in the codomain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Thus, by Zem´anek’s characterisation of the radical, T1 − 1 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' As a result, we have a symmetric situation and can apply the Theorem to either T or its inverse to obtain the following consequence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let T be a surjective spectrum-preserving linear mapping between two semisimple unital Banach algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' If either of them is a unital C*-algebra equipped with a faithful tracial state then T is a Jordan isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 6 MARTIN MATHIEU AND FRANCOIS SCHULZ This is another contribution to a longstanding, still open problem by Kaplansky who asked in 1970 whether the above statement holds without any further assumptions on the Banach algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' For further references, see [2] and [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' All the steps in the proof of the Theorem but the very last one can be performed for each individual tracial state on a unital C*-algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Therefore, the assumption can be relaxed to the existence of a faithful family of tracial states, that is, a family {τi | i ∈ I} of tracial states τi such that τi(a∗a) = 0 for all i ∈ I implies a = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' In particular, since any tracial state on a simple unital C*-algebra is faithful we obtain the following result.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let T : B → A be a surjective unital invertibility-preserving linear mapping into a simple unital C*-algebra A which carries a tracial state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Then T is a Jordan homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Remark 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Our terminology of a “finite”C*-algebra is not quite standard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' In [6, III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1], a unital C*-algebra A is called finite if the identity of A is a finite projection;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' that is, there is no proper subprojection which is Murray–von Neumann equivalent to 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Every unital C*-algebra with a faithful tracial state is finite in this sense but the converse fails in general (though it holds for stably finite exact C*-algebras).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We prefer here a definition that does not make reference to any projections.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' We can also strengthen our main theorem in a different direction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let G1(B) denote the principal component of G(B), where B is a unital Banach algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' It is known, see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=', [1, Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='7], that G1(B) = {ex1 · · · exn | xi ∈ B, n ∈ N}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The associated exponential spectrum of x ∈ B is σε(x) = {λ ∈ C | λ − x /∈ G1(B)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' In certain situations it is more natural and expedient to consider the exponential spec- trum instead of the smaller spectrum, see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=', [1, Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' From the proof of our main result we see that it suffices that the mapping T sends the product of any two exponentials in B onto an invertible element in A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' This gives the following corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let B be a unital complex Banach algebra and let A be a unital finite C*- algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Let T : B → A be a surjective unital linear mapping such that TG1(B) ⊆ G(A).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Then T is a Jordan homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' A spectral isometry between two Banach algebras A and B is a linear mapping S such that ρ(Sx) = ρ(x) for all x ∈ A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Clearly, every spectrum-preserving mapping is a spectral isometry and so is every Jordan isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' A conjecture related to Ka- plansky’s problem mentioned above states that every unital surjective spectral isometry between two C*-algebras is a Jordan isomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' This conjecture has been confirmed in many cases, see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=', [13, 14], but is open in all generality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Notably it was verified in [15] if A is a unital C*-algebra of real rank zero and without tracial states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The above Corollary 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='1 is thus a step forward in the direction of confirming the general conjecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' INVERTIBILITY PRESERVING MAPPINGS ONTO FINITE C*-ALGEBRAS 7 Acknowledgements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' The research for this paper was completed while the second- named author was visiting Queen’s University Belfast.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' He would like to thank both Professor Martin Mathieu and the Mathematical Sciences Research Centre at Queen’s University Belfast for their hospitality, and the National Research Foundation of South Africa for their financial support (NRF Grant Number: 129692).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' References [1] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Aupetit, A Primer on Spectral Theory, Springer-Verlag, New York, 1991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [2] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Aupetit, Sur les transformations qui conservent le spectre, in: Banach Algebras ’97, Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 13th Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Conf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' on Banach Algebras, Blaubeuren, July 20–August 3, 1997, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mathieu and E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Albrecht (eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' ), de Gruyter, Berlin, 1998, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 55–78.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [3] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Aupetit, Spectrum-preserving linear mappings between Banach algebras or Jordan–Banach al- gebras, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' London Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 62 (2000), 917–924.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [4] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Aupetit and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' du T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mouton, Trace and determinant in Banach algebras, Studia Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 121 (1996), 115–136.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [5] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Berberian, Trace and the convex hull of the spectrum in a von Neumann algebra of finite class, Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 23 (1969), 211–212.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [6] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Blackadar, Operator Algebras.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Theory of C*-Algebras and von Neumann Algebras, Encyclopae- dia of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Sciences 122, Springer-Verlag, Berlin, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [7] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Bonsall and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Duncan, Numerical Ranges of Operators and of Elements of Normed Algebras, London Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Lecture Notes 2, Cambridge Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Press, Cambridge, 1971.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [8] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Brits, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Schulz and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Tour´e, A spectral characterization of isomorphisms on C*-algebras, Arch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 113 (2019), 391–398.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [9] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Dixmier, Les alg`ebres d’op´erateurs dans l’espace Hilbertien, Gauthier-Villars, Paris, 1957.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [10] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Harris, Invertibility preserving linear maps of Banach algebras, in: Complex Analysis and Dynamical Systems, Contemp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 364, Israel Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Conf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=', Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=', Provi- dence, RI, 2004;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 59–66.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [11] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Hildebrandt, ¨Uber den numerischen Wertebereich eines Operators, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 163 (1966), 230–247.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [12] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mathieu, Spectrally bounded traces on C*-algebras, Bull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Austral.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 68 (2003), 169– 173.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [13] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mathieu, A collection of problems on spectrally bounded operators, Asian-Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 2 (2009), 487–501.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [14] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mathieu, Interplay between spectrally bounded operators and complex analysis, Irish Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Bull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' 72 (2013), 1–13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' [15] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mathieu, Characterizing Jordan homomorphisms, Acta Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' (Szeged) 86 (2020), 697–701.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Mathematical Sciences Research Centre, Queen’s University Belfast, Belfast BT7 1NN, Northern Ireland Email address: m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='m@qub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='uk Department of Mathematics and Applied Mathematics, Faculty of Science, Univer- sity of Johannesburg, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content=' Box 524, Auckland Park, 2006, South Africa Email address: francoiss@uj.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} +page_content='za' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/mtAyT4oBgHgl3EQfYvcj/content/2301.00208v1.pdf'} diff --git a/nNE5T4oBgHgl3EQfHw7p/vector_store/index.faiss b/nNE5T4oBgHgl3EQfHw7p/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..82e6cef5a00c63c68b22963031000e6596670997 --- /dev/null +++ b/nNE5T4oBgHgl3EQfHw7p/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28e2b888c4f497620c912d3234c3fba0bff4ca53c05066b821791a69b06da98d +size 1441837 diff --git a/ndFLT4oBgHgl3EQffi-_/content/tmp_files/2301.12096v1.pdf.txt b/ndFLT4oBgHgl3EQffi-_/content/tmp_files/2301.12096v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..60f2807f21795772cda0ae504158ddf6c8b76931 --- /dev/null +++ b/ndFLT4oBgHgl3EQffi-_/content/tmp_files/2301.12096v1.pdf.txt @@ -0,0 +1,1516 @@ +QD3SET-1: A Database with Quantum Dissipative +Dynamics Data Sets +Arif Ullah1, Luis E. Herrera Rodr´ıguez2, Pavlo O. Dral1,*, and Alexei A. Kananenka2,* +1State Key Laboratory of Physical Chemistry of Solid Surfaces, Fujian Provincial Key Laboratory of Theoretical and +Computational Chemistry, Department of Chemistry, and College of Chemistry and Chemical Engineering, Xiamen +University, Xiamen 361005, China +2Department of Physics and Astronomy, University of Delaware, Newark, Delaware 19716, United States +*corresponding author(s): Pavlo O. Dral (dral@xmu.edu.cn) and Alexei A. Kananenka (akanane@udel.edu) +ABSTRACT +Simulations of the dynamics of dissipative quantum systems utilize many methods such as physics-based quantum, semiclassi- +cal, and quantum-classical as well as machine learning-based approximations, development and testing of which requires +diverse data sets. Here we present a new database QD3SET-1 containing eight data sets of quantum dynamical data for +two systems of broad interest, spin-boson (SB) model and the Fenna–Matthews–Olson (FMO) complex, generated with two +different methods solving the dynamics, approximate local thermalizing Lindblad master equation (LTLME) and highly accurate +hierarchy equations of motion (HEOM). One data set was generated with the SB model which is a two-level quantum system +coupled to a harmonic environment using HEOM for 1,000 model parameters. Seven data sets were collected for the FMO +complex of different sizes (7- and 8-site monomer and 24-site trimer with LTLME and 8-site monomer with HEOM) for 500–879 +model parameters. Our QD3SET-1 database contains both population and coherence dynamics data and part of it has been +already used for machine learning-based quantum dynamics studies. +Background & Summary +The simulation of the inherently quantum-mechanical dynamics underlying charge, energy, and coherence transfer in the +condensed-phase is one of the most difficult challenges for computational physics and chemistry. The exponential scaling of the +computational cost with system size makes the quantum-mechanically exact simulations of such processes in complex systems +infeasible. With the exception of a few model Hamiltonians whose form makes the numerically exact quantum-dynamics +simulations possible, any simulation of general condensed-phase systems must rely on approximations.1–28 Data-driven +machine learning (ML) methods for quantum dynamics emerged as attractive alternative to the physics-based approximations +due to their low computational cost and high accuracy.29–46 Development and testing of new simulation methodologies, both +physics- and ML-based, would be greatly facilitated if high-quality reference quantum-dynamics data for a diverse set of +quantum systems of interest were available. +Here we present a QD3SET-1 database, a collection of eight data sets of time-evolved population dynamics of the +two systems: spin-boson (SB) model and the Fenna–Matthews–Olson (FMO) light-harvesting complex. The data sets are +summarized in Table 1. The SB model describes a (truncated or intrinsic) two-level quantum system linearly coupled to a +harmonic bath.47,48 The physics of both the ground state and the dynamics of the SB model is very rich. It has been a continuous +object of study during the past decades. SB has become a paradigmatic model system in the development of approximate +quantum-dynamics methods and, nowadays, it is becoming a popular choice for the development of ML models.30,33,40 +The FMO system has become one of the most extensively studied natural light-harvesting complexes.49–55 Under physio- +logical conditions, the FMO complex forms a homotrimer consisting of eight bacteriochlorophyll-a (BChla) molecules per +monomer. The biological function of the FMO trimer is to transfer excitation energy from the chlorosome to the reaction center +(RC).56 An interest in this light-harvesting system sparked when two-dimensional electronic spectroscopy experiments detected +the presence of quantum coherence effects in the FMO complex.57–59 These observations triggered intense debates about the +role this coherence might play in the highly efficient excitation energy transfer (EET). +Early studies of the FMO complex considered only seven-site FMO models comprising of BChla 1–7. Until BChla 8 +was discovered, BChla 1 and 6 were both assumed to be possible locations for accepting the excitation from the chlorosome +because they are believed to be the nearest pigments to the antenna which captures the sunlight.60–62 From there, the energy is +subsequently funneled through two nearly independent routes: from site 1 to 2 (pathway 1) or from site 6 to sites 7, 5, and 4 +(pathway 2). The terminal point of either route is site 3, where the exciton is then transferred to the RC.63 +Ever since the discovery of the eighth BChla, the role of this pigment in the EET has been extensively investigated.53,63–70 +arXiv:2301.12096v1 [physics.chem-ph] 28 Jan 2023 + +In particular, it was shown that while the population dynamics of the eight-site FMO model is markedly different from a +seven-site configuration, the EET efficiencies in both models were predicted to remain comparable and very high.63 BChla 8 +has also been suggested as possible recipient of the initial excitation. +The dynamics of the FMO model has been a subject of numerous computational studies primarily focusing on understanding +the role of the protein environment on the efficiency of EET (see e.g., Refs. 71–74). Numerical simulations typically employ +one of the several parameterized or fitted into the experimental data FMO model Hamiltonians that differ in the BChla excitation +energies and the couplings between different BChla sites.49,56,75–79 Simulations of the full FMO trimer containing 24 BChl +have also been performed.80,81 +Reported in this Data Descriptor a QD3SET-1 database contains seven data sets of time-evolved population dynamics of +FMO models with different system Hamiltonians and initial excitations for several hundreds of bath and system-bath parameters. +Hierarchy of equations of motion (HEOM) approach5,7 was used to simulate the population dynamics of SB and FMO models, +in one of the seven FMO data sets. HEOM is a numerically exact method that can describe the dynamics of a system with a +non-perturbative and non-Markovian system–bath interaction. The high computational cost of HEOM, however, limits the +number of FMO simulations that can be performed with this method. To generate other six FMO data sets, an approximate +method—local thermalizing Lindblad master equation (LTLME)82,83 was used. +Some of our data was already used in previous studies developing and benchmarking ML models for quantum-dynamics +simulations.30–33 Here we regenerate one of the data sets to augment with more data and provide many new data sets generated +from scratch (Table 1). To facilitate their use, we organize the data sets in a coherently formatted database and provided +metadata and extraction scripts. We expect our Database that accompany this Data Descriptor will serve as valuable resources +in the development of new quantum-dynamics methods. +Methods +SB data set +This data set is re-generated with the same settings and the same parameters as in on our previous SB data set33 in order to +include all the elements (populations and coherences) of the system’s reduced density matrix (RDM). The populations and +population differences were published and used before.30,33 Below we provide a brief summary for self-containing presentation +of the data set. +Spin-boson model +The spin-boson model comprises a two-level quantum subsystem (TLS) coupled to a bath of harmonic oscillators. The +Hamiltonian has the following standard system-bath form: ˆH = ˆHs + ˆHb + ˆHsb. The Hamiltonian of the TLS in the local (or +site) basis {|+⟩,|−⟩} is given by (ℏ=1) +ˆHs = ε (|+⟩⟨+|−|−⟩⟨−|) + ∆(|+⟩⟨−|+|−⟩⟨+|), +(1) +where ε is the so-called energetic bias and ∆ is the tunneling splitting. The harmonic bath is an ensemble of independent +harmonic oscillators +ˆHb = +Nb +∑ +j=1 +� +ˆp2 +j +2m j ++ 1 +2m jω2 +j ˆx2 +j +� +, +(2) +where {ˆx j} and { ˆpj} are the coordinates and momenta, respectively, of Nb independent harmonic bath modes with masses +{m j} and frequencies {ωj}. The TLS and bath are coupled through the additional term +ˆHsb = − +Nb +∑ +j=1 +cj ˆxj (|+⟩⟨+|−|−⟩⟨−|), +(3) +where {cj} are the coupling coefficients. +The effects of the bath on the dynamics of TLS are collectively determined by the spectral density function84 +J(ω) = π +2 +Nb +∑ +j=1 +c2 +j +m jωj +δ(ω −ωj). +(4) +In this work we choose to employ the Debye form of the spectral density (Ohmic spectral density with the Drude–Lorentz +cut-off)85 +J(ω) = 2λ +ωγ +ω2 +γ2 , +(5) +2/15 + +where λ is the bath reorganization energy, which controls the strength of system-bath coupling, and the cutoff frequency +γ = 1/τc (τc is the bath relaxation time). +All dynamical properties of the TLS can be obtained from the RDM +˜ραβ(t) = Trb⟨α|e−i ˆHt/ℏ ˆρ(0)ei ˆHt/ℏ|β⟩, +(6) +where α,β ∈ {|+⟩,|−⟩}, ˆρ is the total density operator, and the trace is taken over bath degrees of freedom. For example, +the commonly used in benchmark studies population difference is obtained from the RDM as follows: p+(t) − p−(t) = +˜ρ++(t)− ˜ρ−−(t). +The initial state of the total system is assumed to be a product state of the system and bath in the following form +ˆρ(0) = ˆρs(0) ˆρb(0). +(7) +In Eq. (7) the bath density operator is an equilibrium canonical density operator ˆρb(0) = e−β ˆHb/Trb +� +e−β ˆHb +� +, where β = +(kBT)−1 is the inverse temperature and kB is the Boltzmann constant. The initial density operator of the system is chosen to be +ˆρs(0) = |+⟩⟨+|. These conditions corresponds to instantaneous photoexcitation of the subsystem. +Data generation with spin-boson model and the hierarchy equations of motion approach +The data set for the spin-boson model was generated as described previously.33 We also summarize it below. The following +system and bath parameters were chosen: ˜ε = ε/∆ = {0,1}, ˜λ = λ/∆ = {0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0}, ˜γ = +γ/∆ = {1,2,3,4,5,6,7,8,9,10}, and ˜β = β∆ = {0.1,0.25,0.5,0.75,1}, where the tunneling matrix element ∆ is set as an +energy unit. For all combinations of these parameters the system’s RDM was propagated using HEOM approach implemented +in QUTIP software package.86 The total propagation time was tmax∆ = 20 and the HEOM integration time-step was set to +dt∆ = 0.05. In total, 1,000 of HEOM calculations, 500 for symmetric (ε/∆ = 0) and 500 for asymmetric (ε/∆ = 1) spin-boson +Hamiltonian were performed. The data set contains a set of RDMs from t∆ = 0 to tmax∆ = 20, saved every dt∆ = 0.05, for +every combination of the parameters (˜ε, ˜λ, ˜γ, ˜β) described above. +Fenna–Matthews–Olson complex data sets +In this section we first describe the general theory behind the FMO model Hamiltonian and later for each data set we provide +specific technical details. See also Table 1 for an overview of each data set. +FMO model Hamiltonian +The FMO complex in this work is described by the system-bath Hamiltonian with the renormalization term ˆH = ˆHs + ˆHb + +ˆHsb + ˆHren. The electronic system is described by the Frenkel exciton Hamiltonian +ˆHs = +Ne +∑ +n=1 +En|n⟩⟨n|+ +Ne +∑ +n,m=1,n̸=m +Vnm|n⟩⟨m|, +(8) +where |n⟩ denotes that only the nth site is in its electronically excited state and all other sites are in their electronically ground +states, En are the transition energies, and Vnm is the Coulomb coupling between nth and mth sites. The couplings are assumed +to be constant (the Condon approximation). Note that the overall electronic ground state of the pigment protein complex |0⟩ +is assumed to be only radiatively coupled to the single-excitation manifold and as such it is not included in the dynamics +calculations. In analogy with the SB model the bath is modeled by a set of independent harmonic oscillators. The thermal bath +is coupled to the subsystem’s states |n⟩ through the system-bath interaction term +ˆHsb = +Ne +∑ +n=1 +Nb +∑ +j=1 +cn j ˆxj|n⟩⟨n|, +(9) +where each subsystem’s state is independently coupled to its own harmonic environment and cnj are the pigment-phonon +coupling constants of environmental phonons local to the nth BChla. +The FMO model Hamiltonian contains a reorganization term which counters the shift in the minimum energy positions of +harmonic oscillators introduced by the system-bath coupling. In the case that each state |n⟩ is independently coupled to the +environment the renormalization term takes the following form +ˆHren = +Ne +∑ +n=1 +λn|n⟩⟨n|, +(10) +3/15 + +where λn = ∑j c2 +nj/(2m jω2 +j ) is the bath reorganization energy. The bath spectral density associated with each electronic state is +assumed to be given by the Lorentz–Drude spectral density (Eq. 5). +Analogously to the SB data set the initial state of the total system is assumed to be a product state of the system and bath. +The initial electronic density operator given by ˆρs(0) was varied as described below. The bath density operator is taken to be +the equilibrium canonical density operator. +FMO-Ia, FMO-Ib, and FMO-II data sets: 7-site FMO models with the local thermalizing Lindblad master equation ap- +proach +We generated data sets for the two 7-site system (Ne = 7) Hamiltonians. FMO-I data set was generated for the system +Hamiltonian parameterized by Adolphs and Renger49 and given by (in cm−1) +Hs = +� +� +� +� +� +� +� +� +� +� +200 +−87.7 +5.5 +−5.9 +6.7 +−13.7 +−9.9 +−87.7 +320 +30.8 +8.2 +0.7 +11.8 +4.3 +5.5 +30.8 +0 +−53.5 +−2.2 +−9.6 +6.0 +−5.9 +8.2 +−53.5 +110 +−70.7 +−17.0 +−63.6 +6.7 +0.7 +−2.2 +−70.7 +270 +81.1 +−1.3 +−13.7 +11.8 +−9.6 +−17.0 +81.1 +420 +39.7 +−9.9 +4.3 +6.0 +−63.3 +−1.3 +39.7 +230 +� +� +� +� +� +� +� +� +� +� +, +(11) +FMO-Ia data set comes directly from our previous studies31,32 and FMO-Ib data set was generated here for a broader parameter +space as described below. +FMO-II data set was generated for the Hamiltonian parameterized by Cho et al.76 which takes the following form (in cm−1) +Hs = +� +� +� +� +� +� +� +� +� +� +280 +−106 +8 +−5 +6 +−8 +−4 +−106 +420 +28 +6 +2 +13 +1 +8 +28 +0 +−62 +−1 +−9 +17 +−5 +6 +−62 +175 +−70 +−19 +−57 +6 +2 +−1 +−70 +320 +40 +−2 +−8 +13 +−9 +−19 +40 +360 +32 +−4 +1 +17 +−57 +−2 +32 +260 +� +� +� +� +� +� +� +� +� +� +. +(12) +The diagonal offset of 12210 cm−1 is added to both Hamiltonians. Each site is coupled to its own bath characterized by the +Drude–Lorentz spectral density, Eq. 5, but the bath of each site is described by the same spectral density. +For FMO-Ia data set, the following spectral density parameters and temperatures were employed: λ = {10, 40, 70, . . . , 310} +cm−1, γ = {25, 50, 75, ..., 300} fs rad−1, and T = {30, 50, 70, ..., 310} K. For FMO-Ib and FMO-II data sets, the spectral +density parameters and temperatures were: λ = {10, 40, 70, . . . , 520} cm−1, γ = {25, 50, 75, . . . , 500} cm−1, and T = {30, 50, +70, . . . , 510} K. +For FMO-Ia, FMO-Ib, and FMO-II data sets, the farthest-point sampling87 was employed to select the most distant points +in the Euclidean space32 of parameters which typically more efficiently covers relevant space compared to random sampling.87 +We choose the top 500 (most distant) combinations of (λ , γ, T) based on farthest-point sampling. For each selected set of +parameters the system RDM was calculated using the local thermalizing Lindblad master equation (LTLME) approach82,83 +implemented in the QUANTUM_HEOM package.83,88 Two subsets of the data set were generated, one for the initial electronic +density operator ˆρs(0) = |1⟩⟨1| corresponding to the initial excitation of site-1 and the other one for the initial density operator +ˆρs(0) = |6⟩⟨6| which corresponds to the initial excitation of site-6. In each case, 500 RDM trajectories were generated. The +data set contains both diagonal (populations) and off-diagonal (coherences) elements of the RDM on a time grid from 0 to 1 ns +(in the case of FMO-Ia) and 0 to 50 ps (in the case of FMO-Ib and FMO-II) with the 5 fs time step. +4/15 + +FMO-III and FMO-IV data sets: 8-site FMO models with the local thermalizing Lindblad master equation approach +Using the same LTLME-based approach, we generated a data set for two different Hamiltonians for the 8-site FMO model. The +first Hamiltonian (FMO-III data set) was parameterized by Jia et al.89 The electronic system Hamiltonian is given by (in cm−1) +Hs = +� +� +� +� +� +� +� +� +� +� +� +� +218 +−91.0 +4.1 +−6.3 +6.3 +−8.8 +−7.8 +32.4 +−91.0 +81 +28.7 +8.2 +1.0 +8.8 +3.4 +6.3 +4.1 +28.7 +0 +−46.6 +−4.4 +−9.3 +1.3 +1.3 +−6.3 +8.2 +−46.6 +105 +−73.9 +−17.7 +−59.1 +−1.9 +6.3 +1.0 +−4.4 +−73.9 +105 +76.0 +−3.1 +4.2 +−8.8 +8.8 +−9.3 +−17.7 +76.0 +186 +25.9 +−11.6 +−7.8 +3.4 +1.3 +−59.1 +−3.1 +25.9 +169 +−11.9 +32.4 +6.3 +1.3 +−1.9 +4.2 +−11.6 +−11.9 +154 +� +� +� +� +� +� +� +� +� +� +� +� +, +(13) +with the diagonal offset of 11332 cm−1. +The FMO-IV data set was generated for the Hamiltonian parameterized by Busch et al.64 (site energies) and Olbrich et al.67 +(excitonic couplings) and takes the following form (in cm−1) +Hs = +� +� +� +� +� +� +� +� +� +� +� +� +310 +−80.3 +3.5 +−4.0 +4.5 +−10.2 +−4.9 +21.0 +−80.3 +230 +23.5 +6.7 +0.5 +7.5 +1.5 +3.3 +3.5 +23.5 +0 +−49.8 +−1.5 +−6.5 +1.2 +0.7 +−4.0 +6.7 +−49.8 +180 +63.4 +−13.3 +−42.2 +−1.2 +4.5 +0.5 +−1.5 +63.4 +450 +55.8 +4.7 +2.8 +−10.2 +7.5 +−6.5 +−13.3 +55.8 +320 +33.0 +−7.3 +−4.9 +1.5 +1.2 +−42.2 +4.7 +33.0 +270 +−8.7 +21.0 +3.3 +0.7 +−1.2 +2.8 +−7.3 +−8.7 +505 +� +� +� +� +� +� +� +� +� +� +� +� +, +(14) +with the diagonal offset of 12195 cm−1. +The same set of spectral density parameters and temperatures that was used in generation of the FMO-Ib and FMO-II +data sets was used here. LTLME method was used to propagate system’s RDM from 0 to 50 ps with 5 fs time-step and three +initial states of the electronic system were considered: sites-1, 6 and 8. The data set contains both diagonal (populations) and +off-diagonal (coherences) elements of the RDM. The calculations was performed with the QUANTUM_HEOM package88 +with some local modifications to make it compatible for the Hamiltonians with larger dimension. We will refer to this as +MODIFIED-QUANTUM_HEOM implementation. +FMO-V data set: FMO trimer with local thermalizing Lindblad master equation approach +Additionally, we also generated a data set for the FMO trimer. The overall excitonic Hamiltonian of all three subunits is given +by +Hs = +� +� +HA +HB +HT +B +HT +B +HA +HB +HB +HT +B +HA +� +� +(15) +where HA is the subunit Hamiltonian for which we used the same Hamiltonian as in FMO-IV data set (Eq. 14), while HB is the +intra-subunit Hamiltonian which is taken from the work of Olbrich et al.67 and is given by (in cm−1) +HB = +� +� +� +� +� +� +� +� +� +� +� +� +1.0 +0.3 +−0.6 +0.7 +2.3 +1.5 +0.9 +0.1 +1.5 +−0.4 +−2.5 +−1.5 +7.4 +5.2 +1.5 +0.7 +1.4 +0.1 +−2.7 +5.7 +4.6 +2.3 +4.0 +0.8 +0.3 +0.5 +0.7 +1.9 +−0.6 +−0.4 +1.9 +−0.8 +0.7 +0.9 +1.1 +−0.1 +1.8 +0.1 +−0.7 +1.3 +0.1 +0.7 +0.8 +1.4 +−1.4 +−1.5 +1.6 +−1.0 +0.3 +0.2 +−0.7 +4.8 +−1.6 +0.1 +5.7 +−2.3 +0.1 +0.6 +1.5 +−1.1 +4.0 +−3.1 +−5.2 +3.6 +� +� +� +� +� +� +� +� +� +� +� +� +. +(16) +We propagate dynamics with LTLME from 0 to 50 ps with 5 fs time-step for the same parameters as was adopted in calculations +for the FMO-Ib—FMO-IV data sets. The calculations were performed with the MODIFIED-QUANTUM_HEOM implementation +for the initial excited sites-1, 6 and 8. +5/15 + +FMO-VI data set: 8-site FMO model with the hierarchy of equations of motion approach +The LTLME approach provides only approximate description of quantum dynamics of the FMO complex. Therefore, the +FMO-I—FMO-V data sets are useful merely for the developing machine learning models for quantum dynamics studies. For +example, they can be used to train a neural network model which can then be further improved on more accurate but smaller +data sets (e.g., via transfer learning). However, LTLME dynamics cannot be used to benchmark other quantum dynamics +methods. In the latter case the high-quality reference data is needed. +To generate a data set with accurate FMO dynamics we performed HEOM calculations for the 8-site FMO model with the +Hamiltonian given by Eq. 14. HEOM calculations were performed using the parallel hierarchy integrator (PHI) code.90 The +initial data set was chosen on the basis of farthest-point sampling similar to how it was done in the FMO-Ib—FMO-V data +sets with the only difference being that instead of 500 most distant sets of parameters that were chosen in the preparation of +FMO-Ib—FMO-V data sets, 1100 most distant set of parameters were used to prepare the initial FMO-VI data set. For certain +parameters, the RAM requirements exceeded the RAM of computing nodes available to us (1 TB). Therefore, such parameter +sets were excluded from the data set. Excluded parameters correspond to low temperatures, high reorganization energies, and +low cut-off frequency. Such strong non-Markovian regimes pose significant challenges in the computational studies of open +quantum systems. Approximately 20% of the initial data set was removed because of prohibitive memory requirements. We +note that even though graphics processing units (GPU) implementations of HEOM (e.g., Ref. 91) are much faster than their +CPU-based counterparts, they are still limited by the small amount of memory in presently available GPUs. +For the remaining 80% of the data set HEOM calculations were performed for 2.0 ps. To speed up calculations, an +adaptive integration Runge–Kutta–Fehlberg 4/592 (RKF45) method was used as implemented in the PHI code. Using adaptive +integration reduces both the total computation time and memory requirements but can lead to artifacts if the accuracy threshold +is set too large.90 In this work the PHI default accuracy threshold of 1·10−6 was used. The initial integration time step was +set to 0.1 fs. In RKF45 the integration time step is varied and, therefore, the output comprises time-evolved RDMs on an +unevenly spaced time grid. To obtain the RDMs on an evenly spaced time grid of 0.1 fs, cubic-spline interpolation was used. +The interpolation errors were examined on a few cases where 0.1 fs fixed time step integration was feasible. The errors in the +populations were found to be less than 10−5 which is much smaller compared to the convergence thresholds discussed below in +Technical Validation. The final FMO-VI data set contains 879 entries each comprising all the populations and coherences for +the RDM from 0 to 2 ps with the time-step of 0.1 fs. +Data Records +All data sets can be accessed at https://figshare.com/s/ed24594205ab87404238. The data sets are stored +in standard NumPy93 binary file format (.npy) files. The following format of file names was adopted in the SB data set +2_epsilon-X_Delta-1.0_lambda-Y_gamma-Z_beta-XX.npy, where X denotes the value of the energetic bias +(˜ε), Y is the reorganization energy ˜λ, Z is the cut-off frequency ˜γ, and XX is the inverse temperature ˜β. The following format of +file names was adopted in all FMO data sets X_initial-Y_gamma-Z_lambda-XX_temp-YY.npy, where X denotes +the number of sites in the FMO model, Y is the initial state, Z is the value of bath frequency, XX is the value of reorganization +energy, and YY is the temperature. +Technical Validation +Central to the HEOM approach is the assumption that the bath correlation function Ca(t) for site a can be represented by +an infinite sum of exponentially decaying terms Ca(t) = ∑∞ +k cak exp(−νakt), where νak = 2πk/βℏ are Matsubara frequencies. +Further, each exponential term leads to a set of auxiliary density matrices which take into account the non-Markovian evolution +of the system’s RDM under the influence of bath. In practice, the summation must be truncated at a finite level, K, which +is called Matsubara cut-off and the set of auxiliary density matrices needs to be truncated at a finite number M. In the +truncated set of auxiliary matrices are indexed by n = (n10,...,n1K,nM0,...,nMK). The hierarchy truncation level is given by +L = ∑M +a=1 ∑K +k=0 nak, where nak is the index of an auxiliary density matrix. The computational cost of the HEOM method rises +steeply with the hierarchy level L.90 +The hierarchy truncation level L depends on how non-Markovian the system is. Although, there is some guidance on how +to choose the Matsubara cut-off and hierarchy truncation level based on bath and spectral density parameters,5,50 in practice, +the values of M and K have to be chosen by requiring the convergence of the RDM to acceptable accuracy level. In this work +HEOM calculations for the SB model were performed by setting L = 30 for all temperatures. The Matsubara cut-off was +chosen depending on the temperature as follows: for β = 0.1 K was set to 2; for β = 0.25,K = 3, for β = 0.5,K = 3, for +β = 0.75,K = 4, and for β = 1.0,K = 5. These values are chosen sufficiently high to ensure the convergence of the populations +with respect to K and L. Choosing high truncation levels in the HEOM calculations of a TLS does not present a problem given +the presently available computational resources. +6/15 + +Similar approach of taking excessively large values of K and L, however, is infeasible in the FMO calculations because +the computational cost of HEOM grows steeply with the size of the quantum system. Therefore, the following approach was +adopted for the HEOM calculations of the 8-site FMO model (FMO-VI data set). Starting from K = 0 and L = 1, K was +increased until the maximum difference in the populations between calculations with K and K +1 falls below a threshold ∆, +i.e., +δ = +max +n=1,...,Nel +t=0,...,tmax +����ρK,L +n,n (t)−ρK+1,L +n,n +(t) +���� < ∆. +(17) +When Eq.17 is satisfied for a given ∆ the convergence with respect to Matsubara cut-off is deemed to have been achieved. +Then, for a fixed K a series of calculations were performed with increasing values of L until the maximum difference in +populations between two consecutive calculations becomes less than the same threshold value ∆. When this condition is satisfied +the convergence with respect to hierarchy truncation level as well as the overall convergence is declared. These steps were +performed in the HEOM calculations for each parameter set for an 8-site FMO model until either the overall convergence is +achieved or K and/or L become large enough so the calculation becomes intractable exceeding RAM available on our machines +(1 TB). +In this work we set the threshold ∆ = 0.01. This threshold was chosen such that the population errors would be almost +imperceptible which is illustrated in Figure 1. This data set is converged enough to be helpful in benchmarks of approximate +methods describing quantum dynamics because the errors of these methods often exceed the threshold used in this work. +Additionally, Figures 2 and 3 show the number of Matsubara terms and the hierarchy truncation level required for achieving +the overall convergence depending on spectral density parameters and temperature. +Usage Notes +A Python package for extracting data is provided together with the data set and can be accessed at https://github.com/Arif- +PhyChem/QD3SET. +Code availability +PHI code (version 1.0) used in HEOM calculations was downloaded from http://www.ks.uiuc.edu/Research/phi/. QUTIP +software package (version 4.6) used in HEOM calculations of the spin-boson model and was downloaded from https://qutip.org/. +LTLME calculations of FMO models were performed with the basic QUANTUM_HEOM package https://github.com/jwa7/quantum_HEOM +and was modified to enable compatibility with the Hamiltonian of larger than the default dimension. +References +1. Meyer, H., Gatti, F. & Worth, G. Multidimensional Quantum Dynamics: MCTDH Theory and Applications (Wiley, 2009). +2. Makri, N. Time-dependent quantum methods for large systems. Annu. Rev. Phys. Chem. 50, 167 – 191, 10.1146/annurev. +physchem.50.1.167 (1999). +3. Meyer, H.-D., Manthe, U. & Cederbaum, L. The multi-configurational time-dependent Hartree approach. Chem. Phys. +Lett. 165, 73 – 78, https://doi.org/10.1016/0009-2614(90)87014-I (1990). +4. Wang, H. & Thoss, M. Multilayer formulation of the multiconfiguration time-dependent hartree theory. J. Chem. Phys. +119, 1289–1299 (2003). +5. Tanimura, Y. Numerically “exact” approach to open quantum dynamics: The hierarchical equations of motion (HEOM). J. +Chem. Phys. 153, 020901, 10.1063/5.0011599 (2020). 2006.05501. +6. Tanimura, Y. & Kubo, R. Two-time correlation functions of a system coupled to a heat bath with a Gaussian–Markoffian +interaction. Proc. Jpn. Soc. 58, 1199–1206 (1989). +7. Tanimura, Y. Nonperturbative expansion method for a quantum system coupled to a harmonic-oscillator bath. Phys. Rev. A +41, 6676–6687, 10.1103/PhysRevA.41.6676 (1990). +8. Greene, S. M. & Batista, V. S. Tensor-train split-operator fourier transform (tt-soft) method: Multidimensional nonadiabatic +quantum dynamics. J. Chem. Theory Comput. 13, 4034–4042 (2017). +9. Kapral, R. PROGRESS IN THE THEORY OF MIXED QUANTUM-CLASSICAL DYNAMICS. Annu. Rev. Phys. Chem. +57, 129–157, 10.1146/annurev.physchem.57.032905.104702 (2006). +7/15 + +10. Kapral, R. Surface hopping from the perspective of quantum–classical Liouville dynamics. Chem. Phys. 481, 77 – 83, +10.1016/j.chemphys.2016.05.016 (2016). +11. Min, S. K., Agostini, F., Tavernelli, I. & Gross, E. K. Ab initio nonadiabatic dynamics with coupled trajectories: A rigorous +approach to quantum (de) coherence. The J. Phys. Chem. Lett. 8, 3048–3055 (2017). +12. Min, S. K., Agostini, F. & Gross, E. K. U. Coupled-Trajectory Quantum-Classical Approach to Electronic Decoherence in +Nonadiabatic Processes. Phys. Rev. Lett. 115, 073001, 10.1103/physrevlett.115.073001 (2015). +13. Gao, X. & Geva, E. Improving the Accuracy of Quasiclassical Mapping Hamiltonian Methods by Treating the Window +Function Width as an Adjustable Parameter. The J. Phys. Chem. A 124, 11006–11016, 10.1021/acs.jpca.0c09750 (2020). +14. Crespo-Otero, R. & Barbatti, M. Recent advances and perspectives on nonadiabatic mixed quantum–classical dynamics. +Chem. reviews 118, 7026–7068 (2018). +15. Subotnik, J. E. et al. Understanding the Surface Hopping View of Electronic Transitions and Decoherence. Annu. Rev. +Phys. Chem. 67, 387 – 417, 10.1146/annurev-physchem-040215-112245 (2016). +16. Wang, L., Akimov, A. & Prezhdo, O. V. Recent Progress in Surface Hopping: 2011–2015. The J. Phys. Chem. Lett. 7, +2100 – 2112, 10.1021/acs.jpclett.6b00710 (2016). +17. McLachlan, A. D. A variational solution of the time-dependent Schrodinger equation. Mol. Phys. 8, 39 – 44, 10.1080/ +00268976400100041 (2006). +18. Tully, J. C. Molecular dynamics with electronic transitions. The J. Chem. Phys. 93, 1061–1071, 10.1063/1.459170 (1990). +19. Shushkov, P., Li, R. & Tully, J. C. Ring polymer molecular dynamics with surface hopping. The J. Chem. Phys. 137, +22A549, 10.1063/1.4766449 (2012). +20. Huo, P., Miller, T. F. & Coker, D. F. Communication: Predictive partial linearized path integral simulation of condensed +phase electron transfer dynamics. The J. Chem. Phys. 139, 151103, 10.1063/1.4826163 (2013). +21. Kapral, R. & Ciccotti, G. Mixed quantum-classical dynamics. The J. Chem. Phys. 110, 8919–8929, 10.1063/1.478811 +(1999). +22. Miller, W. H. & Cotton, S. J. Classical molecular dynamics simulation of electronically non-adiabatic processes. Faraday +Discuss. 195, 9 – 30, 10.1039/c6fd00181e (2016). +23. Sun, X. & Geva, E. Equilibrium Fermi’s Golden Rule Charge Transfer Rate Constants in the Condensed Phase: The +Linearized Semiclassical Method vs Classical Marcus Theory. The J. Phys. Chem. A 120, 2976 – 2990, 10.1021/acs.jpca. +5b08280 (2016). +24. Chenu, A. & Scholes, G. D. Coherence in Energy Transfer and Photosynthesis. Annu. Rev. Phys. Chem. 66, 1–28, +10.1146/annurev-physchem-040214-121713 (2015). +25. Han, L. et al. Stochastic equation of motion approach to fermionic dissipative dynamics. i. formalism. The J. Chem. Phys. +152, 204105 (2020). +26. Ullah, A. et al. Stochastic equation of motion approach to fermionic dissipative dynamics. ii. numerical implementation. +The J. Chem. Phys. 152, 204106 (2020). +27. Yan, Y.-A., Zheng, X. & Shao, J. Piecewise ensemble averaging stochastic liouville equations for simulating non-markovian +quantum dynamics. New J. Phys. 24, 103012 (2022). +28. Chen, Z.-H., Wang, Y., Zheng, X., Xu, R.-X. & Yan, Y. Universal time-domain prony fitting decomposition for optimized +hierarchical quantum master equations. The J. Chem. Phys. 156, 221102 (2022). +29. Herrera Rodr’iguez, L. E. & Kananenka, A. A. Convolutional neural networks for long time dissipative quantum dynamics. +J. Phys. Chem. Lett. 12, 2476–2483, 10.1021/acs.jpclett.1c00079 (2021). https://doi.org/10.1021/acs.jpclett.1c00079. +30. Herrera, L. E., Ullah, A., Rueda, K. J., Dral, P. O. & Kananenka, A. A comparative study of different machine learning +methods for dissipative quantum dynamics. Mach. Learn. Sci. Technol. 10.1088/2632-2153/ac9a9d (2022). +31. Ullah, A. & Dral, P. O. One-shot trajectory learning of open quantum systems dynamics. The J. Phys. Chem. Lett. 13, +6037–6041 (2022). +32. Ullah, A. & Dral, P. O. Predicting the future of excitation energy transfer in light-harvesting complex with artificial +intelligence-based quantum dynamics. Nat. Commun. 13, 1–8 (2022). +33. Ullah, A. & Dral, P. O. Speeding up quantum dissipative dynamics of open systems with kernel methods. New J. Phys. +10.1088/1367-2630/ac3261 (2021). +8/15 + +34. Naicker, K., Sinayskiy, I. & Petruccione, F. Machine learning for excitation energy transfer dynamics. Phys. Rev. Res. 4, +033175, 10.1103/physrevresearch.4.033175 (2022). +35. Akimov, A. V. Extending the Time Scales of Nonadiabatic Molecular Dynamics via Machine Learning in the Time Domain. +J. Phys. Chem. Lett. 12, 12119–12128, 10.1021/acs.jpclett.1c03823 (2021). +36. Secor, M., Soudackov, A. V. & Hammes-Schiffer, S. Artificial Neural Networks as Propagators in Quantum Dynamics. J. +Phys. Chem. Lett. 12, 10654–10662, 10.1021/acs.jpclett.1c03117 (2021). +37. Banchi, L., Grant, E., Rocchetto, A. & Severini, S. Modelling non-markovian quantum processes with recurrent neural +networks. New J. Phys. 20, 123030, 10.1088/1367-2630/aaf749 (2018). +38. Bandyopadhyay, S., Huang, Z., Sun, K. & Zhao, Y. Applications of neural networks to the simulation of dynamics of open +quantum systems. Chem. Phys. 515, 272–278 (2018). +39. Yang, B., He, B., Wan, J., Kubal, S. & Zhao, Y. Applications of neural networks to dynamics simulation of Landau–Zener +transitions. Chem. Phys. 528, 110509, https://doi.org/10.1016/j.chemphys.2019.110509 (2020). +40. Wu, D., Hu, Z., Li, J. & Sun, X. Forecasting nonadiabatic dynamics using hybrid convolutional neural network/long +short-term memory network. J. Chem. Phys. 155, 224104 (2021). +41. Lin, K., Peng, J., Gu, F. L. & Lan, Z. Simulation of open quantum dynamics with bootstrap-based long short-term memory +recurrent neural network. J. Phys. Chem. Lett. 12, 10225–10234 (2021). +42. Tang, D., Jia, L., Shen, L. & Fang, W.-H. Fewest-switches surface hopping with long short-term memory networks. The J. +Phys. Chem. Lett. 13, 10377–10387, 10.1021/acs.jpclett.2c02299 (2022). PMID: 36317657, https://doi.org/10.1021/acs. +jpclett.2c02299. +43. Lin, K., Peng, J., Xu, C., Gu, F. L. & Lan, Z. Realization of the trajectory propagation in the mm-sqc dynamics by using +machine learning. arXiv preprint arXiv:2207.05556 (2022). +44. Lin, K., Peng, J., Xu, C., Gu, F. L. & Lan, Z. Automatic evolution of machine-learning-based quantum dynamics with +uncertainty analysis. J. Chem. Theory Comput. 18, 5837–5855, 10.1021/acs.jctc.2c00702 (2022). PMID: 36184823, +https://doi.org/10.1021/acs.jctc.2c00702. +45. Choi, M., Flam-Shepherd, D., Kyaw, T. H. & Aspuru-Guzik, A. Learning quantum dynamics with latent neural ordinary +differential equations. Phys. Rev. A 105, 042403, 10.1103/PhysRevA.105.042403 (2022). +46. Zhang, L., Ullah, A., Pinheiro Jr, M., Dral, P. O. & Barbatti, M. Excited-state dynamics with machine learning. In Quantum +Chemistry in the Age of Machine Learning, 329–353 (Elsevier, 2023). +47. Leggett, A. J. et al. Dynamics of the dissipative two-state system. Rev. Mod. Phys. 59, 1–85 (1987). +48. Weiss, U. Quantum Dissipative Systems. Series in modern condensed matter physics (World Scientific, 2012). +49. Adolphs, J. & Renger, T. How proteins trigger excitation energy transfer in the fmo complex of green sulfur bacteria. +Biophys. J. 91, 2778–2797, https://doi.org/10.1529/biophysj.105.079483 (2006). +50. Ishizaki, A. & Fleming, G. R. Theoretical examination of quantum coherence in a photosynthetic system at physiological +temperature. Proc. Natl. Acad. Sci. U.S.A. 106, 17255–17260, 10.1073/pnas.0908989106 (2009). https://www.pnas.org/ +content/106/41/17255.full.pdf. +51. Panitchayangkoon, G. et al. Long-lived quantum coherence in photosynthetic complexes at physiological temperature. +Proc. Natl. Acad. Sci. U.S.A. 107, 12766–12770, 10.1073/pnas.1005484107 (2010). https://www.pnas.org/content/107/29/ +12766.full.pdf. +52. Harush, E. Z. & Dubi, Y. Do photosynthetic complexes use quantum coherence to increase their efficiency? Probably not. +Sci. Adv. 7, eabc4631, 10.1126/sciadv.abc4631 (2021). https://www.science.org/doi/pdf/10.1126/sciadv.abc4631. +53. Ritschel, G., Roden, J., Strunz, W. T., Aspuru-Guzik, A. & Eisfeld, A. Absence of quantum oscillations and dependence +on site energies in electronic excitation transfer in the Fenna–Matthews–Olson trimer. J. Phys. Chem. Lett. 2, 2912–2917, +10.1021/jz201119j (2011). https://doi.org/10.1021/jz201119j. +54. Shim, S., Rebentrost, P., Valleau, S. & Aspuru-Guzik, A. Atomistic study of the long-lived quantum coherences in the +Fenna–Matthews–Olson complex. Biophys. J. 102, 649–660, https://doi.org/10.1016/j.bpj.2011.12.021 (2012). +55. Fenna, R. E. & Matthews, B. W. Chlorophyll arrangement in a bacteriochlorophyll protein from Chlorobium limicola. +Nature 258, 573–577, 10.1038/258573a0 (1975). +56. Milder, M. T. W., Brüggemann, B., Grondelle, R. v. & Herek, J. L. Revisiting the optical properties of the FMO protein. +Photosynth. Res. 104, 257–274, 10.1007/s11120-010-9540-1 (2010). +9/15 + +57. Engel, G. S. et al. Evidence for wavelike energy transfer through quantum coherence in photosynthetic systems. Nature +446, 782 (2007). +58. Scholes, G. D. et al. Using coherence to enhance function in chemical and biophysical systems. Nature 543, 647–656, +10.1038/nature21425 (2017). +59. Engel, G. S. Quantum coherence in photosynthesis. Procedia Chem. 3, 222–231, https://doi.org/10.1016/j.proche.2011.08. +029 (2011). 22nd Solvay Conference on Chemistry. +60. Renger, T. & May, V. Ultrafast exciton motion in photosynthetic antenna systems: the fmo-complex. The J. Phys. Chem. A +102, 4381–4391 (1998). +61. Louwe, R. J. W., Vrieze, J., Hoff, A. J. & Aartsma, T. J. Toward an integral interpretation of the optical steady-state +spectra of the fmo-complex of prosthecochloris aestuarii. 2. exciton simulations. The J. Phys. Chem. B 101, 11280–11287, +10.1021/jp9722162 (1997). https://doi.org/10.1021/jp9722162. +62. List, N. H., Curutchet, C., Knecht, S., Mennucci, B. & Kongsted, J. Toward reliable prediction of the energy ladder +in multichromophoric systems: A benchmark study on the fmo light-harvesting complex. J. Chem. Theory Comput. 9, +4928–4938, 10.1021/ct400560m (2013). PMID: 26583411, https://doi.org/10.1021/ct400560m. +63. Moix, J., Wu, J., Huo, P., Coker, D. & Cao, J. Efficient Energy Transfer in Light-Harvesting Systems, III: The Influence +of the Eighth Bacteriochlorophyll on the Dynamics and Efficiency in FMO. The J. Phys. Chem. Lett. 2, 3045–3052, +10.1021/jz201259v (2011). +64. Busch, M. S. a., Mü h, F., Madjet, M. E.-A. & Renger, T. The Eighth Bacteriochlorophyll Completes the Excitation Energy +Funnel in the FMO Protein. The J. Phys. Chem. Lett. 2, 93–98, 10.1021/jz101541b (2011). +65. Huang, R. Y.-C., Wen, J., Blankenship, R. E. & Gross, M. L. Hydrogen–deuterium exchange mass spectrometry reveals the +interaction of fenna–matthews–olson protein and chlorosome csma protein. Biochemistry 51, 187–193, 10.1021/bi201620y +(2012). PMID: 22142245, https://doi.org/10.1021/bi201620y. +66. Bina, D. & Blankenship, R. E. Chemical oxidation of the FMO antenna protein from Chlorobaculum tepidum. Photosynth. +Res. 116, 11–19, 10.1007/s11120-013-9878-2 (2013). +67. Olbrich, C. et al. From Atomistic Modeling to Excitation Transfer and Two-Dimensional Spectra of the FMO Light- +Harvesting Complex. The J. Phys. Chem. B 115, 8609–8621, 10.1021/jp202619a (2011). +68. Mühlbacher, L. & Kleinekathöfer, U. Preparational effects on the excitation energy transfer in the fmo complex. The J. +Phys. Chem. B 116, 3900–3906, 10.1021/jp301444q (2012). PMID: 22360690, https://doi.org/10.1021/jp301444q. +69. Tronrud, D. E., Wen, J., Gay, L. & Blankenship, R. E. The structural basis for the difference in absorbance spectra for +the FMO antenna protein from various green sulfur bacteria. Photosynth. Res. 100, 79–87, 10.1007/s11120-009-9430-6 +(2009). +70. Jia, X., Mei, Y., Zhang, J. Z. & Mo, Y. Hybrid QM/MM study of FMO complex with polarized protein-specific charge. Sci. +Reports 5, 17096, 10.1038/srep17096 (2015). +71. Shabani, A., Mohseni, M., Rabitz, H. & Lloyd, S. Efficient estimation of energy transfer efficiency in light-harvesting +complexes. Phys. Rev. E 86, 011915, 10.1103/PhysRevE.86.011915 (2012). +72. Wu, J., Liu, F., Shen, Y., Cao, J. & Silbey, R. J. Efficient energy transfer in light-harvesting systems, i: optimal temperature, +reorganization energy and spatial–temporal correlations. New J. Phys. 12, 105012, 10.1088/1367-2630/12/10/105012 +(2010). +73. Suzuki, Y., Watanabe, H., Okiyama, Y., Ebina, K. & Tanaka, S. Comparative study on model parameter evaluations for the +energy transfer dynamics in Fenna–Matthews–Olson complex. Chem. Phys. 539, 110903, 10.1016/j.chemphys.2020.110903 +(2020). +74. Mohseni, M., Shabani, A., Lloyd, S. & Rabitz, H. Energy-scales convergence for optimal and robust quantum transport in +photosynthetic complexes. The J. Chem. Phys. 140, 035102, 10.1063/1.4856795 (2014). 1104.4812. +75. Vulto, S. I. E. et al. Exciton Simulations of Optical Spectra of the FMO Complex from the Green Sulfur Bacterium +Chlorobium tepidum at 6 K. The J. Phys. Chem. B 102, 9577–9582, 10.1021/jp982095l (1998). +76. Cho, M., Vaswani, H. M., Brixner, T., Stenger, J. & Fleming, G. R. Exciton Analysis in 2D Electronic Spectroscopy. The J. +Phys. Chem. B 109, 10542–10556, 10.1021/jp050788d (2005). +77. Hayes, D. & Engel, G. Extracting the excitonic hamiltonian of the fenna-matthews-olson complex using three-dimensional +third-order electronic spectroscopy. Biophys. J. 100, 2043–2052, https://doi.org/10.1016/j.bpj.2010.12.3747 (2011). +10/15 + +78. Kell, A., Blankenship, R. E. & Jankowiak, R. Effect of spectral density shapes on the excitonic structure and dynamics of +the fenna–matthews–olson trimer from chlorobaculum tepidum. The J. Phys. Chem. A 120, 6146–6154, 10.1021/acs.jpca. +6b03107 (2016). PMID: 27438068, https://doi.org/10.1021/acs.jpca.6b03107. +79. Rolczynski, B. S. et al. Time-domain line-shape analysis from 2d spectroscopy to precisely determine hamiltonian +parameters for a photosynthetic complex. The J. Phys. Chem. B 125, 2812–2820, 10.1021/acs.jpcb.0c08012 (2021). PMID: +33728918, https://doi.org/10.1021/acs.jpcb.0c08012. +80. Ke, Y. & Zhao, Y. Hierarchy of forward-backward stochastic schrödinger equation. The J. Chem. Phys. 145, 024101, +10.1063/1.4955107 (2016). https://doi.org/10.1063/1.4955107. +81. Wilkins, D. M. & Dattani, N. S. Why quantum coherence is not important in the fenna–matthews–olsen complex. J. Chem. +Theory Comput. 11, 3411–3419, 10.1021/ct501066k (2015). PMID: 26575775, https://doi.org/10.1021/ct501066k. +82. Bourne Worster, S., Stross, C., Vaughan, F. M., Linden, N. & Manby, F. R. Structure and efficiency in bacterial +photosynthetic light harvesting. The J. Phys. Chem. Lett. 10, 7383–7390 (2019). +83. Abbott, J. W. Quantum dynamics of bath influenced excitonic energy transfer in photosynthetic pigment-protein complexes. +Master Thesis, University of Bristol United Kingdom https://doi.org/10.5281/zenodo.7229807 (2020). +84. Caldeira, A. & Leggett, A. Path integral approach to quantum Brownian motion. Phys. A: Stat. Mech. its Appl. 121, +587–616, 10.1016/0378-4371(83)90013-4 (1983). +85. Wang, H., Song, X., Chandler, D. & Miller, W. H. Semiclassical study of electronically nonadiabatic dynamics in the +condensed-phase: spin-boson problem with debye spectral density. J. Comp. Phys. 110, 4828 (1999). +86. Johansson, J., Nation, P. & Nori, F. Qutip: An open-source python framework for the dynamics of open quantum systems. +Comput. Phys. Commun. 183, 1760–1772, https://doi.org/10.1016/j.cpc.2012.02.021 (2012). +87. Dral, P. O. Mlatom: A program package for quantum chemical research assisted by machine learning. J. computational +chemistry 40, 2339–2347 (2019). +88. Abbott, J. W. jwa7/quantum_heom. Github repository: https://github.com/jwa7/quantum_HEOM, (accessed on Nov 1, +2022) (2019). +89. Jia, X., Mei, Y., Zhang, J. Z. & Mo, Y. Hybrid qm/mm study of fmo complex with polarized protein-specific charge. Sci. +reports 5, 1–10 (2015). +90. Strümpfer, J. & Schulten, K. Open Quantum Dynamics Calculations with the Hierarchy Equations of Motion on Parallel +Computers. J. Chem. Theory Comput. 8, 2808–2816, 10.1021/ct3003833 (2012). +91. Kreisbeck, C., Kramer, T., Rodr’iguez, M. & Hein, B. High-Performance Solution of Hierarchical Equations of Motion for +Studying Energy Transfer in Light-Harvesting Complexes. J. Chem. Theory Comput. 7, 2166–2174, 10.1021/ct200126d +(2011). 1012.4382. +92. Fehlberg, E. Some old and new Runge-Kutta formulas with stepsize control and their error coefficients. Computing 34, +265–270, 10.1007/bf02253322 (1985). +93. Oliphant, T. E. Python for scientific computing. Comput. science & engineering 9, 10–20 (2007). +Acknowledgements +A.A.K. acknowledges the Ralph E. Powe Junior Faculty Enhancement Award from Oak Ridge Associated Universities. This +work was also supported by General University Research (GUR) Grants and startup funds of the College of Arts and Sciences +and the Department of Physics and Astronomy of the University of Delaware. P.O.D. acknowledges funding by the National +Natural Science Foundation of China (No. 22003051 and funding via the Outstanding Youth Scholars (Overseas, 2021) project), +the Fundamental Research Funds for the Central Universities (No. 20720210092), and via the Lab project of the State Key +Laboratory of Physical Chemistry of Solid Surfaces. This research was supported in part through the use of Data Science +Institute (DSI) computational resources at the University of Delaware. Calculations were also performed with high-performance +computing resources provided by the Xiamen University. +Author contributions statement +P.O.D. and A.U. conceived the idea of creating a HEOM-based spin-boson database. A.U. conceived the idea of creating +LTLME-based database for FMO complex. A.A.K conceived the idea of creating an FMO dataset with the HEOM method. +A.U. performed the HEOM calculations for spin-boson along with the LTLME calculations for FMO complex. A.U. wrote the +11/15 + +provided package for easy extraction of the data. A.A.K and L.E.H.R. performed the calculations and created database files for +the FMO-VI data set. All authors analysed the results. A.A.K. took the lead in writing the original draft of the manuscript. All +authors reviewed and approved the manuscript. +Competing interests +The authors declare no competing interest. +Figures & Tables +0.0 +0.4 +0.8 +1.2 +1.6 +2.0 +Time [ps] +0.0 +0.2 +0.4 +0.6 +0.8 +1.0 +Population +1 +2 +3 +4 +5 +6 +7 +8 +0.0 +0.4 +0.8 +1.2 +1.6 +2.0 +Time [ps] +1 +2 +3 +4 +5 +6 +7 +8 +Figure 1. Example of the technical validation of the convergence of HEOM calculations for the following parameters T = 30 +K, λ = 70 cm−1, γ = 500 cm−1. The convergence within ∆ = 0.01 threshold was achieved for K = 7 and L = 4. Left plot +shows the populations obtained with K = 7 and L = 4 (solid lines) compared to populations obtained with K = 7 and L = 5 +(dashed lines) for all 8 sites. The largest population difference is δ = 3.14·10−4. The right plot shows the populations obtained +with K = 7 and L = 4 (solid lines) compared to the populations obtained with K = 8 and L = 4. The largest population +difference is δ = 4.62·10−3. In both cases the difference is very small illustrating the validity of the chosen threshold. +12/15 + +Data set +System +Hamiltonian(s) +Method +Data set +size +Cases +Propagation +time +(time-step) +Package +Parameter space +References +SB +SB +SB +HEOM +1000 +Symmetric +and +Asymmetric +20/∆ (0.05/∆) +QUTIP86 +E b +Regenerated +based on Ref. 33 +FMO-Ia +7-site FMO +Adolphs and +Renger49 +LTLME +Sites 1 and 6 +1 ns (5 fs) +QUANTUM_HEOM83,88 +F c +Regenerated +based on Ref. 32 +FMO-Ib +50 ps (5 fs) +MODIFIED- +QUANTUM_HEOM +a +G d +This work +FMO-II +Cho76 +FMO-III +8-site FMO +Jia70 +FMO-IV +Busch64 +and +Olbrich67 +1500 +Sites 1, 6 +and 8 +FMO-V +FMO trimer +FMO-VI +8-site FMO +HEOM +879 +Site 1 +2 ps (0.1 fs) +PHI90 +H e +Table 1. Summary of all data sets. See more details in the main text. Here “SB" stands for spin-boson model. +aMODIFIED-QUANTUM_HEOM is the QUANTUM_HEOM package with some local modifications to make it compatible for +larger Hamiltonians. bIn parameter space E , we define ˜ε = ε/∆ = {0,1}, ˜λ = λ/∆ = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, +1.0}, ˜γ = γ/∆ = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, and ˜β = β∆ = {0.1, 0.25, 0.5, 0.75, 1}, where the tunneling matrix element ∆ is +set as an energy unit. For all combinations of these parameters the system’s RDM was propagated. cIn parameter space F, we +choose the top 500 (most distant) combinations of (λ , γ, T) based on farthest-point sampling. Parameter range for each +dimension is λ = {10, 40, 70, ..., 310} cm−1, γ = {25, 50, 75, ..., 300} fs rad−1, and T = {30, 50, 70, ..., 310} K. dIn +parameter space G , we adopt the same procedure as in parameter space F and choose the most distant 500 points (based on +farthest-point sampling) from 3D space (λ , γ, T) where λ = {10, 40, 70, . . . , 520} cm−1, γ = {25, 50, 75, . . . , 500} cm−1, and +T = {30, 50, 70, . .., 510} K. eIn parameter space H , parameters range remains the same as in G . In addition, the same +farthest point sampling was adopted but with the only difference being that instead of 500, 1100 most distant set of parameters +were chosen. Approximately 20% of the initial data set was removed because of the prohibitive memory requirements. For the +remaining 80% of the data set, HEOM calculations were performed for 2.0 ps using 0.1 fs as a time step. +13/15 + +γ(cm−1) +100 +200 +300 +400 +500 +λ(cm−1) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +a) T= 30.0 K +γ(cm−1) +100 +200 +300 +400 +500 +λ(cm−1) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +b) T= 290.0 K +γ(cm−1) +100 +200 +300 +400 +500 +λ(cm−1) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +c) T= 510.0 K +γ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +d) λ = 10 cm−1 +γ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +e) λ = 310 cm−1 +γ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +f) λ = 510 cm−1 +λ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +g) γ = 25.0 cm−1 +λ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +h) γ = 300.0 cm−1 +λ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +i) γ = 500.0 cm−1 +0 +2 +4 +6 +8 +10 +Number of Matsubara terms +Figure 2. Number of Matsubara terms (Matsubara cut-off, M) required for converging populations of an 8-site FMO model +with the system Hamiltonian given by Eq. 14 for three selected temperatures (T), reorganization energies (λ), and bath cut-off +frequencies (γ). +14/15 + +γ(cm−1) +100 +200 +300 +400 +500 +λ(cm−1) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +a) T= 30.0 K +γ(cm−1) +100 +200 +300 +400 +500 +λ(cm−1) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +b) T= 290.0 K +γ(cm−1) +100 +200 +300 +400 +500 +λ(cm−1) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +c) T= 510.0 K +γ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +d) λ = 10 cm−1 +γ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +e) λ = 310 cm−1 +γ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +f) λ = 510 cm−1 +λ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +g) γ = 25.0 cm−1 +λ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +h) γ = 300.0 cm−1 +λ(cm−1) +100 +200 +300 +400 +500 +Temperature (K) +100 +200 +300 +400 +500 +0 +2 +4 +6 +8 +10 +12 +14 +i) γ = 500.0 cm−1 +0 +2 +4 +6 +8 +10 +12 +14 +Hierarchy truncation level +Figure 3. Hierarchy truncation level L required for converging populations of an 8-site FMO model with the system +Hamiltonian given by Eq. 14 for three selected temperatures (T), reorganization energies (λ), and bath cut-off frequencies (γ). +15/15 + diff --git a/ndFLT4oBgHgl3EQffi-_/content/tmp_files/load_file.txt b/ndFLT4oBgHgl3EQffi-_/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..6dc323ad8972033e057d53d30ba8576a4c135f4c --- /dev/null +++ b/ndFLT4oBgHgl3EQffi-_/content/tmp_files/load_file.txt @@ -0,0 +1,1702 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf,len=1701 +page_content='QD3SET-1: A Database with Quantum Dissipative Dynamics Data Sets Arif Ullah1, Luis E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Herrera Rodr´ıguez2, Pavlo O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Dral1,*, and Alexei A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kananenka2,* 1State Key Laboratory of Physical Chemistry of Solid Surfaces, Fujian Provincial Key Laboratory of Theoretical and Computational Chemistry, Department of Chemistry, and College of Chemistry and Chemical Engineering, Xiamen University, Xiamen 361005, China 2Department of Physics and Astronomy, University of Delaware, Newark, Delaware 19716, United States corresponding author(s): Pavlo O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Dral (dral@xmu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='cn) and Alexei A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kananenka (akanane@udel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='edu) ABSTRACT Simulations of the dynamics of dissipative quantum systems utilize many methods such as physics-based quantum, semiclassi- cal, and quantum-classical as well as machine learning-based approximations, development and testing of which requires diverse data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Here we present a new database QD3SET-1 containing eight data sets of quantum dynamical data for two systems of broad interest, spin-boson (SB) model and the Fenna–Matthews–Olson (FMO) complex, generated with two different methods solving the dynamics, approximate local thermalizing Lindblad master equation (LTLME) and highly accurate hierarchy equations of motion (HEOM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' One data set was generated with the SB model which is a two-level quantum system coupled to a harmonic environment using HEOM for 1,000 model parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Seven data sets were collected for the FMO complex of different sizes (7- and 8-site monomer and 24-site trimer with LTLME and 8-site monomer with HEOM) for 500–879 model parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Our QD3SET-1 database contains both population and coherence dynamics data and part of it has been already used for machine learning-based quantum dynamics studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Background & Summary The simulation of the inherently quantum-mechanical dynamics underlying charge, energy, and coherence transfer in the condensed-phase is one of the most difficult challenges for computational physics and chemistry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The exponential scaling of the computational cost with system size makes the quantum-mechanically exact simulations of such processes in complex systems infeasible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' With the exception of a few model Hamiltonians whose form makes the numerically exact quantum-dynamics simulations possible, any simulation of general condensed-phase systems must rely on approximations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1–28 Data-driven machine learning (ML) methods for quantum dynamics emerged as attractive alternative to the physics-based approximations due to their low computational cost and high accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='29–46 Development and testing of new simulation methodologies, both physics- and ML-based, would be greatly facilitated if high-quality reference quantum-dynamics data for a diverse set of quantum systems of interest were available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Here we present a QD3SET-1 database, a collection of eight data sets of time-evolved population dynamics of the two systems: spin-boson (SB) model and the Fenna–Matthews–Olson (FMO) light-harvesting complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The data sets are summarized in Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The SB model describes a (truncated or intrinsic) two-level quantum system linearly coupled to a harmonic bath.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='47,48 The physics of both the ground state and the dynamics of the SB model is very rich.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' It has been a continuous object of study during the past decades.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' SB has become a paradigmatic model system in the development of approximate quantum-dynamics methods and, nowadays, it is becoming a popular choice for the development of ML models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='30,33,40 The FMO system has become one of the most extensively studied natural light-harvesting complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='49–55 Under physio- logical conditions, the FMO complex forms a homotrimer consisting of eight bacteriochlorophyll-a (BChla) molecules per monomer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The biological function of the FMO trimer is to transfer excitation energy from the chlorosome to the reaction center (RC).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='56 An interest in this light-harvesting system sparked when two-dimensional electronic spectroscopy experiments detected the presence of quantum coherence effects in the FMO complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='57–59 These observations triggered intense debates about the role this coherence might play in the highly efficient excitation energy transfer (EET).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Early studies of the FMO complex considered only seven-site FMO models comprising of BChla 1–7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Until BChla 8 was discovered, BChla 1 and 6 were both assumed to be possible locations for accepting the excitation from the chlorosome because they are believed to be the nearest pigments to the antenna which captures the sunlight.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='60–62 From there, the energy is subsequently funneled through two nearly independent routes: from site 1 to 2 (pathway 1) or from site 6 to sites 7, 5, and 4 (pathway 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The terminal point of either route is site 3, where the exciton is then transferred to the RC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='63 Ever since the discovery of the eighth BChla, the role of this pigment in the EET has been extensively investigated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='53,63–70 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='12096v1 [physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='chem-ph] 28 Jan 2023 In particular, it was shown that while the population dynamics of the eight-site FMO model is markedly different from a seven-site configuration, the EET efficiencies in both models were predicted to remain comparable and very high.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='63 BChla 8 has also been suggested as possible recipient of the initial excitation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The dynamics of the FMO model has been a subject of numerous computational studies primarily focusing on understanding the role of the protein environment on the efficiency of EET (see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Refs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 71–74).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Numerical simulations typically employ one of the several parameterized or fitted into the experimental data FMO model Hamiltonians that differ in the BChla excitation energies and the couplings between different BChla sites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='49,56,75–79 Simulations of the full FMO trimer containing 24 BChl have also been performed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='80,81 Reported in this Data Descriptor a QD3SET-1 database contains seven data sets of time-evolved population dynamics of FMO models with different system Hamiltonians and initial excitations for several hundreds of bath and system-bath parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hierarchy of equations of motion (HEOM) approach5,7 was used to simulate the population dynamics of SB and FMO models, in one of the seven FMO data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' HEOM is a numerically exact method that can describe the dynamics of a system with a non-perturbative and non-Markovian system–bath interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The high computational cost of HEOM, however, limits the number of FMO simulations that can be performed with this method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' To generate other six FMO data sets, an approximate method—local thermalizing Lindblad master equation (LTLME)82,83 was used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Some of our data was already used in previous studies developing and benchmarking ML models for quantum-dynamics simulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='30–33 Here we regenerate one of the data sets to augment with more data and provide many new data sets generated from scratch (Table 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' To facilitate their use, we organize the data sets in a coherently formatted database and provided metadata and extraction scripts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' We expect our Database that accompany this Data Descriptor will serve as valuable resources in the development of new quantum-dynamics methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Methods SB data set This data set is re-generated with the same settings and the same parameters as in on our previous SB data set33 in order to include all the elements (populations and coherences) of the system’s reduced density matrix (RDM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The populations and population differences were published and used before.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='30,33 Below we provide a brief summary for self-containing presentation of the data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Spin-boson model The spin-boson model comprises a two-level quantum subsystem (TLS) coupled to a bath of harmonic oscillators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The Hamiltonian has the following standard system-bath form: ˆH = ˆHs + ˆHb + ˆHsb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The Hamiltonian of the TLS in the local (or site) basis {|+⟩,|−⟩} is given by (ℏ=1) ˆHs = ε (|+⟩⟨+|−|−⟩⟨−|) + ∆(|+⟩⟨−|+|−⟩⟨+|), (1) where ε is the so-called energetic bias and ∆ is the tunneling splitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The harmonic bath is an ensemble of independent harmonic oscillators ˆHb = Nb ∑ j=1 � ˆp2 j 2m j + 1 2m jω2 j ˆx2 j � , (2) where {ˆx j} and { ˆpj} are the coordinates and momenta, respectively, of Nb independent harmonic bath modes with masses {m j} and frequencies {ωj}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The TLS and bath are coupled through the additional term ˆHsb = − Nb ∑ j=1 cj ˆxj (|+⟩⟨+|−|−⟩⟨−|), (3) where {cj} are the coupling coefficients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The effects of the bath on the dynamics of TLS are collectively determined by the spectral density function84 J(ω) = π 2 Nb ∑ j=1 c2 j m jωj δ(ω −ωj).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' (4) In this work we choose to employ the Debye form of the spectral density (Ohmic spectral density with the Drude–Lorentz cut-off)85 J(ω) = 2λ ωγ ω2 +γ2 , (5) 2/15 where λ is the bath reorganization energy, which controls the strength of system-bath coupling, and the cutoff frequency γ = 1/τc (τc is the bath relaxation time).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' All dynamical properties of the TLS can be obtained from the RDM ˜ραβ(t) = Trb⟨α|e−i ˆHt/ℏ ˆρ(0)ei ˆHt/ℏ|β⟩, (6) where α,β ∈ {|+⟩,|−⟩}, ˆρ is the total density operator, and the trace is taken over bath degrees of freedom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For example, the commonly used in benchmark studies population difference is obtained from the RDM as follows: p+(t) − p−(t) = ˜ρ++(t)− ˜ρ−−(t).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The initial state of the total system is assumed to be a product state of the system and bath in the following form ˆρ(0) = ˆρs(0) ˆρb(0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' (7) In Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' (7) the bath density operator is an equilibrium canonical density operator ˆρb(0) = e−β ˆHb/Trb � e−β ˆHb � , where β = (kBT)−1 is the inverse temperature and kB is the Boltzmann constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The initial density operator of the system is chosen to be ˆρs(0) = |+⟩⟨+|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' These conditions corresponds to instantaneous photoexcitation of the subsystem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Data generation with spin-boson model and the hierarchy equations of motion approach The data set for the spin-boson model was generated as described previously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='33 We also summarize it below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The following system and bath parameters were chosen: ˜ε = ε/∆ = {0,1}, ˜λ = λ/∆ = {0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9,1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0}, ˜γ = γ/∆ = {1,2,3,4,5,6,7,8,9,10}, and ˜β = β∆ = {0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='25,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='75,1}, where the tunneling matrix element ∆ is set as an energy unit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For all combinations of these parameters the system’s RDM was propagated using HEOM approach implemented in QUTIP software package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='86 The total propagation time was tmax∆ = 20 and the HEOM integration time-step was set to dt∆ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In total, 1,000 of HEOM calculations, 500 for symmetric (ε/∆ = 0) and 500 for asymmetric (ε/∆ = 1) spin-boson Hamiltonian were performed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The data set contains a set of RDMs from t∆ = 0 to tmax∆ = 20, saved every dt∆ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='05, for every combination of the parameters (˜ε, ˜λ, ˜γ, ˜β) described above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Fenna–Matthews–Olson complex data sets In this section we first describe the general theory behind the FMO model Hamiltonian and later for each data set we provide specific technical details.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' See also Table 1 for an overview of each data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' FMO model Hamiltonian The FMO complex in this work is described by the system-bath Hamiltonian with the renormalization term ˆH = ˆHs + ˆHb + ˆHsb + ˆHren.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The electronic system is described by the Frenkel exciton Hamiltonian ˆHs = Ne ∑ n=1 En|n⟩⟨n|+ Ne ∑ n,m=1,n̸=m Vnm|n⟩⟨m|, (8) where |n⟩ denotes that only the nth site is in its electronically excited state and all other sites are in their electronically ground states, En are the transition energies, and Vnm is the Coulomb coupling between nth and mth sites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The couplings are assumed to be constant (the Condon approximation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Note that the overall electronic ground state of the pigment protein complex |0⟩ is assumed to be only radiatively coupled to the single-excitation manifold and as such it is not included in the dynamics calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In analogy with the SB model the bath is modeled by a set of independent harmonic oscillators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The thermal bath is coupled to the subsystem’s states |n⟩ through the system-bath interaction term ˆHsb = Ne ∑ n=1 Nb ∑ j=1 cn j ˆxj|n⟩⟨n|, (9) where each subsystem’s state is independently coupled to its own harmonic environment and cnj are the pigment-phonon coupling constants of environmental phonons local to the nth BChla.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The FMO model Hamiltonian contains a reorganization term which counters the shift in the minimum energy positions of harmonic oscillators introduced by the system-bath coupling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In the case that each state |n⟩ is independently coupled to the environment the renormalization term takes the following form ˆHren = Ne ∑ n=1 λn|n⟩⟨n|, (10) 3/15 where λn = ∑j c2 nj/(2m jω2 j ) is the bath reorganization energy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The bath spectral density associated with each electronic state is assumed to be given by the Lorentz–Drude spectral density (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Analogously to the SB data set the initial state of the total system is assumed to be a product state of the system and bath.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The initial electronic density operator given by ˆρs(0) was varied as described below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The bath density operator is taken to be the equilibrium canonical density operator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' FMO-Ia, FMO-Ib, and FMO-II data sets: 7-site FMO models with the local thermalizing Lindblad master equation ap- proach We generated data sets for the two 7-site system (Ne = 7) Hamiltonians.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' FMO-I data set was generated for the system Hamiltonian parameterized by Adolphs and Renger49 and given by (in cm−1) Hs = � � � � � � � � � � 200 −87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 −87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 320 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 0 −53.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −53.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 110 −70.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −63.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −70.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 270 81.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 −17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 81.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 420 39.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −63.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 39.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 230 � � � � � � � � � � , (11) FMO-Ia data set comes directly from our previous studies31,32 and FMO-Ib data set was generated here for a broader parameter space as described below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' FMO-II data set was generated for the Hamiltonian parameterized by Cho et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='76 which takes the following form (in cm−1) Hs = � � � � � � � � � � 280 −106 8 −5 6 −8 −4 −106 420 28 6 2 13 1 8 28 0 −62 −1 −9 17 −5 6 −62 175 −70 −19 −57 6 2 −1 −70 320 40 −2 −8 13 −9 −19 40 360 32 −4 1 17 −57 −2 32 260 � � � � � � � � � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' (12) The diagonal offset of 12210 cm−1 is added to both Hamiltonians.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Each site is coupled to its own bath characterized by the Drude–Lorentz spectral density, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 5, but the bath of each site is described by the same spectral density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For FMO-Ia data set, the following spectral density parameters and temperatures were employed: λ = {10, 40, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' , 310} cm−1, γ = {25, 50, 75, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', 300} fs rad−1, and T = {30, 50, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', 310} K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For FMO-Ib and FMO-II data sets, the spectral density parameters and temperatures were: λ = {10, 40, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' , 520} cm−1, γ = {25, 50, 75, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' , 500} cm−1, and T = {30, 50, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' , 510} K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For FMO-Ia, FMO-Ib, and FMO-II data sets, the farthest-point sampling87 was employed to select the most distant points in the Euclidean space32 of parameters which typically more efficiently covers relevant space compared to random sampling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='87 We choose the top 500 (most distant) combinations of (λ , γ, T) based on farthest-point sampling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For each selected set of parameters the system RDM was calculated using the local thermalizing Lindblad master equation (LTLME) approach82,83 implemented in the QUANTUM_HEOM package.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='83,88 Two subsets of the data set were generated, one for the initial electronic density operator ˆρs(0) = |1⟩⟨1| corresponding to the initial excitation of site-1 and the other one for the initial density operator ˆρs(0) = |6⟩⟨6| which corresponds to the initial excitation of site-6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In each case, 500 RDM trajectories were generated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The data set contains both diagonal (populations) and off-diagonal (coherences) elements of the RDM on a time grid from 0 to 1 ns (in the case of FMO-Ia) and 0 to 50 ps (in the case of FMO-Ib and FMO-II) with the 5 fs time step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 4/15 FMO-III and FMO-IV data sets: 8-site FMO models with the local thermalizing Lindblad master equation approach Using the same LTLME-based approach, we generated a data set for two different Hamiltonians for the 8-site FMO model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The first Hamiltonian (FMO-III data set) was parameterized by Jia et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='89 The electronic system Hamiltonian is given by (in cm−1) Hs = � � � � � � � � � � � � 218 −91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 81 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 0 −46.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 −4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −46.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 105 −73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 −17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −59.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 105 76.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 76.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 186 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 −11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 −7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −59.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 169 −11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 −11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 154 � � � � � � � � � � � � , (13) with the diagonal offset of 11332 cm−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The FMO-IV data set was generated for the Hamiltonian parameterized by Busch et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='64 (site energies) and Olbrich et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='67 (excitonic couplings) and takes the following form (in cm−1) Hs = � � � � � � � � � � � � 310 −80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 230 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 0 −49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 180 63.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 63.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 450 55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 320 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 270 −8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 505 � � � � � � � � � � � � , (14) with the diagonal offset of 12195 cm−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The same set of spectral density parameters and temperatures that was used in generation of the FMO-Ib and FMO-II data sets was used here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' LTLME method was used to propagate system’s RDM from 0 to 50 ps with 5 fs time-step and three initial states of the electronic system were considered: sites-1, 6 and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The data set contains both diagonal (populations) and off-diagonal (coherences) elements of the RDM.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The calculations was performed with the QUANTUM_HEOM package88 with some local modifications to make it compatible for the Hamiltonians with larger dimension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' We will refer to this as MODIFIED-QUANTUM_HEOM implementation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' FMO-V data set: FMO trimer with local thermalizing Lindblad master equation approach Additionally, we also generated a data set for the FMO trimer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The overall excitonic Hamiltonian of all three subunits is given by Hs = � � HA HB HT B HT B HA HB HB HT B HA � � (15) where HA is the subunit Hamiltonian for which we used the same Hamiltonian as in FMO-IV data set (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 14), while HB is the intra-subunit Hamiltonian which is taken from the work of Olbrich et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='67 and is given by (in cm−1) HB = � � � � � � � � � � � � 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7 −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5 −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 −3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 −5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 � � � � � � � � � � � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' (16) We propagate dynamics with LTLME from 0 to 50 ps with 5 fs time-step for the same parameters as was adopted in calculations for the FMO-Ib—FMO-IV data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The calculations were performed with the MODIFIED-QUANTUM_HEOM implementation for the initial excited sites-1, 6 and 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 5/15 FMO-VI data set: 8-site FMO model with the hierarchy of equations of motion approach The LTLME approach provides only approximate description of quantum dynamics of the FMO complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Therefore, the FMO-I—FMO-V data sets are useful merely for the developing machine learning models for quantum dynamics studies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For example, they can be used to train a neural network model which can then be further improved on more accurate but smaller data sets (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', via transfer learning).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' However, LTLME dynamics cannot be used to benchmark other quantum dynamics methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In the latter case the high-quality reference data is needed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' To generate a data set with accurate FMO dynamics we performed HEOM calculations for the 8-site FMO model with the Hamiltonian given by Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' HEOM calculations were performed using the parallel hierarchy integrator (PHI) code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='90 The initial data set was chosen on the basis of farthest-point sampling similar to how it was done in the FMO-Ib—FMO-V data sets with the only difference being that instead of 500 most distant sets of parameters that were chosen in the preparation of FMO-Ib—FMO-V data sets, 1100 most distant set of parameters were used to prepare the initial FMO-VI data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For certain parameters, the RAM requirements exceeded the RAM of computing nodes available to us (1 TB).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Therefore, such parameter sets were excluded from the data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Excluded parameters correspond to low temperatures, high reorganization energies, and low cut-off frequency.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Such strong non-Markovian regimes pose significant challenges in the computational studies of open quantum systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Approximately 20% of the initial data set was removed because of prohibitive memory requirements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' We note that even though graphics processing units (GPU) implementations of HEOM (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 91) are much faster than their CPU-based counterparts, they are still limited by the small amount of memory in presently available GPUs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For the remaining 80% of the data set HEOM calculations were performed for 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 ps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' To speed up calculations, an adaptive integration Runge–Kutta–Fehlberg 4/592 (RKF45) method was used as implemented in the PHI code.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Using adaptive integration reduces both the total computation time and memory requirements but can lead to artifacts if the accuracy threshold is set too large.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='90 In this work the PHI default accuracy threshold of 1·10−6 was used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The initial integration time step was set to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 fs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In RKF45 the integration time step is varied and, therefore, the output comprises time-evolved RDMs on an unevenly spaced time grid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' To obtain the RDMs on an evenly spaced time grid of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 fs, cubic-spline interpolation was used.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The interpolation errors were examined on a few cases where 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 fs fixed time step integration was feasible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The errors in the populations were found to be less than 10−5 which is much smaller compared to the convergence thresholds discussed below in Technical Validation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The final FMO-VI data set contains 879 entries each comprising all the populations and coherences for the RDM from 0 to 2 ps with the time-step of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 fs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Data Records All data sets can be accessed at https://figshare.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='com/s/ed24594205ab87404238.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The data sets are stored in standard NumPy93 binary file format (.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='npy) files.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The following format of file names was adopted in the SB data set 2_epsilon-X_Delta-1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0_lambda-Y_gamma-Z_beta-XX.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='npy, where X denotes the value of the energetic bias (˜ε), Y is the reorganization energy ˜λ, Z is the cut-off frequency ˜γ, and XX is the inverse temperature ˜β.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The following format of file names was adopted in all FMO data sets X_initial-Y_gamma-Z_lambda-XX_temp-YY.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='npy, where X denotes the number of sites in the FMO model, Y is the initial state, Z is the value of bath frequency, XX is the value of reorganization energy, and YY is the temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Technical Validation Central to the HEOM approach is the assumption that the bath correlation function Ca(t) for site a can be represented by an infinite sum of exponentially decaying terms Ca(t) = ∑∞ k cak exp(−νakt), where νak = 2πk/βℏ are Matsubara frequencies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Further, each exponential term leads to a set of auxiliary density matrices which take into account the non-Markovian evolution of the system’s RDM under the influence of bath.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In practice, the summation must be truncated at a finite level, K, which is called Matsubara cut-off and the set of auxiliary density matrices needs to be truncated at a finite number M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In the truncated set of auxiliary matrices are indexed by n = (n10,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=',n1K,nM0,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=',nMK).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The hierarchy truncation level is given by L = ∑M a=1 ∑K k=0 nak, where nak is the index of an auxiliary density matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The computational cost of the HEOM method rises steeply with the hierarchy level L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='90 The hierarchy truncation level L depends on how non-Markovian the system is.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Although, there is some guidance on how to choose the Matsubara cut-off and hierarchy truncation level based on bath and spectral density parameters,5,50 in practice, the values of M and K have to be chosen by requiring the convergence of the RDM to acceptable accuracy level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In this work HEOM calculations for the SB model were performed by setting L = 30 for all temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The Matsubara cut-off was chosen depending on the temperature as follows: for β = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 K was set to 2;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' for β = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='25,K = 3, for β = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5,K = 3, for β = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='75,K = 4, and for β = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0,K = 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' These values are chosen sufficiently high to ensure the convergence of the populations with respect to K and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Choosing high truncation levels in the HEOM calculations of a TLS does not present a problem given the presently available computational resources.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 6/15 Similar approach of taking excessively large values of K and L, however, is infeasible in the FMO calculations because the computational cost of HEOM grows steeply with the size of the quantum system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Therefore, the following approach was adopted for the HEOM calculations of the 8-site FMO model (FMO-VI data set).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Starting from K = 0 and L = 1, K was increased until the maximum difference in the populations between calculations with K and K +1 falls below a threshold ∆, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', δ = max n=1,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=',Nel t=0,.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=',tmax ����ρK,L n,n (t)−ρK+1,L n,n (t) ���� < ∆.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' (17) When Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='17 is satisfied for a given ∆ the convergence with respect to Matsubara cut-off is deemed to have been achieved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Then, for a fixed K a series of calculations were performed with increasing values of L until the maximum difference in populations between two consecutive calculations becomes less than the same threshold value ∆.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' When this condition is satisfied the convergence with respect to hierarchy truncation level as well as the overall convergence is declared.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' These steps were performed in the HEOM calculations for each parameter set for an 8-site FMO model until either the overall convergence is achieved or K and/or L become large enough so the calculation becomes intractable exceeding RAM available on our machines (1 TB).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In this work we set the threshold ∆ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' This threshold was chosen such that the population errors would be almost imperceptible which is illustrated in Figure 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' This data set is converged enough to be helpful in benchmarks of approximate methods describing quantum dynamics because the errors of these methods often exceed the threshold used in this work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Additionally, Figures 2 and 3 show the number of Matsubara terms and the hierarchy truncation level required for achieving the overall convergence depending on spectral density parameters and temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Usage Notes A Python package for extracting data is provided together with the data set and can be accessed at https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='com/Arif- PhyChem/QD3SET.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Code availability PHI code (version 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0) used in HEOM calculations was downloaded from http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='ks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='uiuc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='edu/Research/phi/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' QUTIP software package (version 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6) used in HEOM calculations of the spin-boson model and was downloaded from https://qutip.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' LTLME calculations of FMO models were performed with the basic QUANTUM_HEOM package https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='com/jwa7/quantum_HEOM and was modified to enable compatibility with the Hamiltonian of larger than the default dimension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' References 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Meyer, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Gatti, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Worth, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Multidimensional Quantum Dynamics: MCTDH Theory and Applications (Wiley, 2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Makri, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Time-dependent quantum methods for large systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Annu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 50, 167 – 191, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1146/annurev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' physchem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='167 (1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Meyer, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Manthe, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Cederbaum, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The multi-configurational time-dependent Hartree approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 165, 73 – 78, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/0009-2614(90)87014-I (1990).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Thoss, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Multilayer formulation of the multiconfiguration time-dependent hartree theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 119, 1289–1299 (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tanimura, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Numerically “exact” approach to open quantum dynamics: The hierarchical equations of motion (HEOM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 153, 020901, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0011599 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='05501.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tanimura, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Kubo, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Two-time correlation functions of a system coupled to a heat bath with a Gaussian–Markoffian interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Jpn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 58, 1199–1206 (1989).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tanimura, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Nonperturbative expansion method for a quantum system coupled to a harmonic-oscillator bath.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A 41, 6676–6687, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1103/PhysRevA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='41.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6676 (1990).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Greene, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Batista, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tensor-train split-operator fourier transform (tt-soft) method: Multidimensional nonadiabatic quantum dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 13, 4034–4042 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kapral, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PROGRESS IN THE THEORY OF MIXED QUANTUM-CLASSICAL DYNAMICS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Annu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 57, 129–157, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1146/annurev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='physchem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='032905.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='104702 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 7/15 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kapral, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Surface hopping from the perspective of quantum–classical Liouville dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 481, 77 – 83, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='chemphys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='016 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Min, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Agostini, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Tavernelli, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Gross, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ab initio nonadiabatic dynamics with coupled trajectories: A rigorous approach to quantum (de) coherence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 8, 3048–3055 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Min, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Agostini, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Gross, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Coupled-Trajectory Quantum-Classical Approach to Electronic Decoherence in Nonadiabatic Processes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 115, 073001, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1103/physrevlett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='115.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='073001 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Gao, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Geva, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Improving the Accuracy of Quasiclassical Mapping Hamiltonian Methods by Treating the Window Function Width as an Adjustable Parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A 124, 11006–11016, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0c09750 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Crespo-Otero, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Barbatti, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Recent advances and perspectives on nonadiabatic mixed quantum–classical dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' reviews 118, 7026–7068 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Subotnik, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Understanding the Surface Hopping View of Electronic Transitions and Decoherence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Annu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 67, 387 – 417, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1146/annurev-physchem-040215-112245 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Akimov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Prezhdo, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Recent Progress in Surface Hopping: 2011–2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 7, 2100 – 2112, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6b00710 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' McLachlan, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A variational solution of the time-dependent Schrodinger equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 8, 39 – 44, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1080/ 00268976400100041 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tully, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Molecular dynamics with electronic transitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 93, 1061–1071, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='459170 (1990).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Shushkov, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Li, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Tully, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ring polymer molecular dynamics with surface hopping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 137, 22A549, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4766449 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Huo, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Miller, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Coker, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Communication: Predictive partial linearized path integral simulation of condensed phase electron transfer dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 139, 151103, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4826163 (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kapral, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Ciccotti, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mixed quantum-classical dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 110, 8919–8929, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='478811 (1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Miller, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Cotton, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Classical molecular dynamics simulation of electronically non-adiabatic processes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Faraday Discuss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 195, 9 – 30, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1039/c6fd00181e (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sun, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Geva, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Equilibrium Fermi’s Golden Rule Charge Transfer Rate Constants in the Condensed Phase: The Linearized Semiclassical Method vs Classical Marcus Theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A 120, 2976 – 2990, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 5b08280 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chenu, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Scholes, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Coherence in Energy Transfer and Photosynthesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Annu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 66, 1–28, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1146/annurev-physchem-040214-121713 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Han, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Stochastic equation of motion approach to fermionic dissipative dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' formalism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 152, 204105 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ullah, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Stochastic equation of motion approach to fermionic dissipative dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' ii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' numerical implementation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 152, 204106 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Yan, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Zheng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Shao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Piecewise ensemble averaging stochastic liouville equations for simulating non-markovian quantum dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' New J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 24, 103012 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chen, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Wang, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Zheng, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Xu, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Yan, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Universal time-domain prony fitting decomposition for optimized hierarchical quantum master equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 156, 221102 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Herrera Rodr’iguez, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Kananenka, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Convolutional neural networks for long time dissipative quantum dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12, 2476–2483, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1c00079 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1c00079.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Herrera, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Ullah, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Rueda, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Dral, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Kananenka, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A comparative study of different machine learning methods for dissipative quantum dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Learn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Technol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1088/2632-2153/ac9a9d (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ullah, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Dral, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' One-shot trajectory learning of open quantum systems dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 13, 6037–6041 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ullah, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Dral, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Predicting the future of excitation energy transfer in light-harvesting complex with artificial intelligence-based quantum dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 13, 1–8 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ullah, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Dral, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Speeding up quantum dissipative dynamics of open systems with kernel methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' New J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1088/1367-2630/ac3261 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 8/15 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Naicker, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Sinayskiy, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Petruccione, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Machine learning for excitation energy transfer dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 4, 033175, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1103/physrevresearch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='033175 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Akimov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Extending the Time Scales of Nonadiabatic Molecular Dynamics via Machine Learning in the Time Domain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12, 12119–12128, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1c03823 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Secor, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Soudackov, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Hammes-Schiffer, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Artificial Neural Networks as Propagators in Quantum Dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12, 10654–10662, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1c03117 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Banchi, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Grant, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Rocchetto, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Severini, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Modelling non-markovian quantum processes with recurrent neural networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' New J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 20, 123030, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1088/1367-2630/aaf749 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 38.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Bandyopadhyay, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Huang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Sun, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Zhao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Applications of neural networks to the simulation of dynamics of open quantum systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 515, 272–278 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 39.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Yang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', He, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Wan, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Kubal, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Zhao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Applications of neural networks to dynamics simulation of Landau–Zener transitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 528, 110509, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='chemphys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='110509 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Wu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Hu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Li, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Sun, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Forecasting nonadiabatic dynamics using hybrid convolutional neural network/long short-term memory network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 155, 224104 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 41.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lin, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Peng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Gu, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Lan, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Simulation of open quantum dynamics with bootstrap-based long short-term memory recurrent neural network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12, 10225–10234 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tang, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Jia, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Shen, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Fang, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Fewest-switches surface hopping with long short-term memory networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 13, 10377–10387, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2c02299 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 36317657, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' jpclett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2c02299.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 43.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lin, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Peng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Xu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Gu, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Lan, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Realization of the trajectory propagation in the mm-sqc dynamics by using machine learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' arXiv preprint arXiv:2207.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='05556 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 44.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lin, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Peng, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Xu, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Gu, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Lan, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Automatic evolution of machine-learning-based quantum dynamics with uncertainty analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 18, 5837–5855, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jctc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2c00702 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 36184823, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jctc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2c00702.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 45.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Choi, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Flam-Shepherd, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Kyaw, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Aspuru-Guzik, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Learning quantum dynamics with latent neural ordinary differential equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A 105, 042403, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1103/PhysRevA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='105.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='042403 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 46.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Zhang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Ullah, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Pinheiro Jr, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Dral, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Barbatti, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Excited-state dynamics with machine learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In Quantum Chemistry in the Age of Machine Learning, 329–353 (Elsevier, 2023).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 47.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Leggett, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Dynamics of the dissipative two-state system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 59, 1–85 (1987).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Weiss, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Quantum Dissipative Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Series in modern condensed matter physics (World Scientific, 2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 49.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Adolphs, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Renger, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' How proteins trigger excitation energy transfer in the fmo complex of green sulfur bacteria.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Biophys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 91, 2778–2797, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1529/biophysj.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='105.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='079483 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ishizaki, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Fleming, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theoretical examination of quantum coherence in a photosynthetic system at physiological temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 106, 17255–17260, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0908989106 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/ content/106/41/17255.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='full.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='pdf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Panitchayangkoon, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Long-lived quantum coherence in photosynthetic complexes at physiological temperature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 107, 12766–12770, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1073/pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1005484107 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='pnas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/content/107/29/ 12766.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='full.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='pdf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 52.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Harush, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Dubi, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Do photosynthetic complexes use quantum coherence to increase their efficiency?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Probably not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 7, eabc4631, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1126/sciadv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='abc4631 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='science.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/doi/pdf/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1126/sciadv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='abc4631.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 53.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ritschel, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Roden, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Strunz, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Aspuru-Guzik, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Eisfeld, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Absence of quantum oscillations and dependence on site energies in electronic excitation transfer in the Fenna–Matthews–Olson trimer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 2, 2912–2917, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jz201119j (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jz201119j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 54.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Shim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Rebentrost, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Valleau, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Aspuru-Guzik, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Atomistic study of the long-lived quantum coherences in the Fenna–Matthews–Olson complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Biophys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 102, 649–660, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='bpj.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='021 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Fenna, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Matthews, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chlorophyll arrangement in a bacteriochlorophyll protein from Chlorobium limicola.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Nature 258, 573–577, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1038/258573a0 (1975).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 56.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Milder, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Brüggemann, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Grondelle, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Herek, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Revisiting the optical properties of the FMO protein.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Photosynth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 104, 257–274, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1007/s11120-010-9540-1 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 9/15 57.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Engel, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Evidence for wavelike energy transfer through quantum coherence in photosynthetic systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Nature 446, 782 (2007).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 58.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Scholes, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Using coherence to enhance function in chemical and biophysical systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Nature 543, 647–656, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1038/nature21425 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 59.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Engel, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Quantum coherence in photosynthesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Procedia Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 3, 222–231, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='proche.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='08.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 029 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 22nd Solvay Conference on Chemistry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 60.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Renger, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & May, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ultrafast exciton motion in photosynthetic antenna systems: the fmo-complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A 102, 4381–4391 (1998).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Louwe, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Vrieze, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Hoff, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Aartsma, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Toward an integral interpretation of the optical steady-state spectra of the fmo-complex of prosthecochloris aestuarii.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' exciton simulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' B 101, 11280–11287, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp9722162 (1997).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp9722162.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' List, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Curutchet, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Knecht, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Mennucci, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Kongsted, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Toward reliable prediction of the energy ladder in multichromophoric systems: A benchmark study on the fmo light-harvesting complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 9, 4928–4938, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/ct400560m (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 26583411, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/ct400560m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 63.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Moix, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Wu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Huo, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Coker, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Cao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Efficient Energy Transfer in Light-Harvesting Systems, III: The Influence of the Eighth Bacteriochlorophyll on the Dynamics and Efficiency in FMO.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 2, 3045–3052, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jz201259v (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 64.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Busch, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Mü h, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Madjet, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Renger, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The Eighth Bacteriochlorophyll Completes the Excitation Energy Funnel in the FMO Protein.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 2, 93–98, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jz101541b (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 65.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Huang, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Wen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Blankenship, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Gross, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hydrogen–deuterium exchange mass spectrometry reveals the interaction of fenna–matthews–olson protein and chlorosome csma protein.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Biochemistry 51, 187–193, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/bi201620y (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 22142245, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/bi201620y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 66.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Bina, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Blankenship, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chemical oxidation of the FMO antenna protein from Chlorobaculum tepidum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Photosynth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 116, 11–19, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1007/s11120-013-9878-2 (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 67.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Olbrich, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' From Atomistic Modeling to Excitation Transfer and Two-Dimensional Spectra of the FMO Light- Harvesting Complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' B 115, 8609–8621, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp202619a (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 68.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mühlbacher, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Kleinekathöfer, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Preparational effects on the excitation energy transfer in the fmo complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' B 116, 3900–3906, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp301444q (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 22360690, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp301444q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Tronrud, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Wen, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Gay, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Blankenship, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The structural basis for the difference in absorbance spectra for the FMO antenna protein from various green sulfur bacteria.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Photosynth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 100, 79–87, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1007/s11120-009-9430-6 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 70.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Jia, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Mei, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Mo, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hybrid QM/MM study of FMO complex with polarized protein-specific charge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Reports 5, 17096, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1038/srep17096 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Shabani, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Mohseni, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Rabitz, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Lloyd, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Efficient estimation of energy transfer efficiency in light-harvesting complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E 86, 011915, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1103/PhysRevE.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='86.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='011915 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 72.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Wu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Liu, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Shen, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Cao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Silbey, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Efficient energy transfer in light-harvesting systems, i: optimal temperature, reorganization energy and spatial–temporal correlations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' New J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12, 105012, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1088/1367-2630/12/10/105012 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Suzuki, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Watanabe, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Okiyama, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Ebina, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Tanaka, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Comparative study on model parameter evaluations for the energy transfer dynamics in Fenna–Matthews–Olson complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 539, 110903, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='chemphys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='110903 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mohseni, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Shabani, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Lloyd, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Rabitz, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Energy-scales convergence for optimal and robust quantum transport in photosynthetic complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 140, 035102, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4856795 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 1104.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4812.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Vulto, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Exciton Simulations of Optical Spectra of the FMO Complex from the Green Sulfur Bacterium Chlorobium tepidum at 6 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' B 102, 9577–9582, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp982095l (1998).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 76.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Cho, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Vaswani, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Brixner, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Stenger, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Fleming, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Exciton Analysis in 2D Electronic Spectroscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' B 109, 10542–10556, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/jp050788d (2005).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 77.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hayes, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Engel, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Extracting the excitonic hamiltonian of the fenna-matthews-olson complex using three-dimensional third-order electronic spectroscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Biophys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 100, 2043–2052, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='bpj.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3747 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 10/15 78.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kell, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Blankenship, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Jankowiak, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Effect of spectral density shapes on the excitonic structure and dynamics of the fenna–matthews–olson trimer from chlorobaculum tepidum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A 120, 6146–6154, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 6b03107 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 27438068, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpca.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6b03107.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Rolczynski, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Time-domain line-shape analysis from 2d spectroscopy to precisely determine hamiltonian parameters for a photosynthetic complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' B 125, 2812–2820, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpcb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0c08012 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 33728918, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/acs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='jpcb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0c08012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Ke, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Zhao, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hierarchy of forward-backward stochastic schrödinger equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 145, 024101, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4955107 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1063/1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4955107.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 81.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Wilkins, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Dattani, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Why quantum coherence is not important in the fenna–matthews–olsen complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 11, 3411–3419, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/ct501066k (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' PMID: 26575775, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/ct501066k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 82.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Bourne Worster, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Stross, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Vaughan, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Linden, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Manby, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Structure and efficiency in bacterial photosynthetic light harvesting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 10, 7383–7390 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Abbott, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Quantum dynamics of bath influenced excitonic energy transfer in photosynthetic pigment-protein complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Master Thesis, University of Bristol United Kingdom https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5281/zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7229807 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 84.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Caldeira, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Leggett, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Path integral approach to quantum Brownian motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A: Stat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mech.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' its Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 121, 587–616, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/0378-4371(83)90013-4 (1983).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 85.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Song, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Chandler, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Miller, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Semiclassical study of electronically nonadiabatic dynamics in the condensed-phase: spin-boson problem with debye spectral density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Comp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 110, 4828 (1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 86.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Johansson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Nation, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Nori, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Qutip: An open-source python framework for the dynamics of open quantum systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 183, 1760–1772, https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='cpc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='02.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='021 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Dral, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Mlatom: A program package for quantum chemical research assisted by machine learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' computational chemistry 40, 2339–2347 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 88.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Abbott, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' jwa7/quantum_heom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Github repository: https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='com/jwa7/quantum_HEOM, (accessed on Nov 1, 2022) (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 89.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Jia, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Mei, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Zhang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Mo, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hybrid qm/mm study of fmo complex with polarized protein-specific charge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' reports 5, 1–10 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 90.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Strümpfer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Schulten, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Open Quantum Dynamics Calculations with the Hierarchy Equations of Motion on Parallel Computers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 8, 2808–2816, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/ct3003833 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Kreisbeck, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Kramer, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', Rodr’iguez, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' & Hein, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' High-Performance Solution of Hierarchical Equations of Motion for Studying Energy Transfer in Light-Harvesting Complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 7, 2166–2174, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1021/ct200126d (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 1012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4382.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Fehlberg, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Some old and new Runge-Kutta formulas with stepsize control and their error coefficients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Computing 34, 265–270, 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1007/bf02253322 (1985).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 93.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Oliphant, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Python for scientific computing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' science & engineering 9, 10–20 (2007).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Acknowledgements A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' acknowledges the Ralph E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Powe Junior Faculty Enhancement Award from Oak Ridge Associated Universities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' This work was also supported by General University Research (GUR) Grants and startup funds of the College of Arts and Sciences and the Department of Physics and Astronomy of the University of Delaware.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' acknowledges funding by the National Natural Science Foundation of China (No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 22003051 and funding via the Outstanding Youth Scholars (Overseas, 2021) project), the Fundamental Research Funds for the Central Universities (No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 20720210092), and via the Lab project of the State Key Laboratory of Physical Chemistry of Solid Surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' This research was supported in part through the use of Data Science Institute (DSI) computational resources at the University of Delaware.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Calculations were also performed with high-performance computing resources provided by the Xiamen University.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Author contributions statement P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' conceived the idea of creating a HEOM-based spin-boson database.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' conceived the idea of creating LTLME-based database for FMO complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='K conceived the idea of creating an FMO dataset with the HEOM method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' performed the HEOM calculations for spin-boson along with the LTLME calculations for FMO complex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' wrote the 11/15 provided package for easy extraction of the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='K and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' performed the calculations and created database files for the FMO-VI data set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' All authors analysed the results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' took the lead in writing the original draft of the manuscript.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' All authors reviewed and approved the manuscript.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Competing interests The authors declare no competing interest.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Figures & Tables 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 Time [ps] 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 Population 1 2 3 4 5 6 7 8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 Time [ps] 1 2 3 4 5 6 7 8 Figure 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Example of the technical validation of the convergence of HEOM calculations for the following parameters T = 30 K, λ = 70 cm−1, γ = 500 cm−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The convergence within ∆ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='01 threshold was achieved for K = 7 and L = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Left plot shows the populations obtained with K = 7 and L = 4 (solid lines) compared to populations obtained with K = 7 and L = 5 (dashed lines) for all 8 sites.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The largest population difference is δ = 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='14·10−4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The right plot shows the populations obtained with K = 7 and L = 4 (solid lines) compared to the populations obtained with K = 8 and L = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' The largest population difference is δ = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='62·10−3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In both cases the difference is very small illustrating the validity of the chosen threshold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 12/15 Data set System Hamiltonian(s) Method Data set size Cases Propagation time (time-step) Package Parameter space References SB SB SB HEOM 1000 Symmetric and Asymmetric 20/∆ (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='05/∆) QUTIP86 E b Regenerated based on Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 33 FMO-Ia 7-site FMO Adolphs and Renger49 LTLME Sites 1 and 6 1 ns (5 fs) QUANTUM_HEOM83,88 F c Regenerated based on Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 32 FMO-Ib 50 ps (5 fs) MODIFIED- QUANTUM_HEOM a G d This work FMO-II Cho76 FMO-III 8-site FMO Jia70 FMO-IV Busch64 and Olbrich67 1500 Sites 1, 6 and 8 FMO-V FMO trimer FMO-VI 8-site FMO HEOM 879 Site 1 2 ps (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 fs) PHI90 H e Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Summary of all data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' See more details in the main text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Here “SB" stands for spin-boson model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' aMODIFIED-QUANTUM_HEOM is the QUANTUM_HEOM package with some local modifications to make it compatible for larger Hamiltonians.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' bIn parameter space E , we define ˜ε = ε/∆ = {0,1}, ˜λ = λ/∆ = {0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='2, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='3, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='4, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='6, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='7, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='8, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='9, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0}, ˜γ = γ/∆ = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, and ˜β = β∆ = {0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='25, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='5, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='75, 1}, where the tunneling matrix element ∆ is set as an energy unit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For all combinations of these parameters the system’s RDM was propagated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' cIn parameter space F, we choose the top 500 (most distant) combinations of (λ , γ, T) based on farthest-point sampling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Parameter range for each dimension is λ = {10, 40, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', 310} cm−1, γ = {25, 50, 75, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', 300} fs rad−1, and T = {30, 50, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=', 310} K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' dIn parameter space G , we adopt the same procedure as in parameter space F and choose the most distant 500 points (based on farthest-point sampling) from 3D space (λ , γ, T) where λ = {10, 40, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' , 520} cm−1, γ = {25, 50, 75, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' , 500} cm−1, and T = {30, 50, 70, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='., 510} K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' eIn parameter space H , parameters range remains the same as in G .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' In addition, the same farthest point sampling was adopted but with the only difference being that instead of 500, 1100 most distant set of parameters were chosen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Approximately 20% of the initial data set was removed because of the prohibitive memory requirements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' For the remaining 80% of the data set, HEOM calculations were performed for 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 ps using 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='1 fs as a time step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 13/15 γ(cm−1) 100 200 300 400 500 λ(cm−1) 100 200 300 400 500 0 2 4 6 8 10 a) T= 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 K γ(cm−1) 100 200 300 400 500 λ(cm−1) 100 200 300 400 500 0 2 4 6 8 10 b) T= 290.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 K γ(cm−1) 100 200 300 400 500 λ(cm−1) 100 200 300 400 500 0 2 4 6 8 10 c) T= 510.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 K γ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 d) λ = 10 cm−1 γ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 e) λ = 310 cm−1 γ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 f) λ = 510 cm−1 λ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 g) γ = 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 cm−1 λ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 h) γ = 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 cm−1 λ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 i) γ = 500.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 cm−1 0 2 4 6 8 10 Number of Matsubara terms Figure 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Number of Matsubara terms (Matsubara cut-off, M) required for converging populations of an 8-site FMO model with the system Hamiltonian given by Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 14 for three selected temperatures (T), reorganization energies (λ), and bath cut-off frequencies (γ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 14/15 γ(cm−1) 100 200 300 400 500 λ(cm−1) 100 200 300 400 500 0 2 4 6 8 10 12 14 a) T= 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 K γ(cm−1) 100 200 300 400 500 λ(cm−1) 100 200 300 400 500 0 2 4 6 8 10 12 14 b) T= 290.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 K γ(cm−1) 100 200 300 400 500 λ(cm−1) 100 200 300 400 500 0 2 4 6 8 10 12 14 c) T= 510.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 K γ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 12 14 d) λ = 10 cm−1 γ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 12 14 e) λ = 310 cm−1 γ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 12 14 f) λ = 510 cm−1 λ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 12 14 g) γ = 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 cm−1 λ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 12 14 h) γ = 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 cm−1 λ(cm−1) 100 200 300 400 500 Temperature (K) 100 200 300 400 500 0 2 4 6 8 10 12 14 i) γ = 500.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content='0 cm−1 0 2 4 6 8 10 12 14 Hierarchy truncation level Figure 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' Hierarchy truncation level L required for converging populations of an 8-site FMO model with the system Hamiltonian given by Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 14 for three selected temperatures (T), reorganization energies (λ), and bath cut-off frequencies (γ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} +page_content=' 15/15' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/ndFLT4oBgHgl3EQffi-_/content/2301.12096v1.pdf'} diff --git a/ndFQT4oBgHgl3EQfpzbH/vector_store/index.pkl b/ndFQT4oBgHgl3EQfpzbH/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..5a64e9eba76ecb2865e4e3be5988de6dadace7bf --- /dev/null +++ b/ndFQT4oBgHgl3EQfpzbH/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4af611a2b25cc69d33bca142a0edfb1637468cdfa62952325f87d277cec6a7ca +size 149949 diff --git a/rtAzT4oBgHgl3EQfrf2X/content/tmp_files/2301.01645v1.pdf.txt b/rtAzT4oBgHgl3EQfrf2X/content/tmp_files/2301.01645v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..17588f379a348b6bb21a143eabd1fabe04062b7d --- /dev/null +++ b/rtAzT4oBgHgl3EQfrf2X/content/tmp_files/2301.01645v1.pdf.txt @@ -0,0 +1,659 @@ +arXiv:2301.01645v1 [cond-mat.stat-mech] 4 Jan 2023 +epl draft +False Onsager relations +Michele Campisi +NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, I-56127 Pisa, Italy +PACS xxx +– +Abstract – Recent research suggests that when a system has a “false time reversal violation” +the Onsager reciprocity relations hold despite the presence of a magnetic field. The purpose of +this work is to clarify that the Onsager relations may well be violated in presence of a “false time +reversal violation”: that rather guarantees the validity of distinct relations, which we dub “false +Onsager relations”. We also point out that for quantum systems “false time reversal violation” is +omnipresent and comment that, per se, this has in general no consequence in regard to the validity +of Onsager relations, or the more general non-equilibrium fluctuation relations, in presence of a +magnetic field. +Our arguments are illustrated with the Heisenberg model of a magnet in an +external magnetic field. +Introduction. – +One open question in current ther- +modyamic research is whether a heat engine may achieve +Carnot efficiency while delivering finite power [1–6]. In a +seminal paper [1], Benenti et al. have established that, +within the framework of linear response theory, Carnot ef- +ficiency could be achieved by a thermoelectric device im- +mersed in a magneic field B provided the Seebeck coeffi- +cient is not even under the reversal of B: +S(B) ̸= S(−B) , +(1) +that is, +if the thermopower displays the so called +“Umkehreffekt”. +While the latter has been experimen- +tally validated in bismuth crystals [7], still the theoretical +understanding of the conditions under which such effect is +expected to appear is the subject of active research [8,9]. +Generally, the origin of its absence should be researched +in the presence of symmetries that prevent its appearance +[10–12]. In particular, it has been suggested [9,13,14] that +the presence of “false time reversal violations” results in +the Onsager reciprocity relations to be satisfied notwith- +standing the presence of a magnetic field, which would +forbid the “Umkehreffekt”. +Following Robnik and Berry [15], with the expression +“false time reversal violation” we denote the case where +the standard textbook “time reversal” (namely the trans- +formation that flips velocities and angular momenta, in- +cluding spins, see Eq. (4) below) is violated while some +other anticanonical (for classical system) or antiunitary +(for quantum system) symmetry is obeyed. All such trans- +formations, which we dub here “unconventional time re- +versals” have the property of inverting time, just like the +standard time reversal does, despite the possible pres- +ence of a magnetic field [15]. +As has been pointed out +in recent years, in a broad class of cases “unconventional +time reversal” symmetries exist, and it has been suggest- +ing that Onsager relations would not break in those cases +[9,13,14,16–18]. +Here we show that i) for any generic quantum system, +one could always find at least one “unconventional time +reversal” symmetry independent of the specific form of +the Hamiltonian, ii) the presence of “unconventional time +reversal” symmetries does not generally imply the valid- +ity of Onsager relations. We illustrate that with an exam- +ple of a quantum system featuring several “unconventional +time reversal” symmetries where the standard Onsager re- +lations are in fact violated. +We shall also remark that the omnipresence of “uncon- +ventional time reversals” does not imply the validity of the +non-equilibrium fluctuation relations [19] in a magnetic +field. We shall shed light onto the fact that, nonethelss, +that happens provided the “unconventional time reversal” +symmetry is one and the same at all times, which often +occurs in standard models studied in the literature. +Onsager, Onsager-Casimir, and false Onsager re- +lations. – +One of the cornerstones of non-equilibrium +thermodynamics are the Onsager relations (ORs) [20–22]. +They dictate that the matrix of phenomenological linear +response coefficients is composed of symmetric and an- +tisymmetric blocks, depending on the time-reversal par- +ity of the thermodynamic forces and fluxes they connect. +Within Kubo’s linear response theory [23], such relations +p-1 + +Michele Campisi +are expressed as +LAB(t) = ϑAϑBLBA(t) , +(2) +where LAB is the relaxation function of the quantity A, +caused by a perturbation of the quantity B. The quanti- +ties A and B in Eq. (2) have definite parity ϑA(B) under +time reversal, namely +Θ†AΘ = ϑAA, +Θ†BΘ = ϑBB , +(3) +where ϑA, ϑB = ±1 and +ΘσiΘ† = −σi, +ΘpiΘ† = −pi, +ΘqiΘ† = qi , +(4) +is the anti-unitary time reversal operator [24]. +In Kubo’s theory, the relaxation function reads, for a +quantum system +LAB(t) = +� β +0 +dsTr ρAH(−is)BH(t) − Tr ρAB , +(5) +where ρ = e−βH/Z is the thermal state, and OH(t) de- +notes the Heisenberg representation of operator O at time +t, that is +OH(t) = U † +t OUt , +(6) +where Ut = e−iHt/ℏ is the unitary time evolution operator +and H the system Hamiltonian. Under the assumption of +time-reversal invariance, i.e., +HΘ = ΘH , +(7) +we have +U−t = ΘUtΘ† , +(8) +which follows directly from the antiunitary character of +the operator Θ. The latter equation says that the appli- +cation of Θ causes the inversion of the time evolution. This +fact is often refreed to as the principle of microreversibility +[24]. Microreversibility combined with the standard rules +of quantum mechanics directly implies the validity of the +ORs Eq. (2). +Notably, in presence of a magnetic field B, the time- +reversal symmetry generally breaks, meaning that gen- +erally the ORs, Eq. +(2) are not valid in presence of a +magnetic field. However, generally, the less stringent “ex- +tended time reversal symmetry” survives +H(B)Θ = ΘH(−B) , +(9) +where we explicitly expressed the dependence of H on +B. The latter implies the following “extended” microre- +versibility principle +U−t(B) = Θ†Ut(−B)Θ , +(10) +where Ut(B) = e−iH(B)t/ℏ. It means that in order to re- +verse the dynamics one needs not only to reverse the mo- +menta and spins, but also the external field. The extended +microreversibility then ensures the validity of the cele- +brated Onsager-Casimir relations (OCRs) [22,23,25–28]: +LAB(t, B) = ϑAϑBLBA(t, −B) , +(11) +that link relaxation functions taken at opposite values of +B. Here the argument B is added to the function LAB to +denote that it refers to the time evolution, Ut(B), relative +to H(B). +Recent research has highlighted the interesting fact that +the dynamical evolution of a system (classical or quantum) +can be reversed in a number of different ways that differ +from the application of Θ, or, in case there is a magnetic +field, the joint reversal of B and the application of Θ [9,13, +14, 16–18]. Most notably, such transformations may well +not involve the reversal of the magnetic field. +For quantum systems, it is in fact straightforward to +note that if the Hamiltonian, H(B), is invariant under the +action of a generic antiunitary operator K, i.e., if there +exists an antiunitary K such that +H(B)K = KH(B) , +(12) +then +U−t(B) = KUt(B)K† , +(13) +meaning that any anti-unitary symmetry of the Hamilto- +nian realises an inversion of the time evolution. We shall +refer to this as an “unconventional time reversal”. A direct +aftermath of an “unconventional time reversal” symmetry +is a set of “false Onsager relations” (FORs) +LAB(t, B) = κAκBLBA(t, B) , +(14) +where A, B are operators with well defined parity under +K, that is +K†AK = κAA, +K†BK = κBB . +(15) +The proof follows exactly the same standard proof of the +OR’s, Eq. (2) with Θ being replaced by K. The relations +in Eq. (14) were first discoverd in Ref. [13] for classical sys- +tems, and then reported in Ref. [14,16] for special classes +of quantum systems. +At first sight one might confuse the FOR’s, Eq. (14), +for the OR’s, Eq. (2), in magnetic field. A closer scrutiny +however reveals that the two sets of relations substantially +differ one from the other in that they refer to quantities +A, B that have different symmetries, a fact that was typ- +ically overlooked so far in the literature. Generally, given +two quantities A and B that have definite parity under Θ +they might not have definite parity under K, or viceversa. +In those cases the validity of the FORs, Eq. (14), does say +nothing about the validity of the OR’s, Eq. (2). Vicev- +ersa, if the Hamiltonian has the K symmetry, Eq. (12), +two observables A and B have definite parity under both +Θ and K, and ϑAϑB = κAκB, then we have +LAB(t, B) = κAκBLBA(t, B) = ϑAϑB(t, B) , +(16) +p-2 + +False Onsager relations +namely, the ORs Eq. (2) would be valid despite the Hamil- +tonian does not have the Θ symmetry. However, that oc- +currence would only be accidental as the ORs, Eq. (2), +would generally be violated if only one looks at different +quantities A, B: hence our expression “false Onsager rela- +tions”. +We shall illustrate all this below with explicit examples +of interacting spin systems in magnetic field. +Omnipresence of false Onsager relations. – +For +quantum systems, the relations in Eq. (14) were first re- +ported in Ref. [16] for interacting spin-less particles in ho- +mogeneous magnetic field, while Ref. [14] reported them +for non-interacting particles with spin. Here we remark a +crucial fact, namely that at least one “unconventional time +reversal symmetry” can always be found in any quantum +system. To see that, recall that any antiunitary operator +can be expressed as the product of the complex conjuga- +tion operator KR relative to some representation R, and +a generic unitary V [24]. Since any unitary can be under- +stood as a change of basis operator, then any antiunitary +operator can be understood as the complex conjugation +operation relative to a specific representation. +Thus, if +there exist a representation R in which the Hamiltonian +is real, then the FORs, Eq. (14), would hold for oper- +ators having definite parity under the complex conjuga- +tion, KR, relative to that representation. Note that the +Hamiltonian is a quantum observable, namely a Hermi- +tian operator with real eigenvalues. Accordingly, in the +representation where H(B) is diagonal, the Hamiltonian +is trivially real. That is, irrespective of the specific form +of the Hamiltonian, the following always holds +H(B) = K† +H(B)H(B)KH(B) , +(17) +where KH(B) is the complex conjugation relative to the +representation where H(B) is diagonal. It follows that for +any couple of operators A and B that are either purely real +or purely imaginary in that representation, the FORs, Eq. +(14), hold. +For a sufficiently complex Hamiltonian this +may well have no consequence whatsoever in regard to +the validity of the ORs, Eq. (2), which refers to quan- +tities A, B with definite parity under Θ. This is further +illustrated below. +Examples. – +Consider an Heisenberg magnet in a +possibly non homogeneous field Bi = (Bx +i , By +i , Bz +i ), +H(B) = J +� +i,j +σi · σj − +� +i +Bi · σi , +(18) +where σi = (σx +i , σy +i , σz +i ), with σα +i denoting Pauli operators. +Since, by definition, all spin operators σα +i are odd under +time reversal Θ, Eq. (4), the Hamiltonian is not invariant +under time reversal Θ, while it is invariant under the joint +action of Θ, and the reversal of B, i.e., it obeys Eq. (9). +Accordingly the OCRs, Eq. (14) are obeyed. +Let us consider first the homogeneous case Bi = B. +Let’s fix the axes so that z is the direction of the applied +field B, B = Bzˆz, then: +H(B) = J +� +σi · σj − Bz +� +i +σz +i +(19) +In the representation where the tensor product ⊗iσz +i is +diagonal, the Hamiltonian is real. That is if K⊗σz +i is the +complex conjugation relative to that representation, it is +H(B) = K† +⊗σz +i H(B)K⊗σz +i . +(20) +In fact, in said representation, the Pauli matrices σz +i and +σx +i are real while the σy +i are imaginary. Thus the term +that couples to the applied magnetic field is real. +The +interaction term is also real because it is the sum of terms +of the type σα +i σα +j , which are real regardless of whether α +is z, x, or y. +Consider now the case when the external field is not +homogeneous in space Bi ̸= Bj for i ̸= j. If it only changes +in modulo, but it has a fixed direction, then the previous +argument will continue to apply unaltered. If its direction +also changes, but remains confined onto one fixed plane, +the argument still applies, with minimal changes. To see +that fix the axes so that the field has only components +along x and z, so that the Hamiltonian reads +H(B) = J +� +σi · σj − +� +(Bx +i σx +i + Bz +i σz +i ) . +(21) +Since in the representation of the basis spanned by the +eigenvectors of ⊗σz +i both the σx +i and the σz +i are all real, +and, as discussed above the interaction term is also real, +then the Hamiltonian is real as well in said representation, +that is, the unconventional time reversal symmetry of Eq. +(20) holds. +Accorodingly, the FORs, Eq. (14), hold for any cou- +ple of observables A, B that have definite parity under the +transformation K⊗σz +i . The spin operators σα +i all have def- +inite parities under K⊗σz +i : the σα +i are even for α = x, z, +and are odd for α = y. Considering that all the operators +σα +i are, by definition, odd under the time reversal Θ, it +follows, for example, that the ORs, Eq. (2) would hold +despite the presence of the magnetic field, for couples of +operators A = σα +i , B = σβ +j , with α and β being both y, or +both either x or z. However for couples of the type α = y, +β = x, z, the FOR, Eq. (14), reads +Lσy +i ,σz(x) +j +(t, B) = −Lσz(x) +j +,σy +i ,(t, B) , +(22) +thus violating the ORs Eq. (2), which would predict a +plus sign on the r.h.s. instead of a minus sign. This is +illustrateed in Fig. 1, top panel. Based on Eq. (22) it is +not hard to get convinced that the ORs, Eq. (2), would +not hold for A and B being spin operators (which are +odd under Θ) pointing in generic directions, i.e. for A = +ai · σi, B = bi · σi where ai, bi denote real unit vectors. +Consider now the general case where Bi is not confined +to a plane but explores all spatial directions as the spatial +index i is varied over all sites of the spin network. We +p-3 + +Michele Campisi +0 +1 +2 +3 +4 +5 +-0.2 +-0.1 +0.0 +0.1 +0.2 +t +B1 +z,B2 +x +�0 +A=σ1 +z +B=σ3 +y +0 +1 +2 +3 +4 +5 +-0.10 +-0.05 +0.00 +0.05 +t +B1 +z,B2 +x,B3 +y≠0 +A=σ1 +z +B=σ3 +y +LAB +LBA +Fig. 1: Breakdown of Onsager relations, Eq. (2). Top: Re- +laxation functions LAB and LBA for A = σz +1, B = σz +3 for +a 3 sites Heisenberg magnet, Eq. (18), with Bα +i = 0 except +Bz +1, Bx +2 = −2. +According to Eq. +(22) it is LAB = −LBA, +whereas the OR’s predicts LAB = LBA. Bottom: same as top +except for By +3, which now has the value By +3 = −2. The ORs, +Eq. (18), predicting LAB = LBA is not obeyed. In both plots +it is J = 1, β = 2, and ℏ = 1. +note that, at variance with the case above, in this case it +is generally not even possible to find an “unconventional +time reversal” symmetry besides the trivial invariance un- +der KH(B). So while in the previous case the OR’s Eq. +(2) remain valid for certain couples of observables, now +we expect them to be generally violated. Fig. 1, bottom +panel, shows an example of such violations. +Fixed field fluctuation relations. – +The presence +of an “unconventional time reversal” symmetry may as +well result in the validity of the nonequilibrium flutuation +relations [19] despite the presence of a magnetic field. At +variance with linear response relations, those relations ap- +ply arbitrarily far from equilibrium and refer to situations +where the Hamiltonian has an explicit time-dependence, +H = H(t, B). +Under the provision that there exist an +“unconventional time reversal” transformation, K , that +commutes with the Hamiltonian at all times during the +driving protocol, i.e., +KH(t, B) = H(t, B)K, +∀t ∈ [0, τ], +(23) +one would have, e.g., for a system prepared in an equi- +librium thermal state that evolves unitarily, the following +“fixed field fluctuation relation” for work +p(w, B) +˜p(−w, B) = e−β(w−∆F ), +(24) +featuring the same B in both the forward work statis- +tics, p(w, B) (obtained from the evolution generated by +H(t, B)), and the backward work statistics, ˜p(w, B) (ob- +tained from the evolution generated by H(τ − t, B), with +τ the duration of the time dependent driving). This is +at variance with the case customarily discussed in the lit- +erature whereby Eq. (9) is assumed to hold at all times +[19, 28, 29]) and the backward probability ˜p is taken at +reversed field −B . +The proof of Eq. (24) follows exactly the standard proof +(see appendices B, and C of Ref. [19]) with Θ being re- +placed by K. +Note that at variance with the Onsager +relations, that are possibly only apparently obeyed in case +of “false time reversal violation”, the fluctuation relation +p(w)/p(−w) = e−β(w−∆F ) is simply obeyed in its standard +form, despite the magnetic field, when Eq. (23) holds. +The validity of Eq. (24) was often observed in the lit- +erature, see e.g., [30–32], although the issue related to +the presence of a magnetic field and the according lack +of microreversibility (which would have ensured its valid- +ity) typically passed unnoticed. Needless to say, all those +previous works considered situations where the Hamilto- +nian was real in some representation, at all times, which +ensured the validity of Eq. (24), a fact that instead was +only acknowledged and discussed in Ref. [32]. +In this regard it is also worth remarking that Eq. (13) +would instantaneously hold at each time, t, with an “un- +conventional time reversal” KH(t,B), that possibly changes +in time. Accordingly, its ubiquitous validity does not en- +sure the validity of Eq. (24) featuring a fixed K, Eq. (23). +It is a simple exercise to show that, even for a single spin +in a time dependent magnetic field H(t, B) = −B(t) · σ, +Eq. (24) does not hold when B(t) explores all three spatial +dimensions [33]. +Discussion. – +We have established two main facts. +First, for quantum systems a trivial “unconventional time +reversal” always exist, and that is the symmetry un- +der complex conjugation in the representation where the +Hamiltonian is diagonal. This very same fact was already +noticed by Robnik and Berry [15] in the context of level +statistics in quantum billiards. Second, in contrast with +what previous research suggested, the presence of one or +more “unconventional time reversals” does not by itself +guarantee the validity of the Onsager relations, Eq. (2), +when a magnetic field is present. +As illustrated by ex- +amples, they indeed can be violated, however if couples +of observables exist that have definite parities under both +p-4 + +False Onsager relations +time reversal and an “unconventional time reversal” which +is a symmetry of the problem, the ORs will be obeyed for +them, provided the products of their parities under time +reversal is the same as that under the “unconventional +time reversal”. +Roughly speaking, the more “unconventional time re- +versal symmetries” a system has, the larger the set of cou- +ples of observables that obey the ORs, Eq. (2), despite a +magnetic field. On the contrary when the only “unconven- +tional time reversal” is the trivial one, Eq. (17) one should +generally expect the ORs Eq. (2) to be violated. The sit- +uation is somewhat similar to dynamical system theory, +where we have two extremes: full integrability (as many +conserved quantities as are the degrees of freedom), and +full ergodicity (the Hamiltonian is the only first integral +of motion). Between these two extremes lie complex sys- +tems displaying both regular and irregular motion. Here, +similarly, we have the case where there are as many “un- +conventional time reversal symmetries” as are the degrees +of freedom 1, the case when only the trivial one exist, and +the complex situation in between featuring both violation +and obedience of the ORs, Eq. (2). +Since, as illustrated by our examples, the Onsager rela- +tions may be violated despite the presence of “false time +reversal violations”, our results leave open the possibility +of the “Umkehreffekt”, which in fact, as mentioned above, +has been experimentally observed. Thus the present study +does not provide any fundamental reason to exclude the +possibility of achieving Carnot efficiency at finite power, +in the way discussed in Ref. [1]. +We further have commented that the omnipresence of +“false time reversal violations” does not, per se, imply the +ubiquitous validity of quantum fluctuation relations, e.g., +the work fluctuation relation, in presence of a magnetic +field. The latter would hold provided the “unconventional +time reversal symmetry” is one and the same during the +whole driving protocol. While many models of many-body +systems that are customarily studied in the literature (e.g., +the driven Ising model in transverse field) satisfy that re- +quirement that is generally not the case. +A question that remains to be answered is whether +“false time reversal violation” is omnipresent as well in +the classical case, and how can one construct the accord- +ing “unconventional time reversal” transformation. While +this issue was easily addressed in the quantum case, the +question does not seem to admit a simple answer for clas- +sical systems. Addressing that might reveal a new discor- +dance between classical and quantum realms 2. +Acknowledgements. – +I would like to thank Giu- +liano Benenti, Lamberto Rondoni and Davide Carbone for +1This would occur, for example, in the case of a set of non- +interacting spins with local magnetic field. In that case all relaxation +functions are null and the ORs are trivially obeyed for any couple of +observables. +2One such discordance whereby deterministic friction appears in +the classical realm but not in the quantum one was reported by Berry +and Robbins [34] +pointing out relevant literature and for the stimulating dis- +cussions that were instrumental to shaping this work. +REFERENCES +[1] Benenti G., Saito K. and Casati G., Phys. Rev. Lett., +106 (2011) 230602. +[2] Allahverdyan A. E., Hovhannisyan K. V., Melkikh +A. V. and Gevorkian S. G., Phys. Rev. Lett., 111 (2013) +050601. +[3] Campisi M. and Fazio R., Nat Commun, 7 (2016) 11895. +[4] Holubec V. and Ryabov A., Phys. Rev. Lett., 121 +(2018) 120601. +[5] Bonanc¸a M. V. S., Journal of Statistical Mechanics: +Theory and Experiment, 2019 (2019) 123203. +[6] Miura K., Izumida Y. and Okuda K., Phys. Rev. E, +105 (2022) 034102. +[7] Wolfe R. and Smith G., Phys. Rev., 129 (1963) . +[8] Saito K., Benenti G., Casati G. and Prosen T. c. v., +Phys. Rev. B, 84 (2011) 201306. +[9] Luo R., Benenti G., Casati G. and Wang J., Phys. +Rev. Research, 2 (2020) 022009. +[10] Gabrielli D., Jona-Lasinio G. and Landim C., Phys. +Rev. Lett., 77 (1996) 1202. +[11] Lebowitz J. L. and Spohn H., Phys. Rev. Lett., 78 +(1997) 394. +[12] Gabrielli D., Jona-Lasinio G. and Landim C., Jour- +nal of Statistical Physics, 96 (1999) 639. +[13] Bonella S., Ciccotti G. and Rondoni L., EPL (Eu- +rophysics Letters), 108 (2014) 60004. +[14] Carbone D., De Gregorio P. and Rondoni L., Annals +of Physics, 441 (2022) 168853. +[15] Robnik M. and Berry M. V., Journal of Physics A: +Mathematical and General, 19 (1986) 669. +[16] Gregorio P. D., Bonella S. and Rondoni L., Sym- +metry, 9 (2017) . +[17] Coretti A., Bonella S., Rondoni L. and Ciccotti +G., Molecular Physics, 116 (2018) 3097. +[18] Coretti A., Rondoni L. and Bonella S., Phys. Rev. +E, 102 (2020) 030101. +[19] Campisi M., H¨anggi P. and Talkner P., Rev. Mod. +Phys., 83 (2011) 771. +[20] Onsager L., Phys. Rev., 37 (1931) 405. +[21] Onsager L., Phys. Rev., 38 (1931) 2265. +[22] de Groot S. R. and Mazur P., Non-equilibrium Ther- +modynamics (Dover, New York) 1984. +[23] Kubo R., J. Phys. Soc. Jpn., 12 (1957) 570. +[24] Messiah A., Quantum Mechanics (North Holland, Ams- +terdam) 1962. +[25] Casimir H. B. G., Rev. Mod. Phys., 17 (1945) 343. +[26] Bochkov G. N. and Kuzovlev Y. E., Zh. Eksp. Teor. +Fiz., 76 (1979) 1071 [Sov. Phys. JETP 49, 543–551 +(1979)]. +[27] Bochkov G. N. and Kuzovlev Y. E., Zh. Eksp. Teor. +Fiz., 72 (1977) 238 [Sov. Phys. JETP 45, 125–130 (1977)]. +[28] Andrieux D. and Gaspard P., Phys. Rev. Lett., 100 +(2008) 230404. +[29] Jarzynski C. and W´ojcik D. K., Phys. Rev. Lett., 92 +(2004) 230602. +[30] Dorner R., Goold J., Cormick C., Paternostro M. +and Vedral V., Phys. Rev. Lett., 109 (2012) 160601. +p-5 + +Michele Campisi +[31] Fusco L., Pigeon S., Apollaro T. J. G., Xuereb A., +Mazzola L., Campisi M., Ferraro A., Paternostro +M. and De Chiara G., Phys. Rev. X, 4 (2014) 031029. +[32] Schmidtke +D., +Knipschild +L., +Campisi +M., +Steinigeweg +R. +and +Gemmer +J., +Phys. +Rev. +E, +98 (2018) 012123. +[33] Orlandini M., BSc. Thesis, University of Florence, +(2022) . +[34] Berry M. V. and Robbins J. M., Proc. R. Soc. London, +Ser. A442, 1993 (659) . +p-6 + diff --git a/rtAzT4oBgHgl3EQfrf2X/content/tmp_files/load_file.txt b/rtAzT4oBgHgl3EQfrf2X/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..555a623a8add3427055122048fe2806d20d220fb --- /dev/null +++ b/rtAzT4oBgHgl3EQfrf2X/content/tmp_files/load_file.txt @@ -0,0 +1,433 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf,len=432 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='01645v1 [cond-mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='stat-mech] 4 Jan 2023 epl draft False Onsager relations Michele Campisi NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, I-56127 Pisa, Italy PACS xxx – Abstract – Recent research suggests that when a system has a “false time reversal violation” the Onsager reciprocity relations hold despite the presence of a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The purpose of this work is to clarify that the Onsager relations may well be violated in presence of a “false time reversal violation”: that rather guarantees the validity of distinct relations, which we dub “false Onsager relations”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We also point out that for quantum systems “false time reversal violation” is omnipresent and comment that, per se, this has in general no consequence in regard to the validity of Onsager relations, or the more general non-equilibrium fluctuation relations, in presence of a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Our arguments are illustrated with the Heisenberg model of a magnet in an external magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Introduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – One open question in current ther- modyamic research is whether a heat engine may achieve Carnot efficiency while delivering finite power [1–6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In a seminal paper [1], Benenti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' have established that, within the framework of linear response theory, Carnot ef- ficiency could be achieved by a thermoelectric device im- mersed in a magneic field B provided the Seebeck coeffi- cient is not even under the reversal of B: S(B) ̸= S(−B) , (1) that is, if the thermopower displays the so called “Umkehreffekt”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' While the latter has been experimen- tally validated in bismuth crystals [7], still the theoretical understanding of the conditions under which such effect is expected to appear is the subject of active research [8,9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Generally, the origin of its absence should be researched in the presence of symmetries that prevent its appearance [10–12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In particular, it has been suggested [9,13,14] that the presence of “false time reversal violations” results in the Onsager reciprocity relations to be satisfied notwith- standing the presence of a magnetic field, which would forbid the “Umkehreffekt”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Following Robnik and Berry [15], with the expression “false time reversal violation” we denote the case where the standard textbook “time reversal” (namely the trans- formation that flips velocities and angular momenta, in- cluding spins, see Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (4) below) is violated while some other anticanonical (for classical system) or antiunitary (for quantum system) symmetry is obeyed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' All such trans- formations, which we dub here “unconventional time re- versals” have the property of inverting time, just like the standard time reversal does, despite the possible pres- ence of a magnetic field [15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' As has been pointed out in recent years, in a broad class of cases “unconventional time reversal” symmetries exist, and it has been suggest- ing that Onsager relations would not break in those cases [9,13,14,16–18].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Here we show that i) for any generic quantum system, one could always find at least one “unconventional time reversal” symmetry independent of the specific form of the Hamiltonian, ii) the presence of “unconventional time reversal” symmetries does not generally imply the valid- ity of Onsager relations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We illustrate that with an exam- ple of a quantum system featuring several “unconventional time reversal” symmetries where the standard Onsager re- lations are in fact violated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We shall also remark that the omnipresence of “uncon- ventional time reversals” does not imply the validity of the non-equilibrium fluctuation relations [19] in a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We shall shed light onto the fact that, nonethelss, that happens provided the “unconventional time reversal” symmetry is one and the same at all times, which often occurs in standard models studied in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Onsager, Onsager-Casimir, and false Onsager re- lations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – One of the cornerstones of non-equilibrium thermodynamics are the Onsager relations (ORs) [20–22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' They dictate that the matrix of phenomenological linear response coefficients is composed of symmetric and an- tisymmetric blocks, depending on the time-reversal par- ity of the thermodynamic forces and fluxes they connect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Within Kubo’s linear response theory [23], such relations p-1 Michele Campisi are expressed as LAB(t) = ϑAϑBLBA(t) , (2) where LAB is the relaxation function of the quantity A, caused by a perturbation of the quantity B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The quanti- ties A and B in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) have definite parity ϑA(B) under time reversal, namely Θ†AΘ = ϑAA, Θ†BΘ = ϑBB , (3) where ϑA, ϑB = ±1 and ΘσiΘ† = −σi, ΘpiΘ† = −pi, ΘqiΘ† = qi , (4) is the anti-unitary time reversal operator [24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In Kubo’s theory, the relaxation function reads, for a quantum system LAB(t) = � β 0 dsTr ρAH(−is)BH(t) − Tr ρAB , (5) where ρ = e−βH/Z is the thermal state, and OH(t) de- notes the Heisenberg representation of operator O at time t, that is OH(t) = U † t OUt , (6) where Ut = e−iHt/ℏ is the unitary time evolution operator and H the system Hamiltonian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Under the assumption of time-reversal invariance, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', HΘ = ΘH , (7) we have U−t = ΘUtΘ† , (8) which follows directly from the antiunitary character of the operator Θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The latter equation says that the appli- cation of Θ causes the inversion of the time evolution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' This fact is often refreed to as the principle of microreversibility [24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Microreversibility combined with the standard rules of quantum mechanics directly implies the validity of the ORs Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Notably, in presence of a magnetic field B, the time- reversal symmetry generally breaks, meaning that gen- erally the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) are not valid in presence of a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' However, generally, the less stringent “ex- tended time reversal symmetry” survives H(B)Θ = ΘH(−B) , (9) where we explicitly expressed the dependence of H on B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The latter implies the following “extended” microre- versibility principle U−t(B) = Θ†Ut(−B)Θ , (10) where Ut(B) = e−iH(B)t/ℏ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' It means that in order to re- verse the dynamics one needs not only to reverse the mo- menta and spins, but also the external field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The extended microreversibility then ensures the validity of the cele- brated Onsager-Casimir relations (OCRs) [22,23,25–28]: LAB(t, B) = ϑAϑBLBA(t, −B) , (11) that link relaxation functions taken at opposite values of B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Here the argument B is added to the function LAB to denote that it refers to the time evolution, Ut(B), relative to H(B).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Recent research has highlighted the interesting fact that the dynamical evolution of a system (classical or quantum) can be reversed in a number of different ways that differ from the application of Θ, or, in case there is a magnetic field, the joint reversal of B and the application of Θ [9,13, 14, 16–18].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Most notably, such transformations may well not involve the reversal of the magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' For quantum systems, it is in fact straightforward to note that if the Hamiltonian, H(B), is invariant under the action of a generic antiunitary operator K, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', if there exists an antiunitary K such that H(B)K = KH(B) , (12) then U−t(B) = KUt(B)K† , (13) meaning that any anti-unitary symmetry of the Hamilto- nian realises an inversion of the time evolution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We shall refer to this as an “unconventional time reversal”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' A direct aftermath of an “unconventional time reversal” symmetry is a set of “false Onsager relations” (FORs) LAB(t, B) = κAκBLBA(t, B) , (14) where A, B are operators with well defined parity under K, that is K†AK = κAA, K†BK = κBB .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (15) The proof follows exactly the same standard proof of the OR’s, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) with Θ being replaced by K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The relations in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14) were first discoverd in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [13] for classical sys- tems, and then reported in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [14,16] for special classes of quantum systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' At first sight one might confuse the FOR’s, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14), for the OR’s, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), in magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' A closer scrutiny however reveals that the two sets of relations substantially differ one from the other in that they refer to quantities A, B that have different symmetries, a fact that was typ- ically overlooked so far in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Generally, given two quantities A and B that have definite parity under Θ they might not have definite parity under K, or viceversa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In those cases the validity of the FORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14), does say nothing about the validity of the OR’s, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Vicev- ersa, if the Hamiltonian has the K symmetry, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (12), two observables A and B have definite parity under both Θ and K, and ϑAϑB = κAκB, then we have LAB(t, B) = κAκBLBA(t, B) = ϑAϑB(t, B) , (16) p-2 False Onsager relations namely, the ORs Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) would be valid despite the Hamil- tonian does not have the Θ symmetry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' However, that oc- currence would only be accidental as the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), would generally be violated if only one looks at different quantities A, B: hence our expression “false Onsager rela- tions”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We shall illustrate all this below with explicit examples of interacting spin systems in magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Omnipresence of false Onsager relations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – For quantum systems, the relations in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14) were first re- ported in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [16] for interacting spin-less particles in ho- mogeneous magnetic field, while Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [14] reported them for non-interacting particles with spin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Here we remark a crucial fact, namely that at least one “unconventional time reversal symmetry” can always be found in any quantum system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' To see that, recall that any antiunitary operator can be expressed as the product of the complex conjuga- tion operator KR relative to some representation R, and a generic unitary V [24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Since any unitary can be under- stood as a change of basis operator, then any antiunitary operator can be understood as the complex conjugation operation relative to a specific representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Thus, if there exist a representation R in which the Hamiltonian is real, then the FORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14), would hold for oper- ators having definite parity under the complex conjuga- tion, KR, relative to that representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Note that the Hamiltonian is a quantum observable, namely a Hermi- tian operator with real eigenvalues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Accordingly, in the representation where H(B) is diagonal, the Hamiltonian is trivially real.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' That is, irrespective of the specific form of the Hamiltonian, the following always holds H(B) = K† H(B)H(B)KH(B) , (17) where KH(B) is the complex conjugation relative to the representation where H(B) is diagonal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' It follows that for any couple of operators A and B that are either purely real or purely imaginary in that representation, the FORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14), hold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' For a sufficiently complex Hamiltonian this may well have no consequence whatsoever in regard to the validity of the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), which refers to quan- tities A, B with definite parity under Θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' This is further illustrated below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Examples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – Consider an Heisenberg magnet in a possibly non homogeneous field Bi = (Bx i , By i , Bz i ), H(B) = J � i,j σi · σj − � i Bi · σi , (18) where σi = (σx i , σy i , σz i ), with σα i denoting Pauli operators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Since, by definition, all spin operators σα i are odd under time reversal Θ, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (4), the Hamiltonian is not invariant under time reversal Θ, while it is invariant under the joint action of Θ, and the reversal of B, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', it obeys Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Accordingly the OCRs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14) are obeyed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Let us consider first the homogeneous case Bi = B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Let’s fix the axes so that z is the direction of the applied field B, B = Bzˆz, then: H(B) = J � σi · σj − Bz � i σz i (19) In the representation where the tensor product ⊗iσz i is diagonal, the Hamiltonian is real.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' That is if K⊗σz i is the complex conjugation relative to that representation, it is H(B) = K† ⊗σz i H(B)K⊗σz i .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (20) In fact, in said representation, the Pauli matrices σz i and σx i are real while the σy i are imaginary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Thus the term that couples to the applied magnetic field is real.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The interaction term is also real because it is the sum of terms of the type σα i σα j , which are real regardless of whether α is z, x, or y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Consider now the case when the external field is not homogeneous in space Bi ̸= Bj for i ̸= j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' If it only changes in modulo, but it has a fixed direction, then the previous argument will continue to apply unaltered.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' If its direction also changes, but remains confined onto one fixed plane, the argument still applies, with minimal changes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' To see that fix the axes so that the field has only components along x and z, so that the Hamiltonian reads H(B) = J � σi · σj − � (Bx i σx i + Bz i σz i ) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (21) Since in the representation of the basis spanned by the eigenvectors of ⊗σz i both the σx i and the σz i are all real, and, as discussed above the interaction term is also real, then the Hamiltonian is real as well in said representation, that is, the unconventional time reversal symmetry of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (20) holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Accorodingly, the FORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14), hold for any cou- ple of observables A, B that have definite parity under the transformation K⊗σz i .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The spin operators σα i all have def- inite parities under K⊗σz i : the σα i are even for α = x, z, and are odd for α = y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Considering that all the operators σα i are, by definition, odd under the time reversal Θ, it follows, for example, that the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) would hold despite the presence of the magnetic field, for couples of operators A = σα i , B = σβ j , with α and β being both y, or both either x or z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' However for couples of the type α = y, β = x, z, the FOR, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (14), reads Lσy i ,σz(x) j (t, B) = −Lσz(x) j ,σy i ,(t, B) , (22) thus violating the ORs Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), which would predict a plus sign on the r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' instead of a minus sign.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' This is illustrateed in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' 1, top panel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Based on Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (22) it is not hard to get convinced that the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), would not hold for A and B being spin operators (which are odd under Θ) pointing in generic directions, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' for A = ai · σi, B = bi · σi where ai, bi denote real unit vectors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Consider now the general case where Bi is not confined to a plane but explores all spatial directions as the spatial index i is varied over all sites of the spin network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We p-3 Michele Campisi 0 1 2 3 4 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='2 t B1 z,B2 x �0 A=σ1 z B=σ3 y 0 1 2 3 4 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='10 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='05 t B1 z,B2 x,B3 y≠0 A=σ1 z B=σ3 y LAB LBA Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' 1: Breakdown of Onsager relations, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Top: Re- laxation functions LAB and LBA for A = σz 1, B = σz 3 for a 3 sites Heisenberg magnet, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (18), with Bα i = 0 except Bz 1, Bx 2 = −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' According to Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (22) it is LAB = −LBA, whereas the OR’s predicts LAB = LBA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Bottom: same as top except for By 3, which now has the value By 3 = −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (18), predicting LAB = LBA is not obeyed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In both plots it is J = 1, β = 2, and ℏ = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' note that, at variance with the case above, in this case it is generally not even possible to find an “unconventional time reversal” symmetry besides the trivial invariance un- der KH(B).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' So while in the previous case the OR’s Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) remain valid for certain couples of observables, now we expect them to be generally violated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' 1, bottom panel, shows an example of such violations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Fixed field fluctuation relations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – The presence of an “unconventional time reversal” symmetry may as well result in the validity of the nonequilibrium flutuation relations [19] despite the presence of a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' At variance with linear response relations, those relations ap- ply arbitrarily far from equilibrium and refer to situations where the Hamiltonian has an explicit time-dependence, H = H(t, B).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Under the provision that there exist an “unconventional time reversal” transformation, K , that commutes with the Hamiltonian at all times during the driving protocol, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', KH(t, B) = H(t, B)K, ∀t ∈ [0, τ], (23) one would have, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', for a system prepared in an equi- librium thermal state that evolves unitarily, the following “fixed field fluctuation relation” for work p(w, B) ˜p(−w, B) = e−β(w−∆F ), (24) featuring the same B in both the forward work statis- tics, p(w, B) (obtained from the evolution generated by H(t, B)), and the backward work statistics, ˜p(w, B) (ob- tained from the evolution generated by H(τ − t, B), with τ the duration of the time dependent driving).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' This is at variance with the case customarily discussed in the lit- erature whereby Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (9) is assumed to hold at all times [19, 28, 29]) and the backward probability ˜p is taken at reversed field −B .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The proof of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (24) follows exactly the standard proof (see appendices B, and C of Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [19]) with Θ being re- placed by K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Note that at variance with the Onsager relations, that are possibly only apparently obeyed in case of “false time reversal violation”, the fluctuation relation p(w)/p(−w) = e−β(w−∆F ) is simply obeyed in its standard form, despite the magnetic field, when Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (23) holds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The validity of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (24) was often observed in the lit- erature, see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', [30–32], although the issue related to the presence of a magnetic field and the according lack of microreversibility (which would have ensured its valid- ity) typically passed unnoticed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Needless to say, all those previous works considered situations where the Hamilto- nian was real in some representation, at all times, which ensured the validity of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (24), a fact that instead was only acknowledged and discussed in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In this regard it is also worth remarking that Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (13) would instantaneously hold at each time, t, with an “un- conventional time reversal” KH(t,B), that possibly changes in time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Accordingly, its ubiquitous validity does not en- sure the validity of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (24) featuring a fixed K, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (23).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' It is a simple exercise to show that, even for a single spin in a time dependent magnetic field H(t, B) = −B(t) · σ, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (24) does not hold when B(t) explores all three spatial dimensions [33].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Discussion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – We have established two main facts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' First, for quantum systems a trivial “unconventional time reversal” always exist, and that is the symmetry un- der complex conjugation in the representation where the Hamiltonian is diagonal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' This very same fact was already noticed by Robnik and Berry [15] in the context of level statistics in quantum billiards.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Second, in contrast with what previous research suggested, the presence of one or more “unconventional time reversals” does not by itself guarantee the validity of the Onsager relations, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), when a magnetic field is present.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' As illustrated by ex- amples, they indeed can be violated, however if couples of observables exist that have definite parities under both p-4 False Onsager relations time reversal and an “unconventional time reversal” which is a symmetry of the problem, the ORs will be obeyed for them, provided the products of their parities under time reversal is the same as that under the “unconventional time reversal”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Roughly speaking, the more “unconventional time re- versal symmetries” a system has, the larger the set of cou- ples of observables that obey the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2), despite a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' On the contrary when the only “unconven- tional time reversal” is the trivial one, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (17) one should generally expect the ORs Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2) to be violated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The sit- uation is somewhat similar to dynamical system theory, where we have two extremes: full integrability (as many conserved quantities as are the degrees of freedom), and full ergodicity (the Hamiltonian is the only first integral of motion).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Between these two extremes lie complex sys- tems displaying both regular and irregular motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Here, similarly, we have the case where there are as many “un- conventional time reversal symmetries” as are the degrees of freedom 1, the case when only the trivial one exist, and the complex situation in between featuring both violation and obedience of the ORs, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Since, as illustrated by our examples, the Onsager rela- tions may be violated despite the presence of “false time reversal violations”, our results leave open the possibility of the “Umkehreffekt”, which in fact, as mentioned above, has been experimentally observed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Thus the present study does not provide any fundamental reason to exclude the possibility of achieving Carnot efficiency at finite power, in the way discussed in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' We further have commented that the omnipresence of “false time reversal violations” does not, per se, imply the ubiquitous validity of quantum fluctuation relations, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', the work fluctuation relation, in presence of a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' The latter would hold provided the “unconventional time reversal symmetry” is one and the same during the whole driving protocol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' While many models of many-body systems that are customarily studied in the literature (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', the driven Ising model in transverse field) satisfy that re- quirement that is generally not the case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' A question that remains to be answered is whether “false time reversal violation” is omnipresent as well in the classical case, and how can one construct the accord- ing “unconventional time reversal” transformation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' While this issue was easily addressed in the quantum case, the question does not seem to admit a simple answer for clas- sical systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Addressing that might reveal a new discor- dance between classical and quantum realms 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Acknowledgements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' – I would like to thank Giu- liano Benenti, Lamberto Rondoni and Davide Carbone for 1This would occur, for example, in the case of a set of non- interacting spins with local magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' In that case all relaxation functions are null and the ORs are trivially obeyed for any couple of observables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' 2One such discordance whereby deterministic friction appears in the classical realm but not in the quantum one was reported by Berry and Robbins [34] pointing out relevant literature and for the stimulating dis- cussions that were instrumental to shaping this work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' REFERENCES [1] Benenti G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Saito K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Casati G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 106 (2011) 230602.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [2] Allahverdyan A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Hovhannisyan K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Melkikh A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Gevorkian S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 111 (2013) 050601.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [3] Campisi M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Fazio R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Nat Commun, 7 (2016) 11895.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [4] Holubec V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Ryabov A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 121 (2018) 120601.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [5] Bonanc¸a M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Journal of Statistical Mechanics: Theory and Experiment, 2019 (2019) 123203.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [6] Miura K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Izumida Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Okuda K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' E, 105 (2022) 034102.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [7] Wolfe R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Smith G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 129 (1963) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [8] Saito K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Benenti G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Casati G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Prosen T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' B, 84 (2011) 201306.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [9] Luo R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Benenti G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Casati G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Wang J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Research, 2 (2020) 022009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [10] Gabrielli D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Jona-Lasinio G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Landim C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 77 (1996) 1202.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [11] Lebowitz J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Spohn H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 78 (1997) 394.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [12] Gabrielli D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Jona-Lasinio G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Landim C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Jour- nal of Statistical Physics, 96 (1999) 639.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [13] Bonella S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Ciccotti G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Rondoni L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', EPL (Eu- rophysics Letters), 108 (2014) 60004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [14] Carbone D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', De Gregorio P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Rondoni L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Annals of Physics, 441 (2022) 168853.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [15] Robnik M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Berry M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Journal of Physics A: Mathematical and General, 19 (1986) 669.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [16] Gregorio P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Bonella S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Rondoni L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Sym- metry, 9 (2017) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [17] Coretti A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Bonella S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Rondoni L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Ciccotti G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Molecular Physics, 116 (2018) 3097.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [18] Coretti A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Rondoni L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Bonella S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' E, 102 (2020) 030101.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [19] Campisi M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', H¨anggi P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Talkner P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Mod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 83 (2011) 771.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [20] Onsager L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 37 (1931) 405.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [21] Onsager L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 38 (1931) 2265.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [22] de Groot S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Mazur P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Non-equilibrium Ther- modynamics (Dover, New York) 1984.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [23] Kubo R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Jpn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 12 (1957) 570.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [24] Messiah A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Quantum Mechanics (North Holland, Ams- terdam) 1962.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [25] Casimir H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Mod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 17 (1945) 343.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [26] Bochkov G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Kuzovlev Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Zh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Eksp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Teor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Fiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 76 (1979) 1071 [Sov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' JETP 49, 543–551 (1979)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [27] Bochkov G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Kuzovlev Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Zh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Eksp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Teor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Fiz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 72 (1977) 238 [Sov.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' JETP 45, 125–130 (1977)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [28] Andrieux D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Gaspard P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 100 (2008) 230404.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [29] Jarzynski C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and W´ojcik D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 92 (2004) 230602.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [30] Dorner R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Goold J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Cormick C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Paternostro M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Vedral V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', 109 (2012) 160601.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' p-5 Michele Campisi [31] Fusco L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Pigeon S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Apollaro T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Xuereb A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Mazzola L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Campisi M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Ferraro A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Paternostro M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and De Chiara G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' X, 4 (2014) 031029.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [32] Schmidtke D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Knipschild L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Campisi M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Steinigeweg R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Gemmer J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' E, 98 (2018) 012123.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [33] Orlandini M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', BSc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Thesis, University of Florence, (2022) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' [34] Berry M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' and Robbins J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=', Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' London, Ser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' A442, 1993 (659) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} +page_content=' p-6' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/rtAzT4oBgHgl3EQfrf2X/content/2301.01645v1.pdf'} diff --git a/sNAzT4oBgHgl3EQfcvwO/content/tmp_files/2301.01406v1.pdf.txt b/sNAzT4oBgHgl3EQfcvwO/content/tmp_files/2301.01406v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..6215b2b6dddd821494aab9bc34bdfb24b0a9bc1a --- /dev/null +++ b/sNAzT4oBgHgl3EQfcvwO/content/tmp_files/2301.01406v1.pdf.txt @@ -0,0 +1,489 @@ + +1 +Atomic and Electronic Structures of Correlated SrRuO3/SrTiO3 +Superlattices +Seung Gyo Jeong and Woo Seok Choi* +Department of Physics, Sungkyunkwan University, Suwon 16419, Korea +Ahmed Yousef Mohamed and Deok-Yong Cho* +IPIT & Department of Physics, Jeonbuk National University, Jeonju 54896, Republic of Korea + +Atomic-scale precision epitaxy of perovskite oxide superlattices provides unique opportunities +for controlling the correlated electronic structures, activating effective control knobs for intriguing +functionalities including electromagnetic, thermoelectric, and electrocatalytic behaviors. In this +study, we investigated the close interplay between the atomic and electronic structures of correlated +superlattices synthesized by atomic-scale precision epitaxy. In particular, we employ superlattices +composed of correlated magnetic SrRuO3 (SRO) and quantum paraelectric SrTiO3 (STO) layers. +In those superlattices, RuO6 octahedral distortion is systematically controlled from 167º to 175º +depending on the thickness of the STO layers, also affecting the TiO6 octahedral distortion within +the STO layer. Customized octahedral distortion within SRO/STO superlattices in turn modifies +the electronic structures of both the Ti and Ru compounds, observed by X-ray absorption +spectroscopy. Our results identify the close correlation between atomic lattice and electronic +structures enabled by the facile controllability of atomic-scale epitaxy, which would be useful for +designing future correlated oxide devices. +Keywords: Atomic-scale epitaxy, Electronic Structure, Oxide superlattice, SrRuO3, Octahedral +distortion +Email: choiws@skku.edu, zax@jbnu.ac.kr + + +2 +I. INTRODUCTION +Atomic-scale heterostructuring offer a facile control knob for customizing correlated functional +oxides. Perovskite ABO3 oxides exhibit a variety of emergent functional phenomena such as electronic +and magnetic phase transitions, superconductivity, ferroelectricity, (anti)ferromagnetism, multiferroicity, +various Hall effects, and topologically non-trivial states [1-8]. Recent technological advances in epitaxy +and microscopy let us better access the atomic-scale precision oxide heterostructures for future quantum +applications [9]. For corner-sharing BO6 networks within the ABO3 perovskites, the bond length and +angle of the B-O-B are crucial in determining the strong interplay between charge, spin, lattice, orbital, +and topological degrees of freedom which gives birth to the intriguing functionality in oxides [10,11]. +Numerous oxide heterostructuring approaches have been proposed to modulate the functionalities, for +example, cation substitutions of the B-site (or A-site), oxygen vacancy engineering, dimensionality +control, imposing epitaxial strain, and interfacial octahedral symmetry coupling [9,12-16]. Atomically +designed superlattices composed of more than two different correlated oxides are useful to modulate the +various degrees of freedom via deliberate heterostructuring. Relevant design parameters of superlattice +include individual layer compounds and thicknesses, number of repetitions, and stacking order [17]. This +provides practical and facile controllability of octahedral distortion and resultant crystalline symmetry of +systems, determining correlated functionality with a possibility of emergent phenomena of the +superlattices. + +X-ray abortion spectroscopy (XAS) provides atomic-selective electronic structures appropriate +for studying perovskite oxide heterostructures [18-22]. XAS is an element-sensitive technique to +detect the local electronic structure of an unoccupied orbital state determined by the specific bonding +geometry of specific atoms and their orbital hybridization. By controlling the photon energy of incident +X-ray, we selectively detect the resonance peaks indicating an electronic transition from the core level +to the unoccupied states with the creation of a core hole. The edge structure of the spectrum contains the + + +3 +chemical information including the oxidation state and coordination environment helpful for figuring out +the modulated electronic correlation. + +Stoner ferromagnetic SrRuO3 is a suitable model system to investigate and understand the +correlation between the atomic and electronic structures leading to versatile functionalities, such +as dimensional crossover with phase instability, anomalous and/or topological Hall effects, +electrocatalyst, THz phonon emission, spin-phonon coupling, and modulation of Weyl fermions [1- +3,13,15,23-27]. A variety of SRO heterostructures have been suggested to engineer the RuO6 octahedral +distortion via stoichiometry and/or epitaxial strain control of a single SRO film. However, most +conventional approaches usually accompany unintended side effects originating from dissimilar +substrates, strain relaxation, cation vacancy-induced electronic change, dislocation, and electronic +reconstruction. + +In this paper, we demonstrate an atomic-scale heterostructure approach for achieving +octahedral distortion control and a systematic modulation of local electronic structure in a +correlated oxide superlattice. To overcome the extrinsic effects in the conventional SRO +heterostructure, we use the correlated oxide superlattice composed of ferromagnetic metallic SRO and +nonmagnetic insulating SrTiO3 (STO). Here, the identical A-site ion (Sr) and the highly suppressed +charge transfer between the SRO and STO layers lead to atomically sharp chemical and electronic +interfaces [4]. The periodic superlattice structures amplify the experimental signals from atomically thin +SRO layers, enabling the conventional characterization by using X-ray diffraction (XRD) and XAS. By +atomically controlling the STO layer thickness within SRO/STO superlattices, the octahedral distortion +can be controlled systematically, despite the same thickness and stoichiometry of the SRO layer. The +customized octahedral distortions for both the TiO6 and RuO6 octahedra modify the collective electronic +structures, which is experimentally observed by XAS measurement in this study. + + +4 + +II. EXPERIMENTS AND DISCUSSION +To control the octahedral distortion in SRO/STO superlattices, we precisely synthesized the epitaxial +oxide superlattice composed of 8 atomic unit cells (u.c. ~0.4 nm) of SRO and y u.c. of STO layers with +10 repetitions ([8|y] superlattice) on (001)-oriented single crystal STO substrates by using pulsed laser +epitaxy [3,4,22,25-27]. We pretreated the STO substrate using buffered HF and annealed it at 1000°C +under atmospheric conditions for the atomically flat surfaces of the substrate. We confirmed the typical +step-terrace structure on STO substrate by using atomic force microscopy. We utilized a KrF excimer +laser (248 nm; IPEX-868, LightMachinery) with 1.5 J/cm2 of fluence and 5 Hz of repetition rate to ablate +stoichiometric ceramic SRO and STO targets. We used 750°C and 100 mtorr of oxygen partial pressures +for obtaining the stoichiometric SRO and STO layers. We verified the atomically controlled periodicity +of superlattice by utilizing a high-resolution XRD with Cu K-α1 (PANalytical). + +We performed the XAS measurements at the 16A1 beamline of the Taiwan Light Source in the +fluorescence yield (FY) mode for the Ru L3-edge spectra, and the 2A beamline of the Pohang Light +Source in the total electron yield (TEY) mode for the O K-edge and Ti L2,3-edge spectra. XAS spectra +were obtained at room temperature with π-polarization, in which the electrical field direction of the X- +rays is perpendicular to the sample plane. To obtain the X-ray linear dichroism (XLD) for Ru L3-edge, +we rotated the samples with respect to the incoming X-rays by an incidence angle, θi = 0° (beam normal) +and θi = 70° (inclined). Whereas the spectra taken at θi = 0° contain horizontal XAS responses (Ix,y) only, +those taken at θi = 70° contain a combination of vertical (Iz) and horizontal (Ix,y) XAS responses as cos270° +× Ix,y + sin270° × Iz. The x, y, and z directions in XAS polarization are the [100]pc, [010]pc, and [001]pc +directions of the thin film, respectively, where [hkl]pc denotes the crystallographic direction in perovskite +pseudo-cubic notation. + + +5 + +XRD θ-2θ measurements confirm the atomic-scale precision control of [8|y] superlattices with +different y, as shown in Fig. 1. XRD θ-2θ results of [8|y] superlattices show expected superlattice Bragg +peaks with well-defined Pendellӧsung fringes (Fig. 1a), indicating high-quality samples with atomically +sharp interfaces and surfaces. The separation between the superlattice peaks increases with decreasing y +(dotted lines), corresponding to the decrease in supercell thickness. The thickness of superlattice films +was obtained by Bragg’s law, +λ +2 (sin θn − sin θn−1)−1, where λ and θn are the X-ray wavelength for Cu +Kα1 (0.154 nm) and the angle position of the nth-order superlattice peak, respectively. The estimated +total thickness of superlattices are 60.50, 54.00, 50.00, 39.24, and 34.66 nm for the y = 8, 6, 4, 2, and 1 +superlattices, respectively. These results represent the well-controlled atomic structures of the SRO/STO +superlattices. The asterisk (*) denotes the STO substrate peaks and the XRD θ-2θ data of the SRO single +film (~19 nm thick) was included for comparison. + +The atomically controlled STO thickness within the superlattice determines the orthorhombicity +(ao/bo) of the SRO layer within the [8|y] superlattices [22]. We performed the off-axis XRD θ-2θ scans +around (204) STO plane with different ϕ (Figs. 1b and S1), to characterize ao/bo of SRO single film and +superlattices. When ϕ = 0º, 90º, 180º, and 270º, the Bragg diffraction peaks shown in Fig. 1b correspond +to (44–4)o, (260)o, (444)o, and (620)o plane [12,28]. Note that (hkl)o denotes the crystallographic plane +for the orthorhombic SRO thin film as schematically shown in Fig. 1 [29]. For the SRO single film, +whereas the Bragg peaks of ϕ = 0º and 180º appear at the same position, those of ϕ = 90º and 270º appear +at asymmetric positions indicating an orthorhombic structure with the ao/bo value deviating from one. +For the [8|y] superlattices, the separation of peak position between the ϕ = 90º and 270º configurations +decreases systematically with increasing y, represented by the decrease of ao/bo, as shown in Fig. S1. +Figure 1c summarizes ao/bo values of SRO single film (y = 0) and [8|y] superlattices, as a function of y. + + +6 +Assuming that the change in the bond length between Ru and O ions is nearly identical, which is the case +for most transition metal oxides, we estimate the orthorhombic tilt angle (θt) as θt = 180º – 2cos–1(bo/ao) +[29,30] (Right panel of Fig. 1c and d schematically defines the ao, bo, and θt in the orthorhombic RuO6 +octahedra.). Fig. 1d shows that the 165º of θt for SRO single film (comparable to 160º of θt for bulk SRO +[29,30]) systematically increases up to 175º for the y = 8 superlattice. This result proves that θt of the +SRO layer can be effectively modulated by the thickness of adjacent cubic STO layers within the +superlattices [3,22], while maintaining the intrinsic properties, e.g., stoichiometry and thickness, of the +SRO layers. + +XAS is employed to scrutinize the electronic structures of the SRO layer under the customized +octahedral distortion. Figure 2a schematically shows the experimental configuration of XAS for the +SRO/STO superlattices with two different detection modes, i.e., FY for Ru L3-edge and TEY for O K- +and Ti L2,3-edge. Figure 2b exemplarily shows the schematic of the XAS excitation process at the Ru L3- +edge (Ru 2p3/2 to 4d transition near a photon energy of 2840 eV) and Ti L-edge (Ti 2p3/2 (or 2p1/2) to 3d +transition near 460 eV (or 466 eV)). The probing depth of the Ru L-edge XAS is expected to be in the +order of microns so that the signals originate throughout the superlattice. Meanwhile, the probing depth +of O K- and Ti L-edge XAS (TEY) is only a few nanometers so that the signals originate predominantly +from the superlattice rather than the STO substrate. + +XAS at O K-edge reflects the unoccupied O 2p state that is hybridized with transition metal ions’ +orbitals. Figure 2c shows the XAS spectra at O K-edge for SRO single film and [8|y] superlattices. The +vertical dotted lines indicate the two peak positions at ~528 and ~530 eV of the unoccupied O 2p states +hybridized with Ru 4d and Ti 3d states, respectively. The following peaks above 530 eV are related to +oxygen hybridization with Sr 4d (533.5–538.0 eV), and Ru 5sp/Ti 4sp (538–547 eV) states [21]. For the +[8|y] superlattices, two peaks are clearly visible, of which their intensities strongly depend on y. With + + +7 +increasing y, the intensity of Ru-O hybridization (528 eV) systematically decreases while that of Ti- +O hybridization (530 eV) significantly increases, manifesting the changes in the abundance of Ru +and Ti ions. On the other hand, for the SRO film (20 nm-thick), only the Ru-O hybridization peak is +observed due to the limited probing depth. + +Figure 3a shows the Ti L2,3-edge XAS spectra of [8|y] superlattice with different y’s in the TEY mode. +The first and the second peaks (L3-edge) can be attributed to Ti 2p3/2→3d-t2g and 3d-eg transitions, +respectively, while the third and the fourth peaks (L2-edge) can be to Ti 2p1/2→3d-t2g and 3d-eg transitions, +respectively. Since the Ti4+ ions have no 3d electrons in their ground state, the electron correlation effect +inside the Ti4+ ion is negligible. Thus, the energy difference between the first and second peaks (or the +third and fourth peaks) can be utilized as a measure of the difference in the effective crystal field strength +between Ti t2g and eg manifolds. The effective crystal field strength (Δ), i.e., the energy difference of the +two adjacent peaks in the L3-edge spectrum is 2.3 eV for y =1, while it slightly but systematically +increases to 2.4 eV for y = 8. The value for y = 8 is consistent with the reported value for cubic STO +single crystal [31]. Although the actual lattice structure of the STO monolayer would be extremely +difficult to characterize, the 0.1 eV decrease as y decreases from 8 to 1, corroborates that the crystal field +effect becomes weaker, plausibly because of the distortion in the TiO6 local structure. + + Figure 3b shows the atomic picture describing the y-dependence in Δ. This scenario is further +supported by the fact that the charge transfer between SRO and STO layers is negligibly small and the +in-plane lattice parameters of the SRO and STO layers within the superlattice are nearly identical to the +STO substrate; The former has been confirmed by the XAS Ti L2,3-edge exhibiting Ti4+ oxidation state +for all the [8|y] superlattices and the latter has been confirmed by both XRD reciprocal space map and +scanning transmission electron microscopy [3,4,25]. This result suggests that the STO layer thickness + + +8 +within the superlattices affects the lattice structures of not only the SRO (octahedral distortion evidently +shown in Fig. 2) but also the STO layers themselves. + +Atomically controlled orthorhombic distortion in SRO/STO superlattices also determines the +Ru 4d orbital state and its anisotropy, as shown in Figure 4. Figures 4a and 4b show XAS at Ru L3- +edge with θi = 0° and 70° for the SRO film and the [8|y] superlattices. The main doublets from Ru 2p3/2 +→ 4d t2g (~2839.5 eV) and eg (~2842.0 eV) states barely show a noticeable evolution in peak energies +with increasing y. This suggests robustness in the Ru4+ valence states for both SRO film and the +superlattices as expected [22]. This confirms that the electronic structure of SRO/STO superlattice is +dominantly determined by structural modification not extrinsic effects including charge transfer, defects, +nor anti-site disorders. To verify the modulated θt-dependent orbital occupation and polarization, we +analyze the XLD signal, namely, the difference between horizontal and vertical polarization components +in the XAS intensity (XLD = Ix,y – Iz), as shown in Figure 4c. The Iz (Ix,y) of Ru t2g states is proportional +to the number of unoccupied states in dxz,yz states (sum of 1/2dxy and 1/2dxz,yz states). Hence, negative and +positive signals in XLD spectra at the Ru t2g region are related to the dxz,yz and dxy orbital states, +respectively, consistent with the previous report on the SRO/STO superlattices [22]. We note that XLD +spectra at Ru eg states show no meaningful trend for y, and in turn to θt. However, as y increases (θt +decreases), the negative XLD signal from dxz,yz orbital disappears, and the positive XLD signal from dxy +becomes enhanced. This implies a decrease in occupancy in dxy states due to the structural modifications +of SRO/STO superlattices. It is noteworthy that the XLD peak position at dxy states slightly increases +(and becomes broader) when θt decreases, suggesting additional Ru t2g splitting between dxz,yz and dxy +states. Figure 4d schematically shows the possible scenario of the electronic structure evolution as a +function of θt within the superlattices. We propose that the Ru t2g splitting increases and partially leads +to the enhancement of the unoccupied dxy state due to the decrease of θt within the SRO/STO superlattices. + + +9 + +III. CONCLUSION +In conclusion, we precisely control the octahedral distortion in orthorhombic SRO heterostructures to +engineer the correlated electronic structures of SRO/STO superlattices. By controlling the STO layer +thickness within the superlattices, the Ru-O-Ru distortion angle in the adjacent 8 u.c. SRO layer increases +systematically from 167º to 175º. Precisely adjusted octahedral distortion in SRO/STO superlattice +determines the electronic structure of both transition metal Ti and Ru compounds within the TiO6 and +RuO6 octahedra respectively, as is confirmed by atomic-selective XAS. These results suggest a versatile +tunability of atomic-scale heterostructuring to customize the strongly correlated electron systems. + +ACKNOWLEDGEMENT +This work was supported by the Basic Science Research Program through the National Research +Foundation of Korea (NRF-2021R1A2C2011340 and 2022R1C1C2006723). D.-Y.C is supported by the +research fund of Jeonbuk National University in 2021. + +REFERENCES +[1] S. W. Cho et al., Tailoring topological Hall effect in SrRuO3/SrTiO3 superlattices. Acta Mater. 216, +117153 (2021). +[2] Z. Fang et al., The Anomalous Hall Effect and Magnetic Monopoles in Momentum Space. Science +302, 92-95 (2003). +[3] S. G. Jeong et al., Symmetry-Driven Spin-Wave Gap Modulation in Nanolayered SrRuO3/SrTiO3 +Heterostructures: Implications for Spintronic Applications. ACS Appl. Nano Mater. 4, 2160-2166 +(2021). +[4] S. G. Jeong et al., Phase Instability amid Dimensional Crossover in Artificial Oxide Crystal. Phys. +Rev. Lett. 124, 026401 (2020). +[5] K. T. Kang et al., Ferroelectricity in SrTiO3 epitaxial thin films via Sr-vacancy-induced tetragonality. +Appl. Surf. Sci. 499, 143930 (2020). +[6] M. Kim et al., Superconductivity in (Ba,K)SbO3. Nat. Mater. 21, 627-633 (2022). +[7] J. Wang et al., Epitaxial BiFeO3 Multiferroic Thin Film Heterostructures. Science 299, 1719-1722 +(2003). + + +10 +[8] J. Yue et al., Anomalous transport in high-mobility superconducting SrTiO3 thin films. Sci. Adv. 8, +eabl5668 (2022). +[9] P. D. C. King et al., Atomic-scale control of competing electronic phases in ultrathin LaNiO3. Nat. +Nanotechnol. 9, 443-447 (2014). +[10] Y. Tokura & N. Nagaosa. Orbital Physics in Transition-Metal Oxides. Science 288, 462-468 (2000). +[11] H. Y. Hwang et al., Emergent phenomena at oxide interfaces. Nat. Mater. 11, 103-113 (2012). +[12] A. Vailionis, W. Siemons & G. Koster. Room temperature epitaxial stabilization of a tetragonal +phase in ARuO3 (A=Ca and Sr) thin films. Appl. Phys. Lett. 93, 051909 (2008). +[13] S. A. Lee et al., Enhanced electrocatalytic activity via phase transitions in strongly correlated +SrRuO3 thin films. Energy Environ. Sci. 10, 924-930 (2017). +[14] D. G. Schlom et al., Elastic strain engineering of ferroic oxides. MRS Bull. 39, 118-130 (2014). +[15] D. Kan et al., Tuning magnetic anisotropy by interfacially engineering the oxygen coordination +environment in a transition metal oxide. Nat. Mater. 15, 432-437 (2016). +[16] J. M. Rondinelli, S. J. May & J. W. Freeland. Control of octahedral connectivity in perovskite oxide +heterostructures: An emerging route to multifunctional materials discovery. MRS Bull. 37, 261- +270 (2012). +[17] R. Ramesh & D. G. Schlom. Creating emergent phenomena in oxide superlattices. Nat. Rev. Mater. +4, 257-268 (2019). +[18] J. Yano & V. K. Yachandra. X-ray absorption spectroscopy. Photosynth. Res. 102, 241 (2009). +[19] Z. Liao et al., Large orbital polarization in nickelate-cuprate heterostructures by dimensional control +of oxygen coordination. Nat. Commun. 10, 589 (2019). +[20] X. Liu et al., Interfacial charge-transfer Mott state in iridate–nickelate superlattices. Proc. Natl. Acad. +Sci. U.S.A. 116, 19863-19868 (2019). +[21] H. Jeong et al., Thickness-dependent orbital hybridization in ultrathin SrRuO3 epitaxial films. Appl. +Phys. Lett. 115, 092906 (2019). +[22] S. G. Jeong et al., Propagation Control of Octahedral Tilt in SrRuO3 via Artificial Heterostructuring. +Adv. Sci. 7, 2001643 (2020). +[23] H. I. Seo et al., Crystalline symmetry-dependent magnon formation in the itinerant ferromagnet +SrRuO3. Phys. Rev. B 103, 045104 (2021). +[24] G. Koster et al., Structure, physical properties, and applications of SrRuO3 thin films. Rev. Mod. +Phys. 84, 253-298 (2012). +[25] S. G. Jeong et al., Unconventional interlayer exchange coupling via chiral phonons in synthetic +magnetic oxide heterostructures. Sci. Adv. 8, eabm4005 (2022). +[26] S. G. Jeong et al., Spin–phonon coupling in epitaxial SrRuO3 heterostructures. Nanoscale 12, 13926- +13932 (2020). +[27] S. G. Jeong, A. Seo & W. S. Choi. Atomistic Engineering of Phonons in Functional Oxide +Heterostructures. Adv. Sci. 9, 2103403 (2022). +[28] D. Kan & Y. Shimakawa. Strain Effect on Structural Transition in SrRuO3 Epitaxial Thin Films. +Cryst. Growth Des. 11, 5483-5487 (2011). +[29] S. H. Chang et al., Thickness-dependent structural phase transition of strained SrRuO3 ultrathin +films: The role of octahedral tilt. Phys. Rev. B 84, 104101 (2011). +[30] A. T. Zayak, X. Huang, J. B. Neaton & K. M. Rabe. Structural, electronic, and magnetic properties +of SrRuO3 under epitaxial strain. Phys. Rev. B 74, 094104 (2006). +[31] W. Fan et al., Evolution of element-specific electronic structures in alkaline titanates. AIP Adv. 9, +065213 (2019). + + + + +11 + +Figure Captions. + +Fig. 1. Structural characterization of atomically designed SRO/STO superlattices: a XRD θ-2θ scans for +SRO single film and [8|y] superlattices. b off-axis XRD scans around (204) plane of STO with different +ϕ angles for SRO single film, [8|2], and [8|8] superlattices. c ao/bo values obtained by off-axis XRD scans +and d estimated tilt angle θt as a function of y. The right panel of c,d shows a schematic illustration of +RuO6 octahedra with θt. The [hkl]pc and [hkl]o indicates the crystallographic direction in perovskite +pseudo-cubic and orthorhombic SRO notation, respectively. + + + + +Intensity (arb. units) +(a) +(b) +SRO +[8]2] +[8]8] +Φ=270° +180° +-90° +[8]8] +10° +Intensity (arb. units) +120 +122 +124 +120 +122 +124 +120 +122 +124 +[8]6] +20 (degrees) +20 (degrees) +20 (degrees) +[8]4] +(c) +1.008 +[8|2] +1.004 +a +1.000 +[8]1] +(d) +180 +11001 +[010]。 +170 +SRO (~19 nm) +[001]pc +[-100]] +[010]。 +160 +20 +30 +40 +50 +0 +2 +4 +6 +8 +2 (degrees) +# of STO layers (u.c.) +12 +Fig. 2. a Schematic representation of XAS configuration for oxide superlattice with two different +detection modes (fluorescence yield and total electron yield). b atomic-selective excitation process of +XAS for Ru and Ti ions by controlling photon energy of X-ray. c O K-edge XAS spectra of [8|y] +superlattice taken at θi = 0°. The vertical lines indicate the peak positions related to Ru-O and Ti-O +hybridization as guides to the eye. + + + +元-polarization +a +1) FY +(c) +X-ray +O K-edge, TEY +2) TEY +[8]8] +0.=.0° +[8]6] +Oxide superlattice +XAS (arb. units) +[8]4] +(b) E个 Ru 4d + 3d +[8|2] +Ee +[8]1] +~2840 eV +SRO +~460 eV +530 +540 +550 +560 +Ru 2p3/2 +Ti 2p3/2 or 2p1/2 +Photon energy (eV) +13 + +Fig. 3. a Ti L2,3-edge XAS results of the [8|y] superlattices taken at θi = 0°. The vertical lines are guides +to the eye. b Schematic diagram of the Ti 3d orbital states with a possible y-dependent octahedral +distortion in the STO layer. + + + +(a) +(b) +Ti L2.3-edge, TEY +Ti 3d +00=0 +TA +0.1eVofenhancement +XAS (arb. units) +[8]8] +[8]8] +[8|1] +[8]6] +SrRuO +[8]4] +SrTiO3 +[8|2] +[8]1] +Possibledistortion +inSrTiOlayer +455 +460 +465 +470 +Photon energy (eV) +14 + +Fig. 4. a XAS spectra at Ru L3-edge for the SRO film and the [8|y] superlattices with a θi = 0° and b θi = +70°. c XLD = Ix,y – Iz. The vertical lines are guides to the eye. d Schematic illustration of possible Ru +orbital states within the SRO/STO superlattice as a function of θt. + + + +(a) +(b) +(c) +(d) +Ru L3-edge, FY +二 + Ru L3-edge, FY +Ru 4d, t2g +0, = 0° +[8]8] +Decreasein +dxy occupation number +XAS (arb. units) +[8]6] +XAS (arb. units) +XLD (arb. units) +[8]4] +d +XZ.VZ +[8|2] +[8]1] +SRO +[8]|1] +[8]8] +Improved distortion +2836 +2840 +2844 +2848 +2836 +2840 +2844 +2848 +2836 +2840 +2844 +2848 +Photon energy (eV) +Photon energy (eV) +Photon energy (eV) \ No newline at end of file diff --git a/sNAzT4oBgHgl3EQfcvwO/content/tmp_files/load_file.txt b/sNAzT4oBgHgl3EQfcvwO/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..b43144d3d5bd47aad87124f5793c191940ef6111 --- /dev/null +++ b/sNAzT4oBgHgl3EQfcvwO/content/tmp_files/load_file.txt @@ -0,0 +1,421 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf,len=420 +page_content='1 Atomic and Electronic Structures of Correlated SrRuO3/SrTiO3 Superlattices Seung Gyo Jeong and Woo Seok Choi* Department of Physics,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sungkyunkwan University,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Suwon 16419,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Korea Ahmed Yousef Mohamed and Deok-Yong Cho* IPIT & Department of Physics,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeonbuk National University,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeonju 54896,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Republic of Korea Atomic-scale precision epitaxy of perovskite oxide superlattices provides unique opportunities for controlling the correlated electronic structures,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' activating effective control knobs for intriguing functionalities including electromagnetic,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' thermoelectric,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' and electrocatalytic behaviors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' In this study, we investigated the close interplay between the atomic and electronic structures of correlated superlattices synthesized by atomic-scale precision epitaxy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' In particular, we employ superlattices composed of correlated magnetic SrRuO3 (SRO) and quantum paraelectric SrTiO3 (STO) layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' In those superlattices, RuO6 octahedral distortion is systematically controlled from 167º to 175º depending on the thickness of the STO layers, also affecting the TiO6 octahedral distortion within the STO layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Customized octahedral distortion within SRO/STO superlattices in turn modifies the electronic structures of both the Ti and Ru compounds, observed by X-ray absorption spectroscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Our results identify the close correlation between atomic lattice and electronic structures enabled by the facile controllability of atomic-scale epitaxy, which would be useful for designing future correlated oxide devices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Keywords: Atomic-scale epitaxy, Electronic Structure, Oxide superlattice, SrRuO3, Octahedral distortion Email: choiws@skku.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='edu, zax@jbnu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='kr 2 I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' INTRODUCTION Atomic-scale heterostructuring offer a facile control knob for customizing correlated functional oxides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Perovskite ABO3 oxides exhibit a variety of emergent functional phenomena such as electronic and magnetic phase transitions, superconductivity, ferroelectricity, (anti)ferromagnetism, multiferroicity, various Hall effects, and topologically non-trivial states [1-8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Recent technological advances in epitaxy and microscopy let us better access the atomic-scale precision oxide heterostructures for future quantum applications [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' For corner-sharing BO6 networks within the ABO3 perovskites, the bond length and angle of the B-O-B are crucial in determining the strong interplay between charge, spin, lattice, orbital, and topological degrees of freedom which gives birth to the intriguing functionality in oxides [10,11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Numerous oxide heterostructuring approaches have been proposed to modulate the functionalities, for example, cation substitutions of the B-site (or A-site), oxygen vacancy engineering, dimensionality control, imposing epitaxial strain, and interfacial octahedral symmetry coupling [9,12-16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Atomically designed superlattices composed of more than two different correlated oxides are useful to modulate the various degrees of freedom via deliberate heterostructuring.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Relevant design parameters of superlattice include individual layer compounds and thicknesses, number of repetitions, and stacking order [17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This provides practical and facile controllability of octahedral distortion and resultant crystalline symmetry of systems, determining correlated functionality with a possibility of emergent phenomena of the superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' X-ray abortion spectroscopy (XAS) provides atomic-selective electronic structures appropriate for studying perovskite oxide heterostructures [18-22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' XAS is an element-sensitive technique to detect the local electronic structure of an unoccupied orbital state determined by the specific bonding geometry of specific atoms and their orbital hybridization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' By controlling the photon energy of incident X-ray, we selectively detect the resonance peaks indicating an electronic transition from the core level to the unoccupied states with the creation of a core hole.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The edge structure of the spectrum contains the 3 chemical information including the oxidation state and coordination environment helpful for figuring out the modulated electronic correlation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Stoner ferromagnetic SrRuO3 is a suitable model system to investigate and understand the correlation between the atomic and electronic structures leading to versatile functionalities, such as dimensional crossover with phase instability, anomalous and/or topological Hall effects, electrocatalyst, THz phonon emission, spin-phonon coupling, and modulation of Weyl fermions [1- 3,13,15,23-27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' A variety of SRO heterostructures have been suggested to engineer the RuO6 octahedral distortion via stoichiometry and/or epitaxial strain control of a single SRO film.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' However, most conventional approaches usually accompany unintended side effects originating from dissimilar substrates, strain relaxation, cation vacancy-induced electronic change, dislocation, and electronic reconstruction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' In this paper, we demonstrate an atomic-scale heterostructure approach for achieving octahedral distortion control and a systematic modulation of local electronic structure in a correlated oxide superlattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' To overcome the extrinsic effects in the conventional SRO heterostructure, we use the correlated oxide superlattice composed of ferromagnetic metallic SRO and nonmagnetic insulating SrTiO3 (STO).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Here, the identical A-site ion (Sr) and the highly suppressed charge transfer between the SRO and STO layers lead to atomically sharp chemical and electronic interfaces [4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The periodic superlattice structures amplify the experimental signals from atomically thin SRO layers, enabling the conventional characterization by using X-ray diffraction (XRD) and XAS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' By atomically controlling the STO layer thickness within SRO/STO superlattices, the octahedral distortion can be controlled systematically, despite the same thickness and stoichiometry of the SRO layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The customized octahedral distortions for both the TiO6 and RuO6 octahedra modify the collective electronic structures, which is experimentally observed by XAS measurement in this study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 4 II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' EXPERIMENTS AND DISCUSSION To control the octahedral distortion in SRO/STO superlattices, we precisely synthesized the epitaxial oxide superlattice composed of 8 atomic unit cells (u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' ~0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='4 nm) of SRO and y u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' of STO layers with 10 repetitions ([8|y] superlattice) on (001)-oriented single crystal STO substrates by using pulsed laser epitaxy [3,4,22,25-27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We pretreated the STO substrate using buffered HF and annealed it at 1000°C under atmospheric conditions for the atomically flat surfaces of the substrate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We confirmed the typical step-terrace structure on STO substrate by using atomic force microscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We utilized a KrF excimer laser (248 nm;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' IPEX-868, LightMachinery) with 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='5 J/cm2 of fluence and 5 Hz of repetition rate to ablate stoichiometric ceramic SRO and STO targets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We used 750°C and 100 mtorr of oxygen partial pressures for obtaining the stoichiometric SRO and STO layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We verified the atomically controlled periodicity of superlattice by utilizing a high-resolution XRD with Cu K-α1 (PANalytical).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We performed the XAS measurements at the 16A1 beamline of the Taiwan Light Source in the fluorescence yield (FY) mode for the Ru L3-edge spectra, and the 2A beamline of the Pohang Light Source in the total electron yield (TEY) mode for the O K-edge and Ti L2,3-edge spectra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' XAS spectra were obtained at room temperature with π-polarization, in which the electrical field direction of the X- rays is perpendicular to the sample plane.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' To obtain the X-ray linear dichroism (XLD) for Ru L3-edge, we rotated the samples with respect to the incoming X-rays by an incidence angle, θi = 0° (beam normal) and θi = 70° (inclined).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Whereas the spectra taken at θi = 0° contain horizontal XAS responses (Ix,y) only, those taken at θi = 70° contain a combination of vertical (Iz) and horizontal (Ix,y) XAS responses as cos270° × Ix,y + sin270° × Iz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The x, y, and z directions in XAS polarization are the [100]pc, [010]pc, and [001]pc directions of the thin film, respectively, where [hkl]pc denotes the crystallographic direction in perovskite pseudo-cubic notation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 5 XRD θ-2θ measurements confirm the atomic-scale precision control of [8|y] superlattices with different y, as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' XRD θ-2θ results of [8|y] superlattices show expected superlattice Bragg peaks with well-defined Pendellӧsung fringes (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1a), indicating high-quality samples with atomically sharp interfaces and surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The separation between the superlattice peaks increases with decreasing y (dotted lines), corresponding to the decrease in supercell thickness.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The thickness of superlattice films was obtained by Bragg’s law, λ 2 (sin θn − sin θn−1)−1, where λ and θn are the X-ray wavelength for Cu Kα1 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='154 nm) and the angle position of the nth-order superlattice peak, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The estimated total thickness of superlattices are 60.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='50, 54.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='00, 50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='00, 39.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='24, and 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='66 nm for the y = 8, 6, 4, 2, and 1 superlattices, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' These results represent the well-controlled atomic structures of the SRO/STO superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The asterisk (*) denotes the STO substrate peaks and the XRD θ-2θ data of the SRO single film (~19 nm thick) was included for comparison.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The atomically controlled STO thickness within the superlattice determines the orthorhombicity (ao/bo) of the SRO layer within the [8|y] superlattices [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We performed the off-axis XRD θ-2θ scans around (204) STO plane with different ϕ (Figs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1b and S1), to characterize ao/bo of SRO single film and superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' When ϕ = 0º, 90º, 180º, and 270º, the Bragg diffraction peaks shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1b correspond to (44–4)o, (260)o, (444)o, and (620)o plane [12,28].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Note that (hkl)o denotes the crystallographic plane for the orthorhombic SRO thin film as schematically shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1 [29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' For the SRO single film, whereas the Bragg peaks of ϕ = 0º and 180º appear at the same position, those of ϕ = 90º and 270º appear at asymmetric positions indicating an orthorhombic structure with the ao/bo value deviating from one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' For the [8|y] superlattices, the separation of peak position between the ϕ = 90º and 270º configurations decreases systematically with increasing y, represented by the decrease of ao/bo, as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' S1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 1c summarizes ao/bo values of SRO single film (y = 0) and [8|y] superlattices, as a function of y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 6 Assuming that the change in the bond length between Ru and O ions is nearly identical, which is the case for most transition metal oxides, we estimate the orthorhombic tilt angle (θt) as θt = 180º – 2cos–1(bo/ao) [29,30] (Right panel of Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1c and d schematically defines the ao, bo, and θt in the orthorhombic RuO6 octahedra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1d shows that the 165º of θt for SRO single film (comparable to 160º of θt for bulk SRO [29,30]) systematically increases up to 175º for the y = 8 superlattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This result proves that θt of the SRO layer can be effectively modulated by the thickness of adjacent cubic STO layers within the superlattices [3,22], while maintaining the intrinsic properties, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', stoichiometry and thickness, of the SRO layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' XAS is employed to scrutinize the electronic structures of the SRO layer under the customized octahedral distortion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 2a schematically shows the experimental configuration of XAS for the SRO/STO superlattices with two different detection modes, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', FY for Ru L3-edge and TEY for O K- and Ti L2,3-edge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 2b exemplarily shows the schematic of the XAS excitation process at the Ru L3- edge (Ru 2p3/2 to 4d transition near a photon energy of 2840 eV) and Ti L-edge (Ti 2p3/2 (or 2p1/2) to 3d transition near 460 eV (or 466 eV)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The probing depth of the Ru L-edge XAS is expected to be in the order of microns so that the signals originate throughout the superlattice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Meanwhile, the probing depth of O K- and Ti L-edge XAS (TEY) is only a few nanometers so that the signals originate predominantly from the superlattice rather than the STO substrate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' XAS at O K-edge reflects the unoccupied O 2p state that is hybridized with transition metal ions’ orbitals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 2c shows the XAS spectra at O K-edge for SRO single film and [8|y] superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The vertical dotted lines indicate the two peak positions at ~528 and ~530 eV of the unoccupied O 2p states hybridized with Ru 4d and Ti 3d states, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The following peaks above 530 eV are related to oxygen hybridization with Sr 4d (533.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='5–538.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='0 eV), and Ru 5sp/Ti 4sp (538–547 eV) states [21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' For the [8|y] superlattices, two peaks are clearly visible, of which their intensities strongly depend on y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' With 7 increasing y, the intensity of Ru-O hybridization (528 eV) systematically decreases while that of Ti- O hybridization (530 eV) significantly increases, manifesting the changes in the abundance of Ru and Ti ions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' On the other hand, for the SRO film (20 nm-thick), only the Ru-O hybridization peak is observed due to the limited probing depth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 3a shows the Ti L2,3-edge XAS spectra of [8|y] superlattice with different y’s in the TEY mode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The first and the second peaks (L3-edge) can be attributed to Ti 2p3/2→3d-t2g and 3d-eg transitions, respectively, while the third and the fourth peaks (L2-edge) can be to Ti 2p1/2→3d-t2g and 3d-eg transitions, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Since the Ti4+ ions have no 3d electrons in their ground state, the electron correlation effect inside the Ti4+ ion is negligible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Thus, the energy difference between the first and second peaks (or the third and fourth peaks) can be utilized as a measure of the difference in the effective crystal field strength between Ti t2g and eg manifolds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The effective crystal field strength (Δ), i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', the energy difference of the two adjacent peaks in the L3-edge spectrum is 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='3 eV for y =1, while it slightly but systematically increases to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='4 eV for y = 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The value for y = 8 is consistent with the reported value for cubic STO single crystal [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Although the actual lattice structure of the STO monolayer would be extremely difficult to characterize, the 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='1 eV decrease as y decreases from 8 to 1, corroborates that the crystal field effect becomes weaker, plausibly because of the distortion in the TiO6 local structure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 3b shows the atomic picture describing the y-dependence in Δ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This scenario is further supported by the fact that the charge transfer between SRO and STO layers is negligibly small and the in-plane lattice parameters of the SRO and STO layers within the superlattice are nearly identical to the STO substrate;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The former has been confirmed by the XAS Ti L2,3-edge exhibiting Ti4+ oxidation state for all the [8|y] superlattices and the latter has been confirmed by both XRD reciprocal space map and scanning transmission electron microscopy [3,4,25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This result suggests that the STO layer thickness 8 within the superlattices affects the lattice structures of not only the SRO (octahedral distortion evidently shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 2) but also the STO layers themselves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Atomically controlled orthorhombic distortion in SRO/STO superlattices also determines the Ru 4d orbital state and its anisotropy, as shown in Figure 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figures 4a and 4b show XAS at Ru L3- edge with θi = 0° and 70° for the SRO film and the [8|y] superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The main doublets from Ru 2p3/2 → 4d t2g (~2839.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='5 eV) and eg (~2842.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='0 eV) states barely show a noticeable evolution in peak energies with increasing y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This suggests robustness in the Ru4+ valence states for both SRO film and the superlattices as expected [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This confirms that the electronic structure of SRO/STO superlattice is dominantly determined by structural modification not extrinsic effects including charge transfer, defects, nor anti-site disorders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' To verify the modulated θt-dependent orbital occupation and polarization, we analyze the XLD signal, namely, the difference between horizontal and vertical polarization components in the XAS intensity (XLD = Ix,y – Iz), as shown in Figure 4c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The Iz (Ix,y) of Ru t2g states is proportional to the number of unoccupied states in dxz,yz states (sum of 1/2dxy and 1/2dxz,yz states).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Hence, negative and positive signals in XLD spectra at the Ru t2g region are related to the dxz,yz and dxy orbital states, respectively, consistent with the previous report on the SRO/STO superlattices [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We note that XLD spectra at Ru eg states show no meaningful trend for y, and in turn to θt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' However, as y increases (θt decreases), the negative XLD signal from dxz,yz orbital disappears, and the positive XLD signal from dxy becomes enhanced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' This implies a decrease in occupancy in dxy states due to the structural modifications of SRO/STO superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' It is noteworthy that the XLD peak position at dxy states slightly increases (and becomes broader) when θt decreases, suggesting additional Ru t2g splitting between dxz,yz and dxy states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Figure 4d schematically shows the possible scenario of the electronic structure evolution as a function of θt within the superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' We propose that the Ru t2g splitting increases and partially leads to the enhancement of the unoccupied dxy state due to the decrease of θt within the SRO/STO superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 9 III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' CONCLUSION In conclusion, we precisely control the octahedral distortion in orthorhombic SRO heterostructures to engineer the correlated electronic structures of SRO/STO superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' By controlling the STO layer thickness within the superlattices, the Ru-O-Ru distortion angle in the adjacent 8 u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' SRO layer increases systematically from 167º to 175º.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Precisely adjusted octahedral distortion in SRO/STO superlattice determines the electronic structure of both transition metal Ti and Ru compounds within the TiO6 and RuO6 octahedra respectively, as is confirmed by atomic-selective XAS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' These results suggest a versatile tunability of atomic-scale heterostructuring to customize the strongly correlated electron systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' ACKNOWLEDGEMENT This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF-2021R1A2C2011340 and 2022R1C1C2006723).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='C is supported by the research fund of Jeonbuk National University in 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' REFERENCES [1] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Cho et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Tailoring topological Hall effect in SrRuO3/SrTiO3 superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Acta Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 216, 117153 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [2] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Fang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', The Anomalous Hall Effect and Magnetic Monopoles in Momentum Space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Science 302, 92-95 (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [3] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Symmetry-Driven Spin-Wave Gap Modulation in Nanolayered SrRuO3/SrTiO3 Heterostructures: Implications for Spintronic Applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' ACS Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nano Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 4, 2160-2166 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [4] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Phase Instability amid Dimensional Crossover in Artificial Oxide Crystal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 124, 026401 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [5] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Kang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Ferroelectricity in SrTiO3 epitaxial thin films via Sr-vacancy-induced tetragonality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Surf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 499, 143930 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [6] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Kim et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Superconductivity in (Ba,K)SbO3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 21, 627-633 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [7] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Wang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Epitaxial BiFeO3 Multiferroic Thin Film Heterostructures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Science 299, 1719-1722 (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 10 [8] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Yue et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Anomalous transport in high-mobility superconducting SrTiO3 thin films.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 8, eabl5668 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [9] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' King et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Atomic-scale control of competing electronic phases in ultrathin LaNiO3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nanotechnol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 9, 443-447 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [10] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Tokura & N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nagaosa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Orbital Physics in Transition-Metal Oxides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Science 288, 462-468 (2000).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [11] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Hwang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Emergent phenomena at oxide interfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 11, 103-113 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [12] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Vailionis, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Siemons & G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Koster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Room temperature epitaxial stabilization of a tetragonal phase in ARuO3 (A=Ca and Sr) thin films.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 93, 051909 (2008).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [13] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Lee et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Enhanced electrocatalytic activity via phase transitions in strongly correlated SrRuO3 thin films.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Energy Environ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 10, 924-930 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [14] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Schlom et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Elastic strain engineering of ferroic oxides.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' MRS Bull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 39, 118-130 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [15] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Kan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Tuning magnetic anisotropy by interfacially engineering the oxygen coordination environment in a transition metal oxide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 15, 432-437 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [16] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rondinelli, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' May & J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Freeland.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Control of octahedral connectivity in perovskite oxide heterostructures: An emerging route to multifunctional materials discovery.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' MRS Bull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 37, 261- 270 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [17] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Ramesh & D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Schlom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Creating emergent phenomena in oxide superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Mater.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 4, 257-268 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [18] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Yano & V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Yachandra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' X-ray absorption spectroscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Photosynth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 102, 241 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [19] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Liao et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Large orbital polarization in nickelate-cuprate heterostructures by dimensional control of oxygen coordination.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 10, 589 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [20] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Liu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Interfacial charge-transfer Mott state in iridate–nickelate superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 116, 19863-19868 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [21] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Thickness-dependent orbital hybridization in ultrathin SrRuO3 epitaxial films.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 115, 092906 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [22] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Propagation Control of Octahedral Tilt in SrRuO3 via Artificial Heterostructuring.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 7, 2001643 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [23] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Seo et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Crystalline symmetry-dependent magnon formation in the itinerant ferromagnet SrRuO3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' B 103, 045104 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [24] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Koster et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Structure, physical properties, and applications of SrRuO3 thin films.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Mod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 84, 253-298 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [25] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Unconventional interlayer exchange coupling via chiral phonons in synthetic magnetic oxide heterostructures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 8, eabm4005 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [26] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Spin–phonon coupling in epitaxial SrRuO3 heterostructures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Nanoscale 12, 13926- 13932 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [27] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Jeong, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Seo & W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Choi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Atomistic Engineering of Phonons in Functional Oxide Heterostructures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 9, 2103403 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [28] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Kan & Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Shimakawa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Strain Effect on Structural Transition in SrRuO3 Epitaxial Thin Films.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Cryst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Growth Des.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 11, 5483-5487 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [29] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Chang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Thickness-dependent structural phase transition of strained SrRuO3 ultrathin films: The role of octahedral tilt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' B 84, 104101 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [30] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Zayak, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Huang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Neaton & K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rabe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Structural, electronic, and magnetic properties of SrRuO3 under epitaxial strain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' B 74, 094104 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' [31] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Fan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=', Evolution of element-specific electronic structures in alkaline titanates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' AIP Adv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 9, 065213 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 11 Figure Captions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Structural characterization of atomically designed SRO/STO superlattices: a XRD θ-2θ scans for SRO single film and [8|y] superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' b off-axis XRD scans around (204) plane of STO with different ϕ angles for SRO single film, [8|2], and [8|8] superlattices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' c ao/bo values obtained by off-axis XRD scans and d estimated tilt angle θt as a function of y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The right panel of c,d shows a schematic illustration of RuO6 octahedra with θt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The [hkl]pc and [hkl]o indicates the crystallographic direction in perovskite pseudo-cubic and orthorhombic SRO notation, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' Intensity (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) (a) (b) SRO [8]2] [8]8] Φ=270° 180° -90° [8]8] 10° Intensity (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) 120 122 124 120 122 124 120 122 124 [8]6] 20 (degrees) 20 (degrees) 20 (degrees) [8]4] (c) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='008 [8|2] 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='004 a 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='000 [8]1] (d) 180 11001 [010]。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 170 SRO (~19 nm) [001]pc [-100]] [010]。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 160 20 30 40 50 0 2 4 6 8 2 (degrees) # of STO layers (u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=') 12 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' a Schematic representation of XAS configuration for oxide superlattice with two different detection modes (fluorescence yield and total electron yield).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' b atomic-selective excitation process of XAS for Ru and Ti ions by controlling photon energy of X-ray.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' c O K-edge XAS spectra of [8|y] superlattice taken at θi = 0°.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The vertical lines indicate the peak positions related to Ru-O and Ti-O hybridization as guides to the eye.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 元 polarization a 1) FY (c) X ray O K edge, TEY 2) TEY [8]8] 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='=.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='0° [8]6] Oxide superlattice XAS (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) [8]4] (b) E个 Ru 4d 3d [8|2] Ee [8]1] ~2840 eV SRO ~460 eV 530 540 550 560 Ru 2p3/2 Ti 2p3/2 or 2p1/2 Photon energy (eV) 13 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' a Ti L2,3-edge XAS results of the [8|y] superlattices taken at θi = 0°.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The vertical lines are guides to the eye.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' b Schematic diagram of the Ti 3d orbital states with a possible y-dependent octahedral distortion in the STO layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' (a) (b) Ti L2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='3 edge, TEY Ti 3d 00=0 TA 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='1eVofenhancement XAS (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) [8]8] [8]8] [8|1] [8]6] SrRuO [8]4] SrTiO3 [8|2] [8]1] Possibledistortion inSrTiOlayer 455 460 465 470 Photon energy (eV) 14 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' a XAS spectra at Ru L3-edge for the SRO film and the [8|y] superlattices with a θi = 0° and b θi = 70°.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' c XLD = Ix,y – Iz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' The vertical lines are guides to the eye.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' d Schematic illustration of possible Ru orbital states within the SRO/STO superlattice as a function of θt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' (a) (b) (c) (d) Ru L3 edge, FY 二 Ru L3 edge, FY Ru 4d, t2g 0, = 0° [8]8] Decreasein dxy occupation number XAS (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) [8]6] XAS (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) XLD (arb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content=' units) [8]4] d XZ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} +page_content='VZ [8|2] [8]1] SRO [8]|1] [8]8] Improved distortion 2836 2840 2844 2848 2836 2840 2844 2848 2836 2840 2844 2848 Photon energy (eV) Photon energy (eV) Photon energy (eV)' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNAzT4oBgHgl3EQfcvwO/content/2301.01406v1.pdf'} diff --git a/sNFKT4oBgHgl3EQfJi1O/content/tmp_files/2301.11738v1.pdf.txt b/sNFKT4oBgHgl3EQfJi1O/content/tmp_files/2301.11738v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..6ad8cd719774d83405e74b7c66a2ecb37068705b --- /dev/null +++ b/sNFKT4oBgHgl3EQfJi1O/content/tmp_files/2301.11738v1.pdf.txt @@ -0,0 +1,1199 @@ +arXiv:2301.11738v1 [math.NA] 27 Jan 2023 +Error estimates for a Gaussian rule involving +Bessel functions +Eleonora Denich∗ +Abstract +This paper deals with the estimation of the quadrature error of +a Gaussian formula for weight functions involving fractional powers, +exponentials and Bessel functions of the first kind. For this purpose, +in this work the averaged and generalized averaged Gaussian rules +are employed, together with a tentative a priori approximation of the +error. +The numerical examples confirm the reliability of these ap- +proaches. +Keywords: Gaussian quadrature, averaged Gauss rule, generalized averaged +Gauss rule +MSC 2020: 33C10, 33C45, 65D32 +1 +Introduction +In this work we consider the approximation of integrals of the type +Iν,α,c(f) = +� ∞ +0 +f(x)xαe−cxJν(x)dx, +(1) +where Jν is the Bessel function of the first kind of order ν ≥ 0 (see [21] for +a background), α > −1, c > 0 and f is a smooth function. +We point +out that integrals of type (1) are strongly related to Hankel transforms, +which commonly appear in problems of mathematical physics and applied +∗Dipartimento di Matematica e Geoscienze, Universit`a di Trieste, Trieste, Italy, +eleonora.denich@phd.units.it +1 + +mathematics having axial symmetry. An example of application arises in +geophysical electromagnetic survey. In particular, electromagnetic fields over +a layered earth due to magnetic dipoles above the surface can be expressed +in integral form as in (1) (see [8]). +Since |Jν(x)| ≤ 1, for ν ≥ 0, x ∈ R (see [1, p.362]), a Gaussian quadrature +rule for the computation of integrals of type (1) was constructed in [7] by +rewriting (1) as +IJ +ν,α,c(f) − IL +α,c(f), +with +IJ +ν,α,c(f) := +� ∞ +0 +f(x)xαe−cx[Jν(x) + 1]dx, +(2) +and +IL +α,c(f) := +� ∞ +0 +f(x)xαe−cxdx. +(3) +In this setting, the authors considered the approximations +IJ +ν,α,c(f) ≈ IJ +n(f) +and +IL +α,c(f) ≈ IL +n (f), +where +IJ +n(f) = +n +� +i=1 +w(n) +i +f +� +x(n) +i +� +(4) +is the Gaussian rule relative to the weight function +wν,α,c(x) := xαe−cx[Jν(x) + 1] +on +[0, +∞), +(5) +and +IL +n (f) = +n +� +i=1 +λ(n) +i f +� +ξ(n) +i +� +(6) +is a slight modification of the Gauss-Laguerre quadrature rule, i.e., relative +to the weight function +wα,c(x) = xαe−cx +on +[0, +∞]. +(7) +Clearly, it is possible to consider a change of variable in (3) in order to work +with the standard Laguerre rule, but our choice allows to treat in the same +way integrals (2) and (3). +2 + +Denoting by In(f) the n-point Gaussian rule for the integral +I(f) = +� +∞ +0 +f(x)w(x)dx, +in which w is a generic weight function, it is not easy, in general, to derive +an accurate estimate of the error +En(f) = I(f) − In(f). +(8) +A classical approach is to consider the (2n+1)-point Gauss-Kronrod quadra- +ture rule associated with the n-point Gaussian formula In(f) (see [14, 9, 15]). +However, in [13] the nonexistence of Gauss-Kronrod rules, for n > 2, with +real nodes and positive weights for the Gauss-Laguerre formula was proved. +As consequence, this approach is not suitable for our case. +An alternative approach was proposed by Laurie [12], who introduced the +so-called anti-Gaussian quadrature rule An+1, corresponding to In. It is a +(n + 1)-point formula of degree 2n − 1 which integrates polynomial of degree +up to 2n + 1 with an error equal in magnitude but of opposite sign to one of +the n-point Gaussian formula. Then the idea is to estimate error (8) as +En(f) ≈ ˜A2n+1(f) − In(f), +where the (2n + 1)-point formula +˜A2n+1(f) = 1 +2 (An+1(f) + In(f)) +is commonly named averaged Gaussian formula. +The anti-Gaussian rule +always exists, it is guaranteed to have real nodes and positive weights, and +at most one of the nodes may be outside the integration interval (see [12]). +A more general formula, given by +H2n+1(f) = +1 +2 + γ ((1 + γ)In(f) + An+1(f)) , +γ > 0, +(9) +was considered by Enrich in [6]. In particular, he constructed (9), for the +Laguerre and Hermite weight functions, with the parameter γ chosen to reach +the highest degree of exactness, that is, 2n + 1. We refer to this formula as +the generalized averaged Gaussian rule. +3 + +More recently, Spalevi´c [18] derived a parameter free method for con- +structing generalized averaged Gauss rules for any weight function for which +all moments exist. We denote these quadrature formulas by ˆA2n+1. They +have degree of exactness at least 2n + 1 and are guaranteed to have real +nodes and positive weights. Then the error is estimated as +En(f) ≈ ˆA2n+1(f) − In(f). +(10) +In this paper the averaged Gauss rules ˜AJ +2n+1, ˜AL +2n+1 and generalized +averaged Gauss rules ˆAJ +2n+1, ˆAL +2n+1, corresponding to IJ +n and IL +n , respectively, +are constructed and used to estimate the error +En,ν,α,c(f) = EJ +n,ν,α,c(f) − EL +n,α,c(f), +(11) +where +EJ +n,ν,α,c(f) = IJ +ν,α,c(f) − IJ +n(f) +and +EL +n,α,c(f) = IL +α,c(f) − IL +n (f). +(12) +Unfortunately, the use of the heavy notation E(·) +n,ν,α,c is necessary to avoid +confusion with some general results reported in the paper. +The averaged and generalized averaged Gaussian rules are easy to con- +struct and, moreover, typically lead to quite accurate estimates of (10) (see +[17]). However, sometimes it can be useful to have at disposal an a priori +estimate of the error to have an idea of the number of points necessary to +reach a prescribed accuracy. In this view, here we also present a tentative a +priori approximation of the quadrature error. In particular, by interpreting +wν,α,c as a perturbation of the weight function of the Gauss-Laguerre rule, +the idea is to employ a result due to Barrett [3] and relative to the asymp- +totic behavior of the error of the Gauss-Laguerre formula, to estimate EL +n,α,c +and EJ +n,ν,α,c. Moreover, similarly to the Gauss-Laguerre rule, it can be ver- +ified that the weights of IJ +n decay exponentially. Hence, having at disposal +a reliable error estimate, a truncated approach can also be introduced, but +not considered in the present paper. +Throughout this work we use the symbol ≈ to indicate a generic approx- +imation. The symbol ∼ is used to express the asymptotic equality. +This paper is organized as follows. Section 2 and 3 deals with the rep- +resentation of averaged and generalized averaged Gaussian rules. Moreover, +some theoretical and experimental properties of the quadrature formulas are +described. In Section 4 a tentative error approximation, which allows to have +4 + +an a priori estimate of the error behavior, is presented. Section 5 deals with +the discussion of some numerical examples, which show the performances of +the error estimates. Concluding remarks can be found in Section 6. +2 +Construction of averaged Gaussian rules +Let w be a generic weight function on [0, +∞) and consider the corresponding +Gaussian quadrature formula +In(f) = +n +� +i=1 +σ(n) +i +f +� +τ (n) +i +� +, +(13) +of degree 2n − 1, for the integral +I(f) = +� +∞ +0 +f(x)w(x)dx. +Denoting by αk ∈ R, βk > 0 the recursion coefficients for the sequences of +monic orthogonal polynomials {πj}j≥0 relative to the weight function w, the +Gaussian rule (13) can be associated with the symmetric tridiagonal matrix +Jn = + + +α0 +√β1 +0 +√β1 +α1 +√β2 +√β2 +α2 +... +... +... +� +βn−1 +0 +� +βn−1 +αn−1 + + +∈ Rn×n. +(14) +It is well known that the eigendecomposition of the above matrix provides +the nodes τ (n) +i +and the weights σ(n) +i +, i = 1, . . . , n, of the Gaussian rule In (see +e.g. [4] and the references therein). Now, the corresponding (n + 1)-point +anti-Gaussian quadrature formula +An+1(f) := +n+1 +� +i=1 +˜σ(n+1) +i +f +� +˜τ (n+1) +i +� +, +(15) +is such that +I(p) − An+1(p) = − (I(p) − In(p)) , +∀p ∈ P2n+1, +(16) +5 + +where P2n+1 denotes the space of polynomials of degree at most 2n+1. Laurie +[12] showed that formula (15) is associated with the symmetric tridiagonal +matrix ˜Jn+1 ∈ R(n+1)×(n+1) defined by +˜Jn+1 = +� +Jn +en +√2βn +eT +n +√2βn +αn +� +, +(17) +where en = (0, . . . , 0, 1)T ∈ Rn. Therefore, having at disposal the recurrence +coefficients αk and βk, it is trivial to compute the nodes ˜τ (n+1) +i +and the weights +˜σ(n+1) +i +, i = 1, . . . , n + 1, of the anti-Gaussian rule. Moreover, quadrature +formula (15) has the following properties (see [12, Theorem 1]): +1. ˜σ(n+1) +i +> 0, i = 1, . . . , n + 1; +2. the nodes ˜τ (n+1) +i +, i = 1, . . . , n + 1, are all real and are interlaced by +those of the Gaussian formula In, that is, +˜τ (n+1) +1 +< τ (n) +1 +< ˜τ (n+1) +2 +< . . . < τ (n) +n +< ˜τ (n+1) +n+1 ; +3. ˜τ (n+1) +i +∈ [0, +∞), for i ≥ 2; +4. ˜τ (n+1) +1 +∈ [0, +∞) if and only if +πn+1(0) +πn−1(0) ≥ βn, +n ≥ 1. +(18) +At this point, the averaged quadrature formula ˜A2n+1 is defined as +˜A2n+1(f) := 1 +2 (In(f) + An+1(f)) . +From property (16) it follows that the degree of exactness of ˜A2n+1 is 2n+ 1, +and the quadrature error can be estimated by +En(f) ≈ ˜A2n+1(f) − In(f) = 1 +2 (An+1(f) − In(f)) . +Going back to our case, we denote by αJ +k, βJ +k and αL +k , βL +k the recursion +coefficients for the sequences of monic orthogonal polynomials relative to +wν,α,c and wα,c (see (5)-(7)), respectively, and by ˜JJ +n+1, ˜JL +n+1 the associated +6 + +tridiagonal matrices of type (17). In particular, we have that αL +k , βL +k are +strongly related to the recurrence coefficients ak, bk of the standard general- +ized Gauss-Laguerre rule. Indeed, it can be easily verified that +αL +k = ak +c +and +βL +k = bk +c2 . +(19) +Moreover, the monic polynomials +� +L(α,c) +k +� +k≥0 defined by +L(α,c) +k +(x) = 1 +ck L(α) +k (cx), +(20) +where L(α) +k (t) is the monic generalized Laguerre polynomial of degree k, are +orthogonal with respect to the weight function wα,c. At this point, we denote +by +AJ +n+1(f) = +n+1 +� +i=1 +˜w(n+1) +i +f +� +˜x(n+1) +i +� +(21) +and +AL +n+1(f) = +n+1 +� +i=1 +˜λ(n+1) +i +f +� +˜ξ(n+1) +i +� +(22) +the anti-Gaussian quadrature rules (cf. (15)) relative to IJ +n and IL +n (see (4) +and (6)), respectively. Then, the corresponding averaged Gauss rules are +˜AJ +2n+1(f) = 1 +2 +� +IJ +n(f) + AJ +n+1(f) +� +and +˜AL +2n+1(f) = 1 +2 +� +IL +n (f) + AL +n+1(f) +� +. +Finally, the error of the Gaussian quadrature formula (see (11)) is estimated +as +En,ν,α,c(f) ≈ ˜En,ν,α,c(f) := ˜EJ +n,ν,α,c(f) − ˜EL +n,α,c(f), +(23) +where +˜EJ +n,ν,α,c(f) = ˜AJ +2n+1(f) − IJ +n(f) +and +˜EL +n,α,c(f) = ˜AL +2n+1(f) − IL +n (f). +For the generalized Gauss-Laguerre rule the recurrence coefficients and +the values of the orthogonal polynomials are explicitly known, and condition +7 + +(18) holds (see [12, Theorem 4]). +Moreover, by using relations (19) and +(20) it is not difficult to prove the same result also for the Gaussian rule +IL +n . Therefore, we have that the anti-Gaussian formula AL +n+1 is internal, i.e., +˜ξ(n+1) +1 +∈ [0, +∞). +For what concerns the anti-Gaussian rule AJ +n+1, we do not have at dis- +posal the recurrence coefficients and the expressions of the corresponding +orthogonal polynomials (see [7]). Hence, condition (18) can only be verified +numerically. In our numerical experiments, independently of ν, c and work- +ing with n = 100, we have observed negative values of ˜x(n+1) +1 +for −1 < α < ˜α, +where ˜α ∈ (−0.8, −0.7). +3 +Construction of generalized averaged Gauss +rules +In this section we describe the generalized averaged Gauss rule ˆA2n+1, intro- +duced in [18], associated with a generic Gauss formula In. It is a (2n+1)-point +formula which can be represented by a single symmetric tridiagonal matrix +ˆJ2n+1 ∈ R(2n+1)×(2n+1), that is, +ˆJ2n+1 = + + +Jn +√βnen +0 +√βneT +n +αn +� +βn+1eT +1 +0 +� +βn+1e1 +J′ +n + + , +where e1 = (1, 0, . . . , 0)T ∈ Rn, Jn is as in (14) and J′ +n is obtained by reversing +the order of the rows and column of Jn, that is, +J′ +n = + + +αn−1 +� +βn−1 +0 +� +βn−1 +αn−2 +� +βn−2 +... +... +... +√β2 +α1 +√β1 +0 +√β1 +α0 + + +∈ Rn×n. +The generalized averaged Gauss formula can also be described in a more +compact form. Indeed, ˆA2n+1 can be written as +ˆA2n+1(f) = +βn+1 +βn + βn+1 +In(f) + +βn +βn + βn+1 +¯An+1(f), +(24) +8 + +where the quadrature formula +¯An+1(f) = +n+1 +� +i=1 +¯σ(n+1) +i +f +� +¯τ (n+1) +i +� +(25) +is determined by the symmetric tridiagonal matrix ¯Jn+1 ∈ R(n+1)×(n+1), de- +fined as +¯Jn+1 = +� +Jn +en +� +βn + βn+1 +eT +n +� +βn + βn+1 +αn +� +, +(26) +(see [16]). By construction, the (n + 1)-point quadrature rule (25) has es- +sentially the same properties of the anti-Gaussian rule An+1, but in this case +the nodes are internal, i.e., ¯τ (n+1) +1 +∈ [0, +∞), if and only if +πn+1(0) +πn−1(0) ≥ βn+1, +n ≥ 1, +(27) +(see [18]). +At this point, we denote the generalized averaged Gaussian rules corre- +sponding to IJ +n and IL +n by +ˆAJ +2n+1(f) = +βJ +n+1 +βJ +n + βJ +n+1 +IJ +n(f) + +βJ +n +βJ +n + βJ +n+1 +¯AJ +n+1(f) +and +ˆAL +2n+1(f) = +βL +n+1 +βL +n + βL +n+1 +IL +n (f) + +βL +n +βL +n + βL +n+1 +¯AL +n+1(f), +where the formulas +¯AJ +n+1(f) = +n+1 +� +i=1 +¯w(n+1) +i +f +� +¯x(n+1) +i +� +and +¯AL +n+1(f) = +1 +cα+1 +n+1 +� +i=1 +¯λ(n+1) +i +f +� +¯ξ(n+1) +i +� +are associated with the tridiagonal matrices ¯JJ +n+1 and ¯JL +n+1, obtained by con- +sidering in (26) the recursion coefficients αJ +k, βJ +k and αL +k , βL +k , respectively. +Finally, the error of the Gaussian quadrature (see (11)) is estimated as +En,ν,α,c(f) ≈ ˆEn,ν,α,c(f) := ˆEJ +n,ν,α,c(f) − ˆEL +n,α,c(f), +(28) +where +ˆEJ +n,ν,α,c(f) = ˆAJ +2n+1(f) − IJ +n(f) +and +ˆEL +n,α,c(f) = ˆAL +2n+1(f) − IL +n (f). +9 + +From (27) and by using again relations (19) and (20), it is trivial to prove that +for the generalized averaged Gauss rule ¯AL +n+1 the condition for the internality +is α ≥ 1. +This means that for −1 < α < 1 the smallest node of ¯AL +n+1 +is negative. As before, the behavior of the rule ¯AJ +n+1 can only be verified +numerically. In particular, independently of ν, c and working with n = 100, +we have found negative values of ¯x(n+1) +1 +for −1 < α < ¯α, with ¯α ∈ (1, 1.1). +4 +A tentative a priori estimate +In the previous sections we have described how, having at disposal the re- +currence coefficients of the corresponding orthogonal polynomials, the av- +eraged and generalized averaged Gaussian rules can be easily constructed +and employed to approximate the quadrature error En,ν,α,c(f) (see (11)). In +this section, by using the results of Barrett [3] regarding the derivation of +an asymptotic expression for the error of the Gauss-Laguerre formula, we +present a tentative a priori estimate of the error En,ν,α,c(f). In particular, +as remarked in the introduction, since |Jν(x)| ≤ 1, for ν ≥ 0, x ∈ R, we +can interpret the weight function wν,α,c as a perturbation of the weight of +the Gauss-Laguerre rule. Therefore, the idea is to employ the error estimate +for the Gauss-Laguerre formula to approximate not only EL +n,α,c(f), but also +EJ +n,ν,α,c(f) (see (12)). First, let consider the error of the Laguerre rule for +the classical case of c = 1, that is, EL +n,α,1(f). For any given R > 1, the set +ΓR = +� +z ∈ C | ℜ√−z = ln R +� +represents a parabola of the complex plane +positively oriented, symmetric with respect to the real axis and with vertex +at − ln R. Barrett [3] showed that, if f(z) has no singularities on or within +ΓR, except for a pair of simple poles z0 and its conjugate ¯z0, then, for n → ∞, +EL +n,α,1(f) ∼ 4πℜ +� +Res(f(z), z0)e−iαπ � +exp √−z0 +�−2√¯n� +, +(29) +where ¯n = 4n + α + 2 and the symbol Res(·, ·) denotes the residue. Formula +(29) follows from the fact that EL +n,α.1(f) can be represented as the contour +integral +EL +n,α,1(f) = +1 +2πi +� +Γ +qL +n(z) +L(α) +n (z) +f(z)dz, +10 + +where L(α) +n (z) is the generalized Laguerre polynomial, qL +n(z) is the associated +function defined by +qL +n(z) = +� +∞ +0 +wα(x)L(α) +n (x) +z − x +dx, +with wα(x) = xαe−x and Γ is a contour containing [0, +∞) and such that no +singularity of f(z) lies on or within the contour (see [19]). Then, by using +the relation (see [5]) +qL +n(z) +L(α) +n (z) +∼ −2e−iπαzαe−z Kα +�√ +¯nze−iπ +� +Iα +�√ +¯nze−iπ +� , +(30) +where Iα, Kα are the modified Bessel functions of order α of the first and +second kind, respectively, and the asymptotic formulas (see [1, p.377, 9.7.1- +9.7.2]) +Iα(t) ∼ +et +√ +2πt, Kα(t) ∼ e−t +� π +2t, +valid for large |t|, | arg(t)| < π/2, we obtain +qL +n(z) +L(α) +n (z) +∼ −2πe−iαπzαe−z � +exp +�√ +−z +��−2√¯n =: Φn(z), +z /∈ [0, +∞). (31) +Finally, formula (29) can be derived by choosing Γ = ΓR ∪C1 ∪C2, where C1, +C2 are two arbitrary small circles surrounding the two poles, and by using +relation (31). After simple computations, it can be verified that, by replacing +wα(x) with wα,c(x), one obtains +EL +n,α,c(f) ∼ 4π +cα−1ℜ +� +Res(f(z), z0)e−iαπ � +exp √−cz0 +�−2√¯n� +. +(32) +Now, let consider the error EJ +n,ν,α,c(f). +It can also be written as the +contour integral +EJ +n,ν,α,c(f) = +1 +2πi +� +Γ +qJ +n(z) +pn(z)f(z)dz, +where pn(z) is the orthogonal polynomial relative to the weight function wν,α,c +and +qJ +n(z) = +� +∞ +0 +wν,α,c(x)pn(x) +z − x +dx. +11 + +However, as remarked before, we do not have at disposal the analytic expres- +sion of pn(z) and, therefore, an asymptotic formula analog to (31) can not +be derived. In order to justify the use of estimate (32) also for EJ +n,ν,α,c(f), +we numerically evaluate the functions +ΨJ +n(z) := qJ +n(z) +pn(z) +and +ΨL +n(z) := qL +n(z) +L(α) +n (z) +(33) +and check if the approximation Φn(z) (cf. (31)) can also be used for ΨJ +n(z). +In Figures 1-2, for different values of n, ν, α, c, we plot the ratios +ΨJ +n(z) +Φn(z) +and +ΨL +n(z) +Φn(z) , +in which z = reiθπ, with r = 4, θ ∈ (0, 2). We remark that ΨL +n(z) +Φn(z) ∼ 1, for +n → ∞, by (31). The results show that approximation (31) works rather good +also for ΨJ +n(z), and the situation is analog for other values of the parameters. +Hence, the idea is to use the approximation (32) also for EJ +n,ν,α,c(f). Finally, +since En,ν,α,c(f) = EJ +n,ν,α,c(f) − EL +n,α,c(f), we consider the estimate +|En,ν,α,c(f)| ≈ 2En,α,c(f), +(34) +with +En,α,c(f) := 4π +cα−1 +���Res(f(z), z0) +� +exp √−cz0 +�−2√¯n��� , +(35) +(cf. (32)). +5 +Numerical examples +In this section we present some numerical experiments which confirm the +reliability of the error estimates (23), (28) and (34). We remark that all the +computations reported in this work are carried out in Matlab by using high- +precision arithmetic, specifically with 120 significant decimal digits. +The +main reason for this choice is that we do not known explicitly the recurrence +coefficients αJ +k, βJ +k and hence we are forced to employ a numerical scheme +to derive them. However, it is well known (see e.g. [10]) that this compu- +tation can be inaccurate for growing k, because the problem is severely ill +conditioned. Therefore, even if in [7] an alternative more stable approach +12 + +0 +0.5 +1 +1.5 +2 +0.8 +0.85 +0.9 +0.95 +1 +1.05 +1.1 +1.15 +1.2 +1.25 +1.3 +0 +0.5 +1 +1.5 +2 +0.8 +0.85 +0.9 +0.95 +1 +1.05 +1.1 +1.15 +1.2 +1.25 +1.3 +Figure 1: The ratios ΨJ +n(z) +Φn(z) (left) and ΨL +n(z) +Φn(z) (right) for z = reiθπ, with r = 4, +θ ∈ (0, 2). In this case ν = 1, c = 0.5, α = 0.3. +0 +0.5 +1 +1.5 +2 +0.6 +0.8 +1 +1.2 +1.4 +1.6 +1.8 +0 +0.5 +1 +1.5 +2 +0.6 +0.8 +1 +1.2 +1.4 +1.6 +1.8 +Figure 2: The ratios ΨJ +n(z) +Φn(z) (left) and ΨL +n(z) +Φn(z) (right) for z = reiθπ, with r = 4, +θ ∈ (0, 2). In this case ν = 0, c = 1, α = −0.5. +13 + +is presented, the use of high-precision arithmetic allows to considerably in- +crease the number of quadrature points and to obtain an absolute error in the +approximations of the order of the machine precision. The Matlab routine +for the computation of the recurrence coefficients αJ +k, βJ +k is taken from [11], +while the code for the implementation of the Gauss-Laguerre quadrature rule +from [20]. +Example 1 Consider the integral +Iν,α,c(f) = +� ∞ +0 +f(x)xαe−cxJν(x)dx, +with +f(x) = +1 +1 + e−x. +In order to use error estimate (34), we start by studing the poles of f. A +simple analysis shows that they are given by the set +zk = −i(π + 2kπ), +k ∈ Z, +and the closest to the real axis are z0 and its conjugate z1, that is, ±iπ. As +for the residue (cf, (32)), we obtain +Res (f(z), z0) = 1. +Therefore, by using (34) and (35), we have that +|En,ν,α,c(f)| ≈ 8πc1−αe− +√ +2cπ¯n. +(36) +In Figure 3 we compare, for different values of ν, α, c, the quadrature error +|En,ν,α,c(f)|, obtained by considering a reference solution, with the approx- +imations | ˜En,ν,α,c(f)|, | ˆEn,ν,α,c(f)| (cf. +(23)-(28)) and estimate (36). +We +can see the very good agreement between the error and the approximations +given by the averaged and generalized averaged Gaussian rules. Moreover, +the examples reveal that the a priori estimate (36) is rather accurate. +Example 2 Consider now the case of +f(x) = +1 +1 + x2. +14 + +0 +20 +40 +60 +80 +10-14 +10-12 +10-10 +10-8 +10-6 +10-4 +10-2 +100 +error +averaged Gauss +generalized averaged Gauss +a priori estimate +0 +20 +40 +60 +80 +10-20 +10-15 +10-10 +10-5 +100 +Figure 3: +The quadrature error and its approximations | ˜En,ν,α,c(f)|, +| ˆEn,ν,α,c(f)| and (36) for ν = 1, c = 0.5, α = 1.7 (left) and ν = 0.5, c = +0.8, α = 1.5 (right). +This function has only two poles ±i. As for the residue (cf. (32)), we obtain +Res (f(z), i) = − i +2. +Therefore, by using (34) and (35), we have that +|En,ν,α,c(f)| ≈ 4πc1−αe− +√ +2cπ¯n. +(37) +In Figure 4 we compare, for different values of ν, α, c, the quadrature error +|En,ν,α,c(f)|, obtained by considering a reference solution, with the approxi- +mations | ˜En,ν,α,c(f)|, | ˆEn,ν,α,c(f)| (cf. (23)-(28)) and estimate (37). +6 +Conclusions +In this work the error estimates of the Gaussian quadrature formula intro- +duced in [7] are considered. In particular, a posteriori error approximations +given by the averaged and generalized averaged Gaussian rules have been +15 + +0 +20 +40 +60 +80 +10-15 +10-10 +10-5 +100 +error +averaged Gauss +generalized averaged Gauss +a priori estimate +0 +20 +40 +60 +80 +10-12 +10-10 +10-8 +10-6 +10-4 +10-2 +100 +Figure 4: +The quadrature error and its approximations | ˜En,ν,α,c(f)|, +| ˆEn,ν,α,c(f)| and (37) for ν = 1, c = 1.5, α = 1 (left) and ν = 0, c = 1, α = 1.5 +(right). +constructed and showed to be very accurate. Moreover, starting from nu- +merical experiments regarding the ratio qJ +n/pn (see (33)), an heuristic but +quite effective a priori error estimate has been introduced. We remark that, +having at disposal an a-priori estimate of the error and by noting that, simi- +lar to the Gauss-Laguerre rule, the weights w(n) +i +, i = 1, . . . , n, (cf. (4)) decay +exponentially (see Figure 5), a truncated approach can also be introduced, +to reduce the number of function evaluation (see [2]). +Acknowledgements +This work was partially supported by GNCS-INdAM and CINECA under +HPC-TRES program award number 2019-04. The author is member of the +INdAM research group GNCS. +16 + +0 +5 +10 +15 +20 +25 +30 +10-50 +10-40 +10-30 +10-20 +10-10 +100 +0 +5 +10 +15 +20 +25 +30 +10-50 +10-40 +10-30 +10-20 +10-10 +100 +1010 +Figure 5: The weights w(n) +i +, ξ(n) +i +, i = 1, . . . , n (cf. (4)-(6)) for ν = 1, α = 0.7, +c = 0.5 (left) and ν = 0, α = −0.5, c = 0.8 (right). In both cases n = 30. +References +[1] +M. Abramowitz and I. A. Stegun (1970), Handbook of Mathematical +Functions with Formulas, Graphs, and Mathematical Tables, 7th Edi- +tion, Dover Publications, Inc., New York. +[2] +L. Aceto and P. Novati, Fast and accurate approximations to fractional +powers of operators, IMA Journal of Numerical Analysis 42 (2022), +1598–1622. +[3] +W. Barrett, Convergence properties of Gaussian quadrature formulae, +Comput. J. 3 (1960/1961), 272–277. +[4] +P. J. Davis and P. Rabinowitz (1975), Methods of numerical integration, +Academic Press, Inc., New York. +[5] +J. D. Donaldson and D. Elliott, A Unified Approach to Quadrature +Rules with Asymptotic Estimates of Their Remainders, SIAM Journal +on Numerical Analysis 9 (1972), 573-602. +17 + +[6] +S. Ehrich, On stratified extensions of Gauss-Laguerre and Gauss- +Hermite quadrature formulas, J. Comput. Appl. Math. 140 (2002), 291- +299. +[7] +E. +Denich +and +P. +Novati, +Gaussian +rule +for +integrals +involving +Bessel +functions, +arXiv:2110.05051v1 +(2021), +https://doi.org/10.48550/arXiv.2110.05051. +[8] +E. Denich, P. Novati and S. Picotti, A fast and accurate numerical ap- +proach for electromagnetic inversion, Journal of Computational Physics +475 (2023), 111846. +[9] +W. Gautschi, Gauss-Kronrod quadrature—A survey, G.V. Milovanovi´c +(Ed.), Numerical Methods and Approximation Theory III, Faculty of +Electronic Engineering, Univ. Niˇs (1988), 39-66 +[10] W. Gautschi, Orthogonal polynomials: applications and computation, +Acta Numerica (1996), Cambridge University Press, 45-119. +[11] W. Gautschi, Orthogonal Polynomials, Quadrature, and Approximation: +Computational Methods and Software (in Matlab), In: Marcell´an F., Van +Assche W. (eds) Orthogonal Polynomials and Special Functions, Lecture +Notes in Mathematics, vol 1883, Springer, Berlin, Heidelberg, 2006. doi: +10.1007/978 − 3 − 540 − 36716 − 1 1 +[12] D. P. Laurie, Anti-Gaussian quadrature formulas, Math. Comp. 65 +(1996), 739-747. +[13] D. K. Kahaner and G. Monegato, Nonexistence of extended Gauss- +Laguerre and Gauss-Hermite quadrature rules with positive weights, Z. +Angew. Math. Phys. 29 (1978), 983-986. +[14] A. S. Kronrod, Integration with control of accuracy (Russian), Dokl. +Acad. Nauk SSSR 154 (1964), 283-286. +[15] S. E. Notaris, Gauss-Kronrod quadrature formulae. A survey of fifty +years of research, ETNA, Electron. Trans. Numer. Anal. 45 (2016), 371- +404. +[16] L. Reichel and M. M. Spalevi´c, A new representation of generalized +averaged Gauss quadrature rules, Applied Numerical Mathematics 165 +(2021), 614-619. +18 + +[17] L. Reichel and M. M. Spalevi´c, Averaged Gauss quadrature formu- +las: Properties and applications, Journal of Computational and Applied +Mathematics 410 (2022), 114232. +[18] M. M. Spalevi´c, On generalized averaged Gaussian formulas, Math. +Comp. 76 (2007), 1483-1492. +[19] G. Szego (1939), Orthogonal polynomials, American Mathematical So- +ciety Colloquium Publications, Vol. XXIII. +[20] N. Trefethen, Approximation Theory and Approximation Practice, Ex- +tended Edition, SIAM, Philadelphia, 2019. +[21] Watson, G. N. (1966), A treatise on the theory of Bessel functions, Cam- +bridge University press, Marcell´an F., Van Assche W. (eds). +19 + diff --git a/sNFKT4oBgHgl3EQfJi1O/content/tmp_files/load_file.txt b/sNFKT4oBgHgl3EQfJi1O/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..eef5a3dbf542792960b03c820c0d2a611cab7827 --- /dev/null +++ b/sNFKT4oBgHgl3EQfJi1O/content/tmp_files/load_file.txt @@ -0,0 +1,368 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf,len=367 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='11738v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='NA] 27 Jan 2023 Error estimates for a Gaussian rule involving Bessel functions Eleonora Denich∗ Abstract This paper deals with the estimation of the quadrature error of a Gaussian formula for weight functions involving fractional powers, exponentials and Bessel functions of the first kind.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' For this purpose, in this work the averaged and generalized averaged Gaussian rules are employed, together with a tentative a priori approximation of the error.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The numerical examples confirm the reliability of these ap- proaches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Keywords: Gaussian quadrature, averaged Gauss rule, generalized averaged Gauss rule MSC 2020: 33C10, 33C45, 65D32 1 Introduction In this work we consider the approximation of integrals of the type Iν,α,c(f) = � ∞ 0 f(x)xαe−cxJν(x)dx, (1) where Jν is the Bessel function of the first kind of order ν ≥ 0 (see [21] for a background), α > −1, c > 0 and f is a smooth function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We point out that integrals of type (1) are strongly related to Hankel transforms, which commonly appear in problems of mathematical physics and applied ∗Dipartimento di Matematica e Geoscienze, Universit`a di Trieste, Trieste, Italy, eleonora.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='denich@phd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='units.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='it 1 mathematics having axial symmetry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' An example of application arises in geophysical electromagnetic survey.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, electromagnetic fields over a layered earth due to magnetic dipoles above the surface can be expressed in integral form as in (1) (see [8]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Since |Jν(x)| ≤ 1, for ν ≥ 0, x ∈ R (see [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='362]), a Gaussian quadrature rule for the computation of integrals of type (1) was constructed in [7] by rewriting (1) as IJ ν,α,c(f) − IL α,c(f), with IJ ν,α,c(f) := � ∞ 0 f(x)xαe−cx[Jν(x) + 1]dx, (2) and IL α,c(f) := � ∞ 0 f(x)xαe−cxdx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (3) In this setting, the authors considered the approximations IJ ν,α,c(f) ≈ IJ n(f) and IL α,c(f) ≈ IL n (f), where IJ n(f) = n � i=1 w(n) i f � x(n) i � (4) is the Gaussian rule relative to the weight function wν,α,c(x) := xαe−cx[Jν(x) + 1] on [0, +∞), (5) and IL n (f) = n � i=1 λ(n) i f � ξ(n) i � (6) is a slight modification of the Gauss-Laguerre quadrature rule, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', relative to the weight function wα,c(x) = xαe−cx on [0, +∞].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (7) Clearly, it is possible to consider a change of variable in (3) in order to work with the standard Laguerre rule, but our choice allows to treat in the same way integrals (2) and (3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 2 Denoting by In(f) the n-point Gaussian rule for the integral I(f) = � +∞ 0 f(x)w(x)dx, in which w is a generic weight function, it is not easy, in general, to derive an accurate estimate of the error En(f) = I(f) − In(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (8) A classical approach is to consider the (2n+1)-point Gauss-Kronrod quadra- ture rule associated with the n-point Gaussian formula In(f) (see [14, 9, 15]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' However, in [13] the nonexistence of Gauss-Kronrod rules, for n > 2, with real nodes and positive weights for the Gauss-Laguerre formula was proved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' As consequence, this approach is not suitable for our case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' An alternative approach was proposed by Laurie [12], who introduced the so-called anti-Gaussian quadrature rule An+1, corresponding to In.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' It is a (n + 1)-point formula of degree 2n − 1 which integrates polynomial of degree up to 2n + 1 with an error equal in magnitude but of opposite sign to one of the n-point Gaussian formula.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Then the idea is to estimate error (8) as En(f) ≈ ˜A2n+1(f) − In(f), where the (2n + 1)-point formula ˜A2n+1(f) = 1 2 (An+1(f) + In(f)) is commonly named averaged Gaussian formula.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The anti-Gaussian rule always exists, it is guaranteed to have real nodes and positive weights, and at most one of the nodes may be outside the integration interval (see [12]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' A more general formula, given by H2n+1(f) = 1 2 + γ ((1 + γ)In(f) + An+1(f)) , γ > 0, (9) was considered by Enrich in [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, he constructed (9), for the Laguerre and Hermite weight functions, with the parameter γ chosen to reach the highest degree of exactness, that is, 2n + 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We refer to this formula as the generalized averaged Gaussian rule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 3 More recently, Spalevi´c [18] derived a parameter free method for con- structing generalized averaged Gauss rules for any weight function for which all moments exist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We denote these quadrature formulas by ˆA2n+1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' They have degree of exactness at least 2n + 1 and are guaranteed to have real nodes and positive weights.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Then the error is estimated as En(f) ≈ ˆA2n+1(f) − In(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (10) In this paper the averaged Gauss rules ˜AJ 2n+1, ˜AL 2n+1 and generalized averaged Gauss rules ˆAJ 2n+1, ˆAL 2n+1, corresponding to IJ n and IL n , respectively, are constructed and used to estimate the error En,ν,α,c(f) = EJ n,ν,α,c(f) − EL n,α,c(f), (11) where EJ n,ν,α,c(f) = IJ ν,α,c(f) − IJ n(f) and EL n,α,c(f) = IL α,c(f) − IL n (f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (12) Unfortunately, the use of the heavy notation E(·) n,ν,α,c is necessary to avoid confusion with some general results reported in the paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The averaged and generalized averaged Gaussian rules are easy to con- struct and, moreover, typically lead to quite accurate estimates of (10) (see [17]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' However, sometimes it can be useful to have at disposal an a priori estimate of the error to have an idea of the number of points necessary to reach a prescribed accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In this view, here we also present a tentative a priori approximation of the quadrature error.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, by interpreting wν,α,c as a perturbation of the weight function of the Gauss-Laguerre rule, the idea is to employ a result due to Barrett [3] and relative to the asymp- totic behavior of the error of the Gauss-Laguerre formula, to estimate EL n,α,c and EJ n,ν,α,c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Moreover, similarly to the Gauss-Laguerre rule, it can be ver- ified that the weights of IJ n decay exponentially.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Hence, having at disposal a reliable error estimate, a truncated approach can also be introduced, but not considered in the present paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Throughout this work we use the symbol ≈ to indicate a generic approx- imation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The symbol ∼ is used to express the asymptotic equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' This paper is organized as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Section 2 and 3 deals with the rep- resentation of averaged and generalized averaged Gaussian rules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Moreover, some theoretical and experimental properties of the quadrature formulas are described.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In Section 4 a tentative error approximation, which allows to have 4 an a priori estimate of the error behavior, is presented.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Section 5 deals with the discussion of some numerical examples, which show the performances of the error estimates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Concluding remarks can be found in Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 2 Construction of averaged Gaussian rules Let w be a generic weight function on [0, +∞) and consider the corresponding Gaussian quadrature formula In(f) = n � i=1 σ(n) i f � τ (n) i � , (13) of degree 2n − 1, for the integral I(f) = � +∞ 0 f(x)w(x)dx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Denoting by αk ∈ R, βk > 0 the recursion coefficients for the sequences of monic orthogonal polynomials {πj}j≥0 relative to the weight function w, the Gaussian rule (13) can be associated with the symmetric tridiagonal matrix Jn = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 α0 √β1 0 √β1 α1 √β2 √β2 α2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' � βn−1 0 � βn−1 αn−1 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fb ∈ Rn×n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (14) It is well known that the eigendecomposition of the above matrix provides the nodes τ (n) i and the weights σ(n) i , i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , n, of the Gaussian rule In (see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [4] and the references therein).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Now, the corresponding (n + 1)-point anti-Gaussian quadrature formula An+1(f) := n+1 � i=1 ˜σ(n+1) i f � ˜τ (n+1) i � , (15) is such that I(p) − An+1(p) = − (I(p) − In(p)) , ∀p ∈ P2n+1, (16) 5 where P2n+1 denotes the space of polynomials of degree at most 2n+1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Laurie [12] showed that formula (15) is associated with the symmetric tridiagonal matrix ˜Jn+1 ∈ R(n+1)×(n+1) defined by ˜Jn+1 = � Jn en √2βn eT n √2βn αn � , (17) where en = (0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , 0, 1)T ∈ Rn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Therefore, having at disposal the recurrence coefficients αk and βk, it is trivial to compute the nodes ˜τ (n+1) i and the weights ˜σ(n+1) i , i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , n + 1, of the anti-Gaussian rule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Moreover, quadrature formula (15) has the following properties (see [12, Theorem 1]): 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' ˜σ(n+1) i > 0, i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , n + 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' the nodes ˜τ (n+1) i , i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , n + 1, are all real and are interlaced by those of the Gaussian formula In, that is, ˜τ (n+1) 1 < τ (n) 1 < ˜τ (n+1) 2 < .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' < τ (n) n < ˜τ (n+1) n+1 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' ˜τ (n+1) i ∈ [0, +∞), for i ≥ 2;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' ˜τ (n+1) 1 ∈ [0, +∞) if and only if πn+1(0) πn−1(0) ≥ βn, n ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (18) At this point, the averaged quadrature formula ˜A2n+1 is defined as ˜A2n+1(f) := 1 2 (In(f) + An+1(f)) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' From property (16) it follows that the degree of exactness of ˜A2n+1 is 2n+ 1, and the quadrature error can be estimated by En(f) ≈ ˜A2n+1(f) − In(f) = 1 2 (An+1(f) − In(f)) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Going back to our case, we denote by αJ k, βJ k and αL k , βL k the recursion coefficients for the sequences of monic orthogonal polynomials relative to wν,α,c and wα,c (see (5)-(7)), respectively, and by ˜JJ n+1, ˜JL n+1 the associated 6 tridiagonal matrices of type (17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, we have that αL k , βL k are strongly related to the recurrence coefficients ak, bk of the standard general- ized Gauss-Laguerre rule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Indeed, it can be easily verified that αL k = ak c and βL k = bk c2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (19) Moreover, the monic polynomials � L(α,c) k � k≥0 defined by L(α,c) k (x) = 1 ck L(α) k (cx), (20) where L(α) k (t) is the monic generalized Laguerre polynomial of degree k, are orthogonal with respect to the weight function wα,c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' At this point, we denote by AJ n+1(f) = n+1 � i=1 ˜w(n+1) i f � ˜x(n+1) i � (21) and AL n+1(f) = n+1 � i=1 ˜λ(n+1) i f � ˜ξ(n+1) i � (22) the anti-Gaussian quadrature rules (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (15)) relative to IJ n and IL n (see (4) and (6)), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Then, the corresponding averaged Gauss rules are ˜AJ 2n+1(f) = 1 2 � IJ n(f) + AJ n+1(f) � and ˜AL 2n+1(f) = 1 2 � IL n (f) + AL n+1(f) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Finally, the error of the Gaussian quadrature formula (see (11)) is estimated as En,ν,α,c(f) ≈ ˜En,ν,α,c(f) := ˜EJ n,ν,α,c(f) − ˜EL n,α,c(f), (23) where ˜EJ n,ν,α,c(f) = ˜AJ 2n+1(f) − IJ n(f) and ˜EL n,α,c(f) = ˜AL 2n+1(f) − IL n (f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' For the generalized Gauss-Laguerre rule the recurrence coefficients and the values of the orthogonal polynomials are explicitly known, and condition 7 (18) holds (see [12, Theorem 4]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Moreover, by using relations (19) and (20) it is not difficult to prove the same result also for the Gaussian rule IL n .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Therefore, we have that the anti-Gaussian formula AL n+1 is internal, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', ˜ξ(n+1) 1 ∈ [0, +∞).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' For what concerns the anti-Gaussian rule AJ n+1, we do not have at dis- posal the recurrence coefficients and the expressions of the corresponding orthogonal polynomials (see [7]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Hence, condition (18) can only be verified numerically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In our numerical experiments, independently of ν, c and work- ing with n = 100, we have observed negative values of ˜x(n+1) 1 for −1 < α < ˜α, where ˜α ∈ (−0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8, −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='7).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 3 Construction of generalized averaged Gauss rules In this section we describe the generalized averaged Gauss rule ˆA2n+1, intro- duced in [18], associated with a generic Gauss formula In.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' It is a (2n+1)-point formula which can be represented by a single symmetric tridiagonal matrix ˆJ2n+1 ∈ R(2n+1)×(2n+1), that is, ˆJ2n+1 = \uf8ee \uf8f0 Jn √βnen 0 √βneT n αn � βn+1eT 1 0 � βn+1e1 J′ n \uf8f9 \uf8fb , where e1 = (1, 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , 0)T ∈ Rn, Jn is as in (14) and J′ n is obtained by reversing the order of the rows and column of Jn, that is, J′ n = \uf8ee \uf8ef\uf8ef\uf8ef\uf8ef\uf8ef\uf8f0 αn−1 � βn−1 0 � βn−1 αn−2 � βn−2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' √β2 α1 √β1 0 √β1 α0 \uf8f9 \uf8fa\uf8fa\uf8fa\uf8fa\uf8fa\uf8fb ∈ Rn×n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The generalized averaged Gauss formula can also be described in a more compact form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Indeed, ˆA2n+1 can be written as ˆA2n+1(f) = βn+1 βn + βn+1 In(f) + βn βn + βn+1 ¯An+1(f), (24) 8 where the quadrature formula ¯An+1(f) = n+1 � i=1 ¯σ(n+1) i f � ¯τ (n+1) i � (25) is determined by the symmetric tridiagonal matrix ¯Jn+1 ∈ R(n+1)×(n+1), de- fined as ¯Jn+1 = � Jn en � βn + βn+1 eT n � βn + βn+1 αn � , (26) (see [16]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' By construction, the (n + 1)-point quadrature rule (25) has es- sentially the same properties of the anti-Gaussian rule An+1, but in this case the nodes are internal, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', ¯τ (n+1) 1 ∈ [0, +∞), if and only if πn+1(0) πn−1(0) ≥ βn+1, n ≥ 1, (27) (see [18]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' At this point,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' we denote the generalized averaged Gaussian rules corre- sponding to IJ n and IL n by ˆAJ 2n+1(f) = βJ n+1 βJ n + βJ n+1 IJ n(f) + βJ n βJ n + βJ n+1 ¯AJ n+1(f) and ˆAL 2n+1(f) = βL n+1 βL n + βL n+1 IL n (f) + βL n βL n + βL n+1 ¯AL n+1(f),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' where the formulas ¯AJ n+1(f) = n+1 � i=1 ¯w(n+1) i f � ¯x(n+1) i � and ¯AL n+1(f) = 1 cα+1 n+1 � i=1 ¯λ(n+1) i f � ¯ξ(n+1) i � are associated with the tridiagonal matrices ¯JJ n+1 and ¯JL n+1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' obtained by con- sidering in (26) the recursion coefficients αJ k,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' βJ k and αL k ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' βL k ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Finally, the error of the Gaussian quadrature (see (11)) is estimated as En,ν,α,c(f) ≈ ˆEn,ν,α,c(f) := ˆEJ n,ν,α,c(f) − ˆEL n,α,c(f), (28) where ˆEJ n,ν,α,c(f) = ˆAJ 2n+1(f) − IJ n(f) and ˆEL n,α,c(f) = ˆAL 2n+1(f) − IL n (f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 9 From (27) and by using again relations (19) and (20), it is trivial to prove that for the generalized averaged Gauss rule ¯AL n+1 the condition for the internality is α ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' This means that for −1 < α < 1 the smallest node of ¯AL n+1 is negative.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' As before, the behavior of the rule ¯AJ n+1 can only be verified numerically.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, independently of ν, c and working with n = 100, we have found negative values of ¯x(n+1) 1 for −1 < α < ¯α, with ¯α ∈ (1, 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 4 A tentative a priori estimate In the previous sections we have described how, having at disposal the re- currence coefficients of the corresponding orthogonal polynomials, the av- eraged and generalized averaged Gaussian rules can be easily constructed and employed to approximate the quadrature error En,ν,α,c(f) (see (11)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In this section, by using the results of Barrett [3] regarding the derivation of an asymptotic expression for the error of the Gauss-Laguerre formula, we present a tentative a priori estimate of the error En,ν,α,c(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, as remarked in the introduction, since |Jν(x)| ≤ 1, for ν ≥ 0, x ∈ R, we can interpret the weight function wν,α,c as a perturbation of the weight of the Gauss-Laguerre rule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Therefore, the idea is to employ the error estimate for the Gauss-Laguerre formula to approximate not only EL n,α,c(f), but also EJ n,ν,α,c(f) (see (12)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' First, let consider the error of the Laguerre rule for the classical case of c = 1, that is, EL n,α,1(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' For any given R > 1, the set ΓR = � z ∈ C | ℜ√−z = ln R � represents a parabola of the complex plane positively oriented, symmetric with respect to the real axis and with vertex at − ln R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Barrett [3] showed that, if f(z) has no singularities on or within ΓR, except for a pair of simple poles z0 and its conjugate ¯z0, then, for n → ∞, EL n,α,1(f) ∼ 4πℜ � Res(f(z), z0)e−iαπ � exp √−z0 �−2√¯n� , (29) where ¯n = 4n + α + 2 and the symbol Res(·, ·) denotes the residue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Formula (29) follows from the fact that EL n,α.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='1(f) can be represented as the contour integral EL n,α,1(f) = 1 2πi � Γ qL n(z) L(α) n (z) f(z)dz, 10 where L(α) n (z) is the generalized Laguerre polynomial, qL n(z) is the associated function defined by qL n(z) = � +∞ 0 wα(x)L(α) n (x) z − x dx, with wα(x) = xαe−x and Γ is a contour containing [0, +∞) and such that no singularity of f(z) lies on or within the contour (see [19]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Then, by using the relation (see [5]) qL n(z) L(α) n (z) ∼ −2e−iπαzαe−z Kα �√ ¯nze−iπ � Iα �√ ¯nze−iπ � , (30) where Iα, Kα are the modified Bessel functions of order α of the first and second kind, respectively, and the asymptotic formulas (see [1, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='377, 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='1- 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='2]) Iα(t) ∼ et √ 2πt, Kα(t) ∼ e−t � π 2t, valid for large |t|, | arg(t)| < π/2, we obtain qL n(z) L(α) n (z) ∼ −2πe−iαπzαe−z � exp �√ −z ��−2√¯n =: Φn(z), z /∈ [0, +∞).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (31) Finally, formula (29) can be derived by choosing Γ = ΓR ∪C1 ∪C2, where C1, C2 are two arbitrary small circles surrounding the two poles, and by using relation (31).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' After simple computations, it can be verified that, by replacing wα(x) with wα,c(x), one obtains EL n,α,c(f) ∼ 4π cα−1ℜ � Res(f(z), z0)e−iαπ � exp √−cz0 �−2√¯n� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (32) Now, let consider the error EJ n,ν,α,c(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' It can also be written as the contour integral EJ n,ν,α,c(f) = 1 2πi � Γ qJ n(z) pn(z)f(z)dz, where pn(z) is the orthogonal polynomial relative to the weight function wν,α,c and qJ n(z) = � +∞ 0 wν,α,c(x)pn(x) z − x dx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 11 However, as remarked before, we do not have at disposal the analytic expres- sion of pn(z) and, therefore, an asymptotic formula analog to (31) can not be derived.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In order to justify the use of estimate (32) also for EJ n,ν,α,c(f), we numerically evaluate the functions ΨJ n(z) := qJ n(z) pn(z) and ΨL n(z) := qL n(z) L(α) n (z) (33) and check if the approximation Φn(z) (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (31)) can also be used for ΨJ n(z).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In Figures 1-2, for different values of n, ν, α, c, we plot the ratios ΨJ n(z) Φn(z) and ΨL n(z) Φn(z) , in which z = reiθπ, with r = 4, θ ∈ (0, 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We remark that ΨL n(z) Φn(z) ∼ 1, for n → ∞, by (31).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The results show that approximation (31) works rather good also for ΨJ n(z), and the situation is analog for other values of the parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Hence, the idea is to use the approximation (32) also for EJ n,ν,α,c(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Finally, since En,ν,α,c(f) = EJ n,ν,α,c(f) − EL n,α,c(f), we consider the estimate |En,ν,α,c(f)| ≈ 2En,α,c(f), (34) with En,α,c(f) := 4π cα−1 ���Res(f(z), z0) � exp √−cz0 �−2√¯n��� , (35) (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (32)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 5 Numerical examples In this section we present some numerical experiments which confirm the reliability of the error estimates (23), (28) and (34).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We remark that all the computations reported in this work are carried out in Matlab by using high- precision arithmetic, specifically with 120 significant decimal digits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The main reason for this choice is that we do not known explicitly the recurrence coefficients αJ k, βJ k and hence we are forced to employ a numerical scheme to derive them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' However, it is well known (see e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [10]) that this compu- tation can be inaccurate for growing k, because the problem is severely ill conditioned.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Therefore, even if in [7] an alternative more stable approach 12 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='85 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='9 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='95 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='05 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='15 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='25 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='3 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='85 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='9 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='95 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='05 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='15 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='25 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='3 Figure 1: The ratios ΨJ n(z) Φn(z) (left) and ΨL n(z) Φn(z) (right) for z = reiθπ, with r = 4, θ ∈ (0, 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In this case ν = 1, c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5, α = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='4 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='6 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='4 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='6 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 Figure 2: The ratios ΨJ n(z) Φn(z) (left) and ΨL n(z) Φn(z) (right) for z = reiθπ, with r = 4, θ ∈ (0, 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In this case ν = 0, c = 1, α = −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 13 is presented, the use of high-precision arithmetic allows to considerably in- crease the number of quadrature points and to obtain an absolute error in the approximations of the order of the machine precision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The Matlab routine for the computation of the recurrence coefficients αJ k, βJ k is taken from [11], while the code for the implementation of the Gauss-Laguerre quadrature rule from [20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Example 1 Consider the integral Iν,α,c(f) = � ∞ 0 f(x)xαe−cxJν(x)dx, with f(x) = 1 1 + e−x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In order to use error estimate (34), we start by studing the poles of f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' A simple analysis shows that they are given by the set zk = −i(π + 2kπ), k ∈ Z, and the closest to the real axis are z0 and its conjugate z1, that is, ±iπ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' As for the residue (cf, (32)), we obtain Res (f(z), z0) = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Therefore, by using (34) and (35), we have that |En,ν,α,c(f)| ≈ 8πc1−αe− √ 2cπ¯n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (36) In Figure 3 we compare, for different values of ν, α, c, the quadrature error |En,ν,α,c(f)|, obtained by considering a reference solution, with the approx- imations | ˜En,ν,α,c(f)|, | ˆEn,ν,α,c(f)| (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (23)-(28)) and estimate (36).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We can see the very good agreement between the error and the approximations given by the averaged and generalized averaged Gaussian rules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Moreover, the examples reveal that the a priori estimate (36) is rather accurate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Example 2 Consider now the case of f(x) = 1 1 + x2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 14 0 20 40 60 80 10-14 10-12 10-10 10-8 10-6 10-4 10-2 100 error averaged Gauss generalized averaged Gauss a priori estimate 0 20 40 60 80 10-20 10-15 10-10 10-5 100 Figure 3: The quadrature error and its approximations | ˜En,ν,α,c(f)|, | ˆEn,ν,α,c(f)| and (36) for ν = 1, c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5, α = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='7 (left) and ν = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5, c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8, α = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 (right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' This function has only two poles ±i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' As for the residue (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (32)), we obtain Res (f(z), i) = − i 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Therefore, by using (34) and (35), we have that |En,ν,α,c(f)| ≈ 4πc1−αe− √ 2cπ¯n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (37) In Figure 4 we compare, for different values of ν, α, c, the quadrature error |En,ν,α,c(f)|, obtained by considering a reference solution, with the approxi- mations | ˜En,ν,α,c(f)|, | ˆEn,ν,α,c(f)| (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (23)-(28)) and estimate (37).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 6 Conclusions In this work the error estimates of the Gaussian quadrature formula intro- duced in [7] are considered.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In particular, a posteriori error approximations given by the averaged and generalized averaged Gaussian rules have been 15 0 20 40 60 80 10-15 10-10 10-5 100 error averaged Gauss generalized averaged Gauss a priori estimate 0 20 40 60 80 10-12 10-10 10-8 10-6 10-4 10-2 100 Figure 4: The quadrature error and its approximations | ˜En,ν,α,c(f)|, | ˆEn,ν,α,c(f)| and (37) for ν = 1, c = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5, α = 1 (left) and ν = 0, c = 1, α = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 (right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' constructed and showed to be very accurate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Moreover, starting from nu- merical experiments regarding the ratio qJ n/pn (see (33)), an heuristic but quite effective a priori error estimate has been introduced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' We remark that, having at disposal an a-priori estimate of the error and by noting that, simi- lar to the Gauss-Laguerre rule, the weights w(n) i , i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , n, (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (4)) decay exponentially (see Figure 5), a truncated approach can also be introduced, to reduce the number of function evaluation (see [2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Acknowledgements This work was partially supported by GNCS-INdAM and CINECA under HPC-TRES program award number 2019-04.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' The author is member of the INdAM research group GNCS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 16 0 5 10 15 20 25 30 10-50 10-40 10-30 10-20 10-10 100 0 5 10 15 20 25 30 10-50 10-40 10-30 10-20 10-10 100 1010 Figure 5: The weights w(n) i , ξ(n) i , i = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' , n (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (4)-(6)) for ν = 1, α = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='7, c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5 (left) and ν = 0, α = −0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='5, c = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='8 (right).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' In both cases n = 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' References [1] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Abramowitz and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Stegun (1970), Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, 7th Edi- tion, Dover Publications, Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', New York.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [2] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Aceto and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Novati, Fast and accurate approximations to fractional powers of operators, IMA Journal of Numerical Analysis 42 (2022), 1598–1622.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [3] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Barrett, Convergence properties of Gaussian quadrature formulae, Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 3 (1960/1961), 272–277.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [4] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Davis and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Rabinowitz (1975), Methods of numerical integration, Academic Press, Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', New York.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [5] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Donaldson and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Elliott, A Unified Approach to Quadrature Rules with Asymptotic Estimates of Their Remainders, SIAM Journal on Numerical Analysis 9 (1972), 573-602.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 17 [6] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Ehrich, On stratified extensions of Gauss-Laguerre and Gauss- Hermite quadrature formulas, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 140 (2002), 291- 299.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [7] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Denich and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Novati, Gaussian rule for integrals involving Bessel functions, arXiv:2110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='05051v1 (2021), https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='48550/arXiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='2110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='05051.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [8] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Denich, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Novati and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Picotti, A fast and accurate numerical ap- proach for electromagnetic inversion, Journal of Computational Physics 475 (2023), 111846.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [9] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Gautschi, Gauss-Kronrod quadrature—A survey, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Milovanovi´c (Ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' ), Numerical Methods and Approximation Theory III, Faculty of Electronic Engineering, Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Niˇs (1988), 39-66 [10] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Gautschi, Orthogonal polynomials: applications and computation, Acta Numerica (1996), Cambridge University Press, 45-119.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [11] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Gautschi, Orthogonal Polynomials, Quadrature, and Approximation: Computational Methods and Software (in Matlab), In: Marcell´an F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', Van Assche W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (eds) Orthogonal Polynomials and Special Functions, Lecture Notes in Mathematics, vol 1883, Springer, Berlin, Heidelberg, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content='1007/978 − 3 − 540 − 36716 − 1 1 [12] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Laurie, Anti-Gaussian quadrature formulas, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Comp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 65 (1996), 739-747.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [13] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Kahaner and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Monegato, Nonexistence of extended Gauss- Laguerre and Gauss-Hermite quadrature rules with positive weights, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Angew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 29 (1978), 983-986.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [14] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Kronrod, Integration with control of accuracy (Russian), Dokl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Nauk SSSR 154 (1964), 283-286.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [15] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Notaris, Gauss-Kronrod quadrature formulae.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' A survey of fifty years of research, ETNA, Electron.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Trans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Numer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Anal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 45 (2016), 371- 404.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [16] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Reichel and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Spalevi´c, A new representation of generalized averaged Gauss quadrature rules, Applied Numerical Mathematics 165 (2021), 614-619.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 18 [17] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Reichel and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Spalevi´c, Averaged Gauss quadrature formu- las: Properties and applications, Journal of Computational and Applied Mathematics 410 (2022), 114232.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [18] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Spalevi´c, On generalized averaged Gaussian formulas, Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Comp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 76 (2007), 1483-1492.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [19] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Szego (1939), Orthogonal polynomials, American Mathematical So- ciety Colloquium Publications, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' XXIII.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [20] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' Trefethen, Approximation Theory and Approximation Practice, Ex- tended Edition, SIAM, Philadelphia, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' [21] Watson, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (1966), A treatise on the theory of Bessel functions, Cam- bridge University press, Marcell´an F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=', Van Assche W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' (eds).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} +page_content=' 19' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/sNFKT4oBgHgl3EQfJi1O/content/2301.11738v1.pdf'} diff --git a/t9E1T4oBgHgl3EQfQgOR/content/2301.03041v1.pdf b/t9E1T4oBgHgl3EQfQgOR/content/2301.03041v1.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a8b7fd08475be5de9ebf7b33728470cf1a986082 --- /dev/null +++ b/t9E1T4oBgHgl3EQfQgOR/content/2301.03041v1.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c4c8234287cb20e6db9c1d0b3bb785b1bf850228208a5f7ea9915682adfa572 +size 4744980 diff --git a/t9E1T4oBgHgl3EQfQgOR/vector_store/index.faiss b/t9E1T4oBgHgl3EQfQgOR/vector_store/index.faiss new file mode 100644 index 0000000000000000000000000000000000000000..59650369dd47083fc4aefaae9fb26ff1f224cb0d --- /dev/null +++ b/t9E1T4oBgHgl3EQfQgOR/vector_store/index.faiss @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:502b43f76ab8ad57293119018d63bd59df853372f900865190276cc047f43a01 +size 6094893 diff --git a/t9E1T4oBgHgl3EQfQgOR/vector_store/index.pkl b/t9E1T4oBgHgl3EQfQgOR/vector_store/index.pkl new file mode 100644 index 0000000000000000000000000000000000000000..5d34e8e23ce693e8025b5ef477fdce9a11c2fd90 --- /dev/null +++ b/t9E1T4oBgHgl3EQfQgOR/vector_store/index.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee5d29493b1d6eda328eefd39255e58056e54932f4d2f8ea3907733ef5bf6518 +size 208112 diff --git a/vdFQT4oBgHgl3EQfuzZ_/content/tmp_files/2301.13395v1.pdf.txt b/vdFQT4oBgHgl3EQfuzZ_/content/tmp_files/2301.13395v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..722f5b8a7d25a7af74c7c93dcc2f9e7a04e815d6 --- /dev/null +++ b/vdFQT4oBgHgl3EQfuzZ_/content/tmp_files/2301.13395v1.pdf.txt @@ -0,0 +1,970 @@ +Faster Predict-and-Optimize +with Three-Operator Splitting +Daniel McKenzie∗ +Samy Wu Fung∗† +Howard Heaton‡ +Abstract +In many practical settings, a combinatorial problem must be repeatedly solved with +similar, but distinct parameters w. Yet, w is not directly observed; only contextual +data d that correlates with w is available. It is tempting to use a neural network +to predict w given d, but training such a model requires reconciling the discrete +nature of combinatorial optimization with the gradient-based frameworks used +to train neural networks. One approach to overcoming this issue is to consider a +continuous relaxation of the combinatorial problem. +While existing such approaches have shown to be highly effective on small prob- +lems (10–100 variables) they do not scale well to large problems. In this work, +we show how recent results in operator splitting can be used to design such a +system which is easy to train and scales effortlessly to problems with thousands of +variables. +1 +Introduction +Many high-stakes decision problems in healthcare[ZT21], logistics and scheduling [SE10, KKM21], +and transportation [WT21] can be phrased as combinatorial optimization problems, i.e. to compute +x⋆ = argmin +x∈X +f(x; w), +(1) +where X ⊂ Rn is a finite constraint set and f(x) = w⊤x is a linear function4. Depending on the +structure of X, the problem (1) may be straightforward (e.g. shortest path) or NP-hard (e.g. Traveling +Sales Person problem [Kar72]). In both cases, this problem is well-studied and there are a plethora of +robust and reliable algorithms for solving (1) given both the constraint X and objective f. Our present +interest is settings where f is only partially known. Specifically, we aim to repeatedly compute +x⋆(d) = argmin +x∈X +w(d)⊤x, +(2) +for different data d, where w is a function of d. Importantly, the precise relationship between d and w +is unknown. We propose learning a model to approximate the unknown objective: wΘ(d) ≈ w(d). +The data d is observed and is called the context. +As an illustrative example to keep in mind, consider the shortest path prediction problem studied +in [PPM+19, BBT+20] (see Figure 1). Here, the goal is to find the shortest path (from top-left to +bottom-right) through a randomly generated terrain map from the Warcraft II tileset [Guy]. Each +8-by-8 square in the map represents a vertex in a 12-by-12 grid graph (with diagonal edges). The time +required to traverse this vertex depends on the visual content of the square (e.g. a square containing +water takes longer to traverse than a square containing land). Here d is the image representing the map, +∗Dept. of Applied Mathematics and Statistics, Colorado School of Mines +†Dept. of Computer Science, Colorado School of Mines +‡Typal Research, Typal LLC +4While we focus on linear objective functions, our techniques can be applied to more general, convex f. +Preprint. Under review. +arXiv:2301.13395v1 [cs.LG] 31 Jan 2023 + +0 +20 +40 +60 +80 +0 +20 +40 +60 +80 +(a) Map +0 +2 +4 +6 +8 +10 +0 +2 +4 +6 +8 +10 +(b) Costs for Each Square +0 +2 +4 +6 +8 +10 +0 +2 +4 +6 +8 +10 +(c) Shortest Path +Figure 1: The shortest path prediction problem [PPM+19]. The contextual data d in (a) maps to the +cost of traversing each square in (b), with darker shading representing lower cost. The true shortest +path is shown in (c). +w(d) is a vector of traversal times (i.e. costs) for each vertex5, the true objective is f(x) = w(d)⊤x, +and x⋆(d) is the true shortest path (See (c) in 1). +Suppose we wish to use gradient-based methods (e.g. Stochastic Gradient Descent) to train wΘ(d). +The primary obstacle to doing so is “differentiating through” the solution to +ˆx(d) = argmin +x∈X +wΘ(d)⊤x, +(3) +so as to compute an informative gradient with which to update Θ. Given the combinatorial nature of +X, this is a nontrivial task. Specifically, for many small perturbations of Θ, the solution ˆx(d) will +be unchanged; yet, for some perturbations the solution ˆx(d) will “jump” to a different point in X. +Hence the gradient dˆx +� +dwΘ is always either zero or undefined. We emphasize that this behaviour is +common to all problems of the form (1) with linear f(·; w) . +To compute an informative gradient, we follow recent works (e.g. [WDT19]) and relax (3) to a +quadratic program over the convex hull of X (see (13)) by adding a small regularizer. Building upon +[HWFGY21], we use Davis-Yin Three Operator splitting [DY17] and Jacobian-Free Backpropagation +(JFB) [FHL+22] to build a solver for this relaxation that is efficient and easy to backpropagate through. +Our approach is fast, easy to implement using our provided code6, and unlike several prior works (e.g. +see [PPM+19, BBT+20]), runs completely on GPU. Although prior works are mostly constrained +to problems with fewer than one hundred variables, numerical examples herein demonstrate our +approach, run using only standard computing resources, easily scales to combinatorial problems with +tens of thousands of variables. +Paper Organization +The rest of this paper is laid out as follows. We introduce the Predict-then- +Optimize paradigm in Section 2, which is followed by discussion of related works in Section 3. Our +proposed combination of a three-operator splitting architecture, called DYS-Net, and Jacobian-Free +Backpropagation (JFB) is presented in Section 4. Numerical examples are provided in Section 5 and +concluding remarks in Section 6. +2 +The Predict-and-Optimize Paradigm +2.1 +LP reformulation +We begin by observing that the combinatorial optimization problem (2) is equivalent to a linear +program: +x⋆(d) = argmin +x∈X +w(d)⊤x = +argmin +x∈Conv(X) +w(d)⊤x, +(4) +5Note [PPM+19] considers the vertex-weighted variant of the shortest problem. In our experiments we +consider the edge-weighted version. We discuss this further in Appendix. +6See https://github.com/mines-opt-ml/SPO_with_DYS +2 + +where Conv(X) is the convex hull of X. This is because for generic7 w(d) the solution to the linear +program lies at a corner of the polytope Conv(X), which must be an element of X. Equivalently: +ˆx(d) = argmin +x∈X +wΘ(d)⊤x = +argmin +x∈Conv(X) +wΘ(d)⊤x +(5) +Henceforth we follow [WDT19, MG20] and others and focus exclusively on this LP reformulation. +2.2 +The Loss and Training Data +Recall that we wish to train a neural network wΘ(d) such that ˆx(d) ≈ x⋆(d), where ˆx(d) is defined +in (5) and x⋆(d) is defined in (4). What loss function, and what training data, should one use? Several +works [WXQ+06, FNP+16] suggest gathering training data of the form (d, w(d)) and then training +so as to minimize the discrepancy8 between w(d) and wΘ(d) . Frequently, this is referred to as the +“two-stage” approach. For large or complex problems this can be sub-optimal. Intuitively [Ben97] +this is because small errors in the approximation wΘ(d) ≈ w(d) in areas crucial to the optimization +problem (4) can yield wildly different minimizers. Instead, one should choose a loss aligned with the +end goal, i.e. one that seeks to minimize the discrepancy between x⋆ +Θ(d) and x(d) [WDT19]. We +discuss two such losses here. +Regret Loss +To define this loss, also known as the “Smart Predict-then Optimize” (SPO) loss +[EG22] or task loss [MG20], we first consider the regret associated to solving (2), given a particular +d, with ˆx(d) instead of x⋆(d): +R(Θ, d, w) := w(d)⊤ˆxΘ(d) − w(d)⊤x⋆(d) +(6) +As the second term is independent of model parameters (Θ) we may drop it and consider the loss: +LR(Θ) = Ed∼D +� +w(d)⊤ˆxΘ(d) +� +(7) +The required training data in this setting is context-weight pairs (d, w(d)) where d is drawn from the +distribution in question D. Note that finding a model with low task loss guarantees that cost of the +model output (i.e. w⊤ˆx(d)) is close to the true optimal cost (i.e. w⊤x⋆(d)). However, it provides no +guarantees as to how well ˆx(d) approximates x⋆(d). +Argmin loss +If w(d) is never observed directly, only x⋆(d), one can minimize the mismatch +between ˆx(d) and x⋆(d): +LA(Θ) = 1 +2Ed∼D +� +∥x⋆(d) − ˆxΘ(d)∥2� +(8) +A similar loss is used in [PPM+19], except that the Hamming distance between x⋆(d) and ˆx(d), +not the ℓ2 distance, is computed. The required training data for this loss is context-solution pairs +(d, x⋆(d)). +2.3 +Argmin differentiation +In practice we replace the true risks for either loss by empirical ones: +LR(Θ) = +m +� +i=1 +w⊤ +i ˆxΘ(di) +(9) +LA(Θ) = 1 +2 +m +� +i=1 +∥x⋆(di) − ˆxΘ(di)∥2 +(10) +7For certain w(d) it may happen that an entire face of Conv(X) is a solution to the linear program. In this +case, (??) is no longer strictly correct; rather the left-hand side is a subset of the right-hand side. In the interest +of clarity we ignore this edge case. +8for example the least-square discrepancy ∥w(d) − wΘ(d)∥ +3 + +where wi are the observed/true weights corresponding to di. In order to use gradient based methods, +one has to compute the derivative with respect to Θ. Appealing to the chain rule: +dLR +dΘ = +m +� +i=1 +w⊤ +i +∂ˆxi,Θ +dwi,Θ +dwi,Θ +dΘ +(11) +dLA +dΘ = +m +� +i=1 +(x⋆(di) − ˆxΘ(di))⊤ ∂ˆxi,Θ +dwi,Θ +dwi,Θ +dΘ , +(12) +where wi,Θ := wΘ(di) is the neural network approximation evaluated at di. For both LR and +LA, the key difficulty is making sense of dˆxΘ/dw. As discussed in Section 1, ˆxΘ(d) is piecewise +constant as a function of w, and this remains true for the LP reformulation (5). So, for all w either +dx⋆ +Θ(d)/dw = 0 or dx⋆ +Θ(d)/dw is undefined—both cases do not yield an informative gradient. To +remedy this, [WDT19, MG20] propose adding a small amount of regularization to the objective +function in (5) so that the objective function becomes strongly convex, whence ˆxΘ(d) becomes a +continuously differentiable function of w. We shall follow [WDT19] and add a small quadratic +regularizer, so that (5) becomes: +ˆxΘ(d) = +argmin +x∈Conv(X) +wΘ(d)⊤x + γ∥x∥2 +2 +� +�� +� +≜fΘ(x;d) +. +(13) +Thus, for the rest of this paper we shall consider the problem +ˆxΘ(d) = argmin +x∈C +fΘ(x; d) +(14) +where C is a polytope (i.e. the convex hull of a finite set) and fΘ(x; d) is strongly convex with +respect to x. Our goal is to solve (14) and simultaneously computing the derivative9 dˆxΘ/dΘ; this +problem is frequently referred to as argmin differentiation as has received much attention lately +[AK17, AAB+19b, ABB+19, BKK19, FHL+22]. Crucially, we note that every polytope may be +expressed in the standardized form10: +C = {x ∈ Rn : Ax = b and x ≥ 0} , +(15) +where the inequality x ≥ 0 is interpreted componentwise, see [Zie12]. +3 +Prior Work +The most common approach to computing dˆxΘ/dΘ, proposed in [AK17] and used in [WDT19, +MG20, FWDT20], starts with the KKT conditions for constrained optimality: +∂fΘ +∂x (ˆxΘ) + A⊤ˆλ + ˆν = 0 +Aˆx − b = 0 +D(ˆν)ˆxΘ = 0 +where ˆλ and ˆν ≥ 0 are Lagrange multipliers associated to the optimal solution ˆxΘ [Ber97] and +D(ˆν) is a matrix with ˆν along its diagonal. Differentiating these equations with respect to Θ and +rearranging one obtains: +� +� +∂2fΘ +∂x2 +A +I +A⊤ +0 +0 +D(ˆν) +0 +D(ˆxΘ) +� +� +� +� +dˆxΘ +dΘ +dˆλ +dΘ +dˆν +dΘ +� +� = +� +� +∂2fΘ +∂x∂θ +0 +0 +� +� . +(16) +Solving (16) then yields dˆxΘ +dΘ (as well as dˆλ +dΘ and dˆν +dΘ). The computational bottleneck in this approach +is computing ˆλ and ˆν in addition to ˆxΘ for (14). If dim(x) = n and A ∈ Rm×n this can be done with +a primal-dual interior point method at a cost of O +� +max{n, m}3� +[AK17]. In principle it is possible +9For notational convenience we suppress the dependence on d +10The standardization process may add dummy variables, and so increase the dimension of the problem n +4 + +to exploit sparsity in A or ∂fΘ +∂x (ˆxΘ) to solve (16) faster, but in practice we observe the state-of-the-art +implementation of this approach, cvxpylayers [AAB+19b], struggles with problems containing +more than 100 variables, see Section 5. +Another approach, proposed for deep equilibrium models in [BKK19] and adapted to constrained +optimization layers in [CDB+21, BBC+21] is to re-formulate (14) as a fixed point problem: +Find ˆxΘ such that ˆxΘ = PC +� +ˆxΘ − α∂fΘ +∂x (ˆxΘ; d) +� +(17) +and then apply the implicit function theorem to obtain an explicit formula for dˆxΘ/dΘ. However, the +cost of computing PC can be prohibitive, see Section 4.1. +Finally, many works use a perturbation-based approach to define a continuously differentiable proxy +for the solution to the unregularized optimization problem (5), which we rewrite here as +g(w) = +min +x∈Conv(X) w⊤x, +(18) +omitting the dependence of w on d for notational clarity. For example, [PPM+19] define a piecewise- +affine interpolant to g(w). The gradients of gλ(w) are strikingly easy to compute, requiring just one +additional solve of (18) with perturbed cost w′. In [BBT+20] a stochastic perturbation is considered: +gε(w) = EZ +� +min +x∈Conv(X) (w + εZ)⊤ x +� +, +(19) +analogous to Nesterov-Spokoiny smoothing [NS17] in zeroth-order optimization. By Danskin’s +theorem [Dan66] the gradients of gε(w) are also easy to compute: +∇wgε(w) = EZ +� +argmin +x∈Conv(X) +(w + εZ)⊤ x +� +≈ 1 +m +m +� +i=1 +argmin +x∈Conv(X) +(w + εZi)⊤ x. +(20) +We implement this approach as PertOpt-net in Section 5. We highlight that the advantage of such +approaches is they easily wrap around existing combinatorial solvers (e.g. Dijkstra for the shortest +path problem), as only repeated solves of (18) are required for computing gradients. The disadvantage +is that such solvers are usually run on CPU. Thus, data needs to be shuttled between CPU and GPU +when training. +4 +DYS-Net +We now describe our proposed architecture, DYS-Net. We use this term to refer to the neural +architecture (see Section 4.1) and the custom backprop procedure (see Section 4.2). DYS-Net models +may be trained using either the regret loss or the argmin loss. +4.1 +The Forward Pass +We wish to compute ˆxΘ and dˆxΘ/dΘ where +ˆxΘ = argmin +x∈C⊂Rn fΘ(x; d). +(21) +As we are interested in the case where n is very large, we are led to consider first-order methods +(e.g. gradient descent) over second-order methods (e.g. Newton’s method). For example, one could +consider Projected Gradient Descent (PGD): +xk+1 = PC +� +xk − α∂fΘ +∂x (xk; d) +� +for k = 0, . . . , K − 1 +ˆxΘ(d) ≈ xK +(22) +This works for simple C for which there exists an explicit form of PC, e.g. when C is the probability +simplex [DSSSC08, Con16, LMY21]. However, for general polytopes C no such form exists, and so +one must resort to a second iterative procedure, run at every k, to compute PC(xk) as the solution to +PC(xk) = argmin +y∈C +∥y − xk∥2. +(23) +5 + +Our core idea is to solve (21) using Davis-Yin splitting (DYS) [DY17] to avoid computation of PC in +the forward pass. To this end, we recall the standardized representation of polytopes (15) and write: +C = {x : Ax = b and x ≥ 0} = {x : Ax = b} +� +�� +� +≜C1 +∩ {x : x ≥ 0} +� +�� +� +≜C2 += C1 ∩ C2, +(24) +where we note PC is hard to compute but PC1 and PC2 can be computed cheaply in many settings.11 +The following theorem formulates (21) as a fixed point problem involving only PC1 and PC2, not PC. +Theorem 1. Suppose C = C1 ∩ C2 for convex C1 and C2. Suppose both Ci are polyhedral or have +relative interiors with a point in common and fΘ(x; d) be strongly convex and L-smooth in x. For +any α > 0 define: +TΘ(x) ≜ x−PC1(x) + PC2 (2PC1(x)−x−α∇fΘ(PC1(x))) , +(25) +Then ˆxΘ solves (21) if and only if +ˆxΘ = PC1(ˆzΘ) where ˆzΘ = TΘ(ˆzΘ). +(26) +Proof. This follows from [HML+21, Theorem 3.2] by first noting ˆxΘ solves (21) if and only if it is +the solution to the variational inequality +∇xfΘ(ˆxΘ; d)⊤ (x − ˆxΘ) ≥ 0 +for all x ∈ C. +(27) +As the operator ∇xfΘ(·; d) is L-Lipschitz continuous, it is 1/L-cocoercive by the Baillon-Haddad +theorem [BH77, BC09]. +The next result shows that the simple fixed point iteration method, applied with TΘ, will converge. +Corollary 1. Suppose fΘ is L-smooth. Take α ∈ (0, 2/L), and define the sequence +zk+1 = TΘ(zk). +(28) +Then xk := PC1(zk) → ˆxΘ with rate O(1/k). +Proof. See [RY22, Sec 2.2.1] +The upshot is that we can compute ˆxΘ using only PC1 and PC2. Finally we note that if fΘ(·; d) +is strongly convex with respect to x then x⋆ +Θ(d) is unique for fixed d and Θ, and the mapping +Θ �→ x⋆ +Θ(d) is differentiable [Daf88, FP03, HML+21]. +4.2 +Efficient Backpropagation +Differentiating both sides of (26), we obtain +dˆzΘ +dΘ = ∂TΘ +∂Θ + ∂TΘ +∂z +dˆzΘ +dΘ +(29) +⇒ dˆzΘ +dΘ = J −1 +Θ +∂TΘ +∂Θ +where +JΘ(z) = I − ∂TΘ +∂z +(30) +Thus, one may compute the gradient of the loss using the chain rule as follows: +dℓ +dΘ = dℓ +dx +dˆxΘ +dΘ = dℓ +dx +�dPC1 +dz +dˆzΘ +dΘ +� += dℓ +dx +dPC1 +dz J −1 +Θ +∂TΘ +∂Θ +(31) +This approach requires solving a linear system with JΘ which becomes particularly expensive when +dim(ˆxΘ) is large. Instead, we use the recently introduced Jacobian-Free Backpropagation (JFB) in +which the Jacobian JΘ is replaced with the identity matrix. This leads to an approximation of the +true gradient dℓ +dΘ using +pΘ = +� ∂ℓ +∂x +dPC1 +dz +∂TΘ +∂Θ +� +Θ,x=x⋆ +Θ +. +(32) +In [FHL+22] it is shown pΘ is a descent direction, i.e. +� +pΘ, dℓ +dΘ +� +> 0. Moreover, JFB has been +found to be effective in a wide variety of application domains imaging [HML+21, HWFGY21, HF22, +RMB+22]. Using JFB yields a massive computational saving, as we no longer need to solve a linear +system with JΘ. +11Further splitting of C1 can yield simpler projections if a singular value decomposition of A is unavailable. +6 + +Algorithm 1 DYS-Net Trained with JFB +1: Input: A and b defining C, fΘ +2: Initialize Θ0 at random +3: Precompute SVD for PC1 +4: for m = 0, . . . , M − 1 do +5: +Compute xK = PC1(zK) ≈ ˆxΘm using iteration (28) with TΘm. +6: +Compute pΘ(xK) ≈ pΘ(ˆxΘ) using (32). +7: +Θm+1 = Θm − ηpΘm(xK). +8: end for +4.3 +Implementation +The combination of DYS for the forward pass and JFB for backpropagation makes the algorithm +simple to implement. First, we note that PC1 is a projection onto the intersection of hyperplanes +which is given by the following lemma [HML+21]: +Lemma 1. Let C1 = {x : Ax = b} where A ∈ Rm×n and b ∈ Rm. Then the projection onto C1 is +given by +PC1(z) = argmin +x∈C1 +1 +2∥x − z∥2 = z − A†(Az − b) +(33) +where A† = UΣ−1V ⊤ and UΣV ⊤ is the compact singular value decomposition of A such that U +and V have orthonormal columns and Σ is invertible. +Thus, to efficiently apply PC1, a singular value decomposition of A is first computed off-line. Also, +once a parameterization is chosen for fΘ, standard autodifferentiation tools, e.g. PyTorch [PGM+19], +can be used to compute its gradient with respect to its inputs. Moreover, backpropagation can also +be simply implemented as follows. We first compute ˆxΘ without tracking its gradients. Next, since +JFB is equivalent to backpropagating through the last layer of the fixed point iteration [FHL+22], +we apply TΘ from (25) once more to ˆxΘ to track the gradients and backpropagate through the last +application of TΘ. Thus, given ˆxΘ, we backpropagate in four lines of code as shown in Fig. 2. +Backpropagation through x⋆ +Θ +y = apply_T(x_star) +loss = criterion(y) +loss.backward() +optimizer.step() +Figure 2: Sample PyTorch code for backpropagation through DYS. +The entire training process can thus be described in Algorithm 1. +5 +Numerical Experiments +We evaluate the performance of DYS-Net on a shortest path prediction problem inspired by [PPM+19] +(see the discussion of Section1). +5.1 +Problem Setup +In [PPM+19] a shortest path prediction problem is introduced for k-by-k grid graphs. The graph is +constructed by partitioning a 8k-by-8k pixel image representing a randomly generated terrain from +the computer game Warcraft II into 8-by-8 squares. Each square represents a vertex12, and the time +12Note [PPM+19] considers the vertex-weighted variant of the shortest problem. In our experiments we +consider the edge-weighted version. We discuss this further in Appendix. +7 + +10-by-10 +20-by-20 +30-by-30 +Test MSE Loss +0 +25 +50 +75 +100 +epochs +10 1 +DYS +CVX +PertOpt +0 +25 +50 +75 +100 +epochs +10 2 +10 1 +DYS +CVX +PertOpt +0 +25 +50 +75 +100 +epochs +10 2 +10 1 +DYS +CVX +PertOpt +Training Time +0 +25 +50 +75 +100 +epochs +0 +20 +40 +60 +80 +DYS +CVX +PertOpt +0 +25 +50 +75 +100 +epochs +0 +100 +200 +300 +DYS +CVX +PertOpt +0 +25 +50 +75 +100 +epochs +0 +200 +400 +600 +800 +DYS +CVX +PertOpt +Figure 3: Comparison of of DYS-Net, cvxpylayers , and PertOptNet [BBT+20] for three different +grid sizes: 10 × 10 (first column), 20 × 20 (second column), and 30 × 30 (third column). The first +row shows the MSE loss vs. epochs of the testing dataset. The second row shows the training time vs. +epochs. +grid size +number of variables +network size +5-by-5 +40 +500 +10-by-10 +180 +2040 +20-by-20 +760 +8420 +30-by-30 +1740 +19200 +50-by-50 +4900 +53960 +100-by-100 +19800 +217860 +Table 1: Number of variables (i.e. number of edges) per grid size for the shortest path problem +described in Section 5. Third column: number of parameters for all three models used: DYS-Net, +cvxpyayers and PertOpt-Net +taken to traverse this vertex depends on the visual content of the square (e.g. a square containing +water takes longer to traverse than a square containing land). The context d ∈ R8k×8k is then the +visual information encoded by all these squares, i.e. an input image. In [PPM+19] problem instances +are considered for k ∈ {12, 18, 24, 30}. +Data Generation +We extend this experiment to larger graphs while keeping the intrinsic learning +complexity of the task fixed. We do so by constructing a series of k × k grid graphs (see Fig 4) +for k ∈ {5, 10, 20, 30, 50, 100} where the weights are an unknown function of a context parameter +d ∈ R5. We do not allow diagonal edges thus each vertex (excepting those at corners and on sides) +has four edges. In all experiments we draw d uniformly at random from the unit cube [0, 1]5. We +assume the true edge weights are given by w(d) = Wd where W ∈ R|E|×5, and E is the set of edges +in the graph. While more complicated relationships between w and d can certainly be considered13, +we deliberately choose a simple relationship between context and edge weights so that the forward +and backward pass through the combinatorial solver is the bottleneck in training. Further details are +presented in Appendix A. +Models +We test three different differential combinatorial solvers, based on cvxpylayers +[AAB+19a], the perturbed optimization approach of Berthet et al [BBT+20], as well as the proposed +13and are easily implementable using our code: https://github.com/mines-opt-ml/SPO_with_DYS +8 + +True Path +DYS-net +CVX-net +PertOpt-net +10-by-10 +20-by-20 +30-by-30 +Figure 4: True paths (column 1), paths predicted by DYS-net (column 2), CVX-net (column 3), and +PertOpt-net (column 4). Samples are taken from different grid sizes: 10-by-10 (row 1), 20-by-20 +(row 2), and 30-by-30 (row 3). +DYS-Net. We use the exact same neural network architecture for wΘ(d) in all three approaches; a two +layer fully connected neural network with leaky ReLU activation functions. Network sizes can be seen +in Table 1. We refer to the network equipped with the various solvers as CVX-net, PertOpt-net, +and DYS-net respectively. CVX-net and DYS-net use the regularized, linear program form (13) of +the shortest path problem. We tuned the hyperparameters for each architecture to the best of our +ability on the smallest problem (5-by-5 grid graphs) and then used these hyperparameter values for +all other graph sizes. We train all three architectures for 100 epochs total on each problem. +5.2 +Results +In Figure 3, we show the test loss and training time per epoch for all three architectures: DYS-net, +CVX-net, and PertOpt-net for 10-by-10, 20-by-20, and 30-by-30 grids. In terms of MSE loss, +CVX-net and DYS-net lead to comparable performance. In the second row of Figure 3, we observe +the benefits of combining the three-operator splitting with JFB [FHL+22]; in particular, DYS-Net +trains much faster. Figure 4 shows some randomly selected outputs for the three architectures once +fully trained. Interpreting the outputs of DYS-net and CVX-net as (unnormalized) probabilities +over the grid, one can use a greedy decoder to determine the most probable path from top-left to +bottom-right. For small grids, e.g. 5-by-5, this most probable path coincides exactly with the true +path for most d (see Fig. 6). For larger grids, we find there are often slight differences between the +predicted and true paths. This is not surprising, as the number of possible paths grows exponentially +with k. Thus, the number of “almost shortest paths” grows too, making exact prediction of x⋆ +d an +unreasonable task. Hence, we use the regret (see (6)) of the trained model, averaged over the test set, +9 + +Wa) Test MSE Loss +b) Training Time (in minutes) +c) Regret Values +25 +50 +75 +100 +grid size +10 2 +10 1 +DYS +CVX +PertOpt +25 +50 +75 +100 +grid size +103 +104 +105 +106 +DYS +CVX +PertOpt +25 +50 +75 +100 +grid size +10 3 +10 2 +10 1 +100 +101 +DYS +CVX +PertOpt +Figure 5: a) Test MSE loss (left), b) training time in minutes (middle), and c) regret values (right) vs. +gridsize for three methods: DYS-Net, cvxpylayers[AAB+19b], and PertOpt [BBT+20]. The grid +sizes are chosen to be 5-by-5, 10-by-10, 20-by-20, 30-by-30, 50-by-50, and 100-by-100. All three +algorithms are shown up to gridsize 30-by-30, however, CVX is unable to load or run problems with +gridsize over 30. Indeed, this is because the optimization variable x is too large. Dimensions of the +variables can be found in Table 1. PertOpt [BBT+20] can be trained on the larger problems but takes +a substantial amount to train. +as a measure of accuracy. Fig. 5 shows that while CVX-net and PertOpt-net achieve low regret +for small grids, DYS-net model achieves a low regret for all grids. In addition to training faster, +DYS-net can also be trained for much larger problems, e.g., 100-by-100 grids, as shown in Figure 5. +We found that CVX-net could not handle grids larger than 30-by-30, i.e. , problems with more than +1740 variables14 (see Table 1). Importantly, PertOpt-net takes close to a week to train for the +100-by-100 problem, whereas DYS-net takes about a day (see right Figure 5b). For details on the +training setup, see Appendix A.2. +6 +Conclusions +Predicted Path Accuracy % +0 +25 +50 +75 +100 +epochs +0 +25 +50 +75 +100 +DYS +CVX +PertOpt +Figure 6: Accuracy (in percentage) of predicted +paths on 5-by-5 grid during training. +We present a new data-driven method for solv- +ing combinatorial optimization problems where +parameters are unknown. The core idea is to +use an implicit learn-to-optimize approach and +learn an optimization algorithm based on three- +operator-splitting. We call this model DYS-net. +Importantly, DYS-net is trained using Jacobian- +Free Backpropagation (JFB), thus avoiding com- +puationally expensive backpropagation through +projections on to the feasible set. +Our experiments shows DYS-net leads to better +accuracy with substantially lower training times. +Our approach is also easy to implement as a +consequence of JFB. While this work focused +on linear objective functions, DYS-Net can be +applied to more general objective functions with +uniformly-Lipschitz gradients [RY21]. Moreover, as the dimensions of problems increase, this +problem becomes more akin to using deep learning for optimal control problem [ONL+21, ONL+22], +where the aim is to find an optimal path that minimizes an energy functional. Investigating these +connections is a direction of future work. +References +[AAB+19a] A. Agrawal, B. Amos, S. Barratt, S. Boyd, S. Diamond, and Z. Kolter. Differentiable +convex optimization layers. In Advances in Neural Information Processing Systems, +2019. +14This is to be expected, as discussed in in [AK17, AAB+19a] +10 + +[AAB+19b] Akshay Agrawal, Brandon Amos, Shane Barratt, Stephen Boyd, Steven Diamond, +and J Zico Kolter. Differentiable convex optimization layers. Advances in neural +information processing systems, 32, 2019. +[ABB+19] Akshay Agrawal, Shane Barratt, Stephen Boyd, Enzo Busseti, and Walaa M Moursi. +Differentiating through a cone program. arXiv preprint arXiv:1904.09043, 2019. +[AK17] Brandon Amos and J Zico Kolter. Optnet: Differentiable optimization as a layer in +neural networks. In International Conference on Machine Learning, pages 136–145. +PMLR, 2017. +[BBC+21] Mathieu Blondel, Quentin Berthet, Marco Cuturi, Roy Frostig, Stephan Hoyer, Felipe +Llinares-López, Fabian Pedregosa, and Jean-Philippe Vert. Efficient and modular +implicit differentiation. arXiv preprint arXiv:2105.15183, 2021. +[BBT+20] Quentin Berthet, Mathieu Blondel, Olivier Teboul, Marco Cuturi, Jean-Philippe Vert, +and Francis Bach. Learning with differentiable pertubed optimizers. Advances in +neural information processing systems, 33:9508–9519, 2020. +[BC09] Heinz H Bauschke and Patrick L Combettes. The baillon-haddad theorem revisited. +arXiv preprint arXiv:0906.0807, 2009. +[Ben97] Yoshua Bengio. Using a financial training criterion rather than a prediction criterion. +International journal of neural systems, 8(04):433–443, 1997. +[Ber97] Dimitri P Bertsekas. Nonlinear programming. Journal of the Operational Research +Society, 48(3):334–334, 1997. +[BH77] Jean-Bernard Baillon and Georges Haddad. Quelques propriétés des opérateurs angle- +bornés et n-cycliquement monotones. Israel Journal of Mathematics, 26:137–150, +1977. +[BKK19] Shaojie Bai, J Zico Kolter, and Vladlen Koltun. Deep equilibrium models. Advances +in Neural Information Processing Systems, 32, 2019. +[CDB+21] Bingqing Chen, Priya L Donti, Kyri Baker, J Zico Kolter, and Mario Bergés. Enforcing +policy feasibility constraints through differentiable projection for energy optimization. +In Proceedings of the Twelfth ACM International Conference on Future Energy Systems, +pages 199–210, 2021. +[Con16] Laurent Condat. Fast projection onto the simplex and the ℓ1 ball. Mathematical +Programming, 158(1):575–585, 2016. +[Daf88] Stella Dafermos. Sensitivity analysis in variational inequalities. Mathematics of +Operations Research, 13(3):421–434, 1988. +[Dan66] John M Danskin. The theory of max-min, with applications. SIAM Journal on Applied +Mathematics, 14(4):641–664, 1966. +[DSSSC08] John Duchi, Shai Shalev-Shwartz, Yoram Singer, and Tushar Chandra. Efficient +projections onto the ℓ1-ball for learning in high dimensions. In Proceedings of the +25th international conference on Machine learning, pages 272–279, 2008. +[DY17] Damek Davis and Wotao Yin. A three-operator splitting scheme and its optimization +applications. Set-valued and variational analysis, 25(4):829–858, 2017. +[EG22] Adam N Elmachtoub and Paul Grigas. Smart “predict, then optimize”. Management +Science, 68(1):9–26, 2022. +[FHL+22] Samy Wu Fung, Howard Heaton, Qiuwei Li, Daniel McKenzie, Stanley Osher, and +Wotao Yin. Jfb: Jacobian-free backpropagation for implicit networks. In Proceedings +of the AAAI Conference on Artificial Intelligence, 2022. +11 + +[FNP+16] Fei Fang, Thanh Hong Nguyen, Rob Pickles, Wai Y Lam, Gopalasamy R Clements, +Bo An, Amandeep Singh, Milind Tambe, Andrew Lemieux, et al. Deploying paws: +Field optimization of the protection assistant for wildlife security. In AAAI, volume 16, +pages 3966–3973, 2016. +[FP03] Francisco Facchinei and Jong-Shi Pang. Finite-dimensional variational inequalities +and complementarity problems. Springer, 2003. +[FWDT20] Aaron Ferber, Bryan Wilder, Bistra Dilkina, and Milind Tambe. Mipaal: Mixed integer +program as a layer. In Proceedings of the AAAI Conference on Artificial Intelligence, +volume 34, pages 1504–1511, 2020. +[Guy] Jean Guyomarch. Warcraft ii open-source map editor, 2017. URL http://github. +com/war2/war2edit. +[HF22] Howard Heaton and Samy Wu Fung. Explainable ai via learning to optimize. arXiv +preprint arXiv:2204.14174, 2022. +[HML+21] Howard Heaton, Daniel McKenzie, Qiuwei Li, Samy Wu Fung, Stanley Osher, and +Wotao Yin. Learn to predict equilibria via fixed point networks. arXiv preprint +arXiv:2106.00906, 2021. +[HWFGY21] Howard Heaton, Samy Wu Fung, Aviv Gibali, and Wotao Yin. Feasibility-based fixed +point networks. Fixed Point Theory and Algorithms for Sciences and Engineering, +2021(1):1–19, 2021. +[Kar72] Richard M Karp. Reducibility among combinatorial problems. In Complexity of +computer computations, pages 85–103. Springer, 1972. +[KKM21] Imed Kacem, Hans Kellerer, and A Ridha Mahjoub. Preface: New trends on combi- +natorial optimization for network and logistical applications. Annals of Operations +Research, 298(1):1–5, 2021. +[LMY21] Qiuwei Li, Daniel McKenzie, and Wotao Yin. +From the simplex to the sphere: +Faster constrained optimization using the hadamard parametrization. arXiv preprint +arXiv:2112.05273, 2021. +[MG20] Jayanta Mandi and Tias Guns. Interior point solving for lp-based prediction+ optimi- +sation. Advances in Neural Information Processing Systems, 33:7272–7282, 2020. +[NS17] Yurii Nesterov and Vladimir Spokoiny. Random gradient-free minimization of convex +functions. Foundations of Computational Mathematics, 17:527–566, 2017. +[ONL+21] Derek Onken, Levon Nurbekyan, Xingjian Li, Samy Wu Fung, Stanley Osher, and +Lars Ruthotto. A neural network approach applied to multi-agent optimal control. In +2021 European Control Conference (ECC), pages 1036–1041. IEEE, 2021. +[ONL+22] Derek Onken, Levon Nurbekyan, Xingjian Li, Samy Wu Fung, Stanley Osher, and Lars +Ruthotto. A neural network approach for high-dimensional optimal control applied to +multiagent path finding. IEEE Transactions on Control Systems Technology, 2022. +[PGM+19] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory +Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: +An imperative style, high-performance deep learning library. Advances in neural +information processing systems, 32, 2019. +[PPM+19] Marin Vlastelica Poganˇci´c, Anselm Paulus, Vit Musil, Georg Martius, and Michal Ro- +linek. Differentiation of blackbox combinatorial solvers. In International Conference +on Learning Representations, 2019. +[RMB+22] Zaccharie Ramzi, Florian Mannel, Shaojie Bai, Jean-Luc Starck, Philippe Ciuciu, +and Thomas Moreau. Shine: Sharing the inverse estimate from the forward pass for +bi-level optimization and implicit models. In ICLR 2022-International Conference on +Learning Representations, 2022. +12 + +[RY21] Ernest K Ryu and Wotao Yin. Large-scale convex optimization via monotone operators, +2021. +[RY22] Ernest Ryu and Wotao Yin. Large-Scale Convex Optimization: Algorithm Designs via +Monotone Operators. Cambridge University Press, Cambridge, England, 2022. +[SE10] Abdelkader Sbihi and Richard W Eglese. Combinatorial optimization and green +logistics. Annals of Operations Research, 175(1):159–175, 2010. +[WDT19] Bryan Wilder, Bistra Dilkina, and Milind Tambe. Melding the data-decisions pipeline: +Decision-focused learning for combinatorial optimization. In Proceedings of the AAAI +Conference on Artificial Intelligence, volume 33, pages 1658–1665, 2019. +[WT21] Qi Wang and Chunlei Tang. Deep reinforcement learning for transportation network +combinatorial optimization: A survey. Knowledge-Based Systems, 233:107526, 2021. +[WXQ+06] Hao Wang, Haiyong Xie, Lili Qiu, Yang Richard Yang, Yin Zhang, and Albert Green- +berg. Cope: Traffic engineering in dynamic networks. In Proceedings of the 2006 +conference on Applications, technologies, architectures, and protocols for computer +communications, pages 99–110, 2006. +[Zie12] Günter M Ziegler. Lectures on polytopes, volume 152. Springer Science & Business +Media, 2012. +[ZT21] Liwei Zhong and Guochun Tang. Preface: Combinatorial optimization drives the +future of health care. Journal of Combinatorial Optimization, 42(4):675–676, 2021. +A +Experimental Details +A.1 +Additional Model Details +Our implementation of PertOpt-net used a PyTorch implementation15 of the original TensorFlow +code16 associated to the paper [BBT+20]. We train PertOpt-net using the argmin loss (see +Section 2.2), also referred to as MSE loss in the text. We do so for consistency with the other two +models tested. We experimented with various hyperparameter settings for 5-by-5 grids and found +setting the number of samples equal to 3, the temperature (i.e. ε) to 1 and using Gumbel noise to +work best, so we used these values for all other experiments. +A.2 +Additional Training Details +To train DYS-Net and cvxpylayers, we use an initial learning rate of 10−2 and use a scheduler that +reduces whenever the loss plateaus - we found this to perform the best for these two models. For +PertOpt-Net, however, we found that using a fixed learning rate of 10−2 performed the best. All +networks were trained using a AMD Threadripper Pro 3955WX: 16 cores, 3.90 GHz, 64 MB cache, +PCIe 4.0 CPU and an NVIDIA RTX A6000 GPU. Training times can be seen in Figure 5. +15See code at github.com/tuero/perturbations-differential-pytorch +16See code at github.com/google-research/google-research/tree/master/perturbations +13 + diff --git a/vdFQT4oBgHgl3EQfuzZ_/content/tmp_files/load_file.txt b/vdFQT4oBgHgl3EQfuzZ_/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..21ddfe1c455e2a69683238b97fe94feda0b74220 --- /dev/null +++ b/vdFQT4oBgHgl3EQfuzZ_/content/tmp_files/load_file.txt @@ -0,0 +1,476 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf,len=475 +page_content='Faster Predict-and-Optimize with Three-Operator Splitting Daniel McKenzie∗ Samy Wu Fung∗† Howard Heaton‡ Abstract In many practical settings, a combinatorial problem must be repeatedly solved with similar, but distinct parameters w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Yet, w is not directly observed;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' only contextual data d that correlates with w is available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' It is tempting to use a neural network to predict w given d, but training such a model requires reconciling the discrete nature of combinatorial optimization with the gradient-based frameworks used to train neural networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' One approach to overcoming this issue is to consider a continuous relaxation of the combinatorial problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' While existing such approaches have shown to be highly effective on small prob- lems (10–100 variables) they do not scale well to large problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In this work, we show how recent results in operator splitting can be used to design such a system which is easy to train and scales effortlessly to problems with thousands of variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 1 Introduction Many high-stakes decision problems in healthcare[ZT21], logistics and scheduling [SE10, KKM21], and transportation [WT21] can be phrased as combinatorial optimization problems, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' to compute x⋆ = argmin x∈X f(x;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w), (1) where X ⊂ Rn is a finite constraint set and f(x) = w⊤x is a linear function4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Depending on the structure of X, the problem (1) may be straightforward (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' shortest path) or NP-hard (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Traveling Sales Person problem [Kar72]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In both cases, this problem is well-studied and there are a plethora of robust and reliable algorithms for solving (1) given both the constraint X and objective f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Our present interest is settings where f is only partially known.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Specifically, we aim to repeatedly compute x⋆(d) = argmin x∈X w(d)⊤x, (2) for different data d, where w is a function of d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Importantly, the precise relationship between d and w is unknown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We propose learning a model to approximate the unknown objective: wΘ(d) ≈ w(d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The data d is observed and is called the context.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' As an illustrative example to keep in mind, consider the shortest path prediction problem studied in [PPM+19, BBT+20] (see Figure 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Here, the goal is to find the shortest path (from top-left to bottom-right) through a randomly generated terrain map from the Warcraft II tileset [Guy].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Each 8-by-8 square in the map represents a vertex in a 12-by-12 grid graph (with diagonal edges).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The time required to traverse this vertex depends on the visual content of the square (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' a square containing water takes longer to traverse than a square containing land).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Here d is the image representing the map, ∗Dept.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' of Applied Mathematics and Statistics, Colorado School of Mines †Dept.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' of Computer Science, Colorado School of Mines ‡Typal Research, Typal LLC 4While we focus on linear objective functions, our techniques can be applied to more general, convex f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Preprint.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Under review.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='13395v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='LG] 31 Jan 2023 0 20 40 60 80 0 20 40 60 80 (a) Map 0 2 4 6 8 10 0 2 4 6 8 10 (b) Costs for Each Square 0 2 4 6 8 10 0 2 4 6 8 10 (c) Shortest Path Figure 1: The shortest path prediction problem [PPM+19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The contextual data d in (a) maps to the cost of traversing each square in (b), with darker shading representing lower cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The true shortest path is shown in (c).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w(d) is a vector of traversal times (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' costs) for each vertex5, the true objective is f(x) = w(d)⊤x, and x⋆(d) is the true shortest path (See (c) in 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Suppose we wish to use gradient-based methods (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Stochastic Gradient Descent) to train wΘ(d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The primary obstacle to doing so is “differentiating through” the solution to ˆx(d) = argmin x∈X wΘ(d)⊤x, (3) so as to compute an informative gradient with which to update Θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Given the combinatorial nature of X, this is a nontrivial task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Specifically, for many small perturbations of Θ, the solution ˆx(d) will be unchanged;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' yet, for some perturbations the solution ˆx(d) will “jump” to a different point in X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Hence the gradient dˆx � dwΘ is always either zero or undefined.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We emphasize that this behaviour is common to all problems of the form (1) with linear f(·;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' To compute an informative gradient, we follow recent works (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [WDT19]) and relax (3) to a quadratic program over the convex hull of X (see (13)) by adding a small regularizer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Building upon [HWFGY21], we use Davis-Yin Three Operator splitting [DY17] and Jacobian-Free Backpropagation (JFB) [FHL+22] to build a solver for this relaxation that is efficient and easy to backpropagate through.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Our approach is fast, easy to implement using our provided code6, and unlike several prior works (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' see [PPM+19, BBT+20]), runs completely on GPU.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Although prior works are mostly constrained to problems with fewer than one hundred variables, numerical examples herein demonstrate our approach, run using only standard computing resources, easily scales to combinatorial problems with tens of thousands of variables.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Paper Organization The rest of this paper is laid out as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We introduce the Predict-then- Optimize paradigm in Section 2, which is followed by discussion of related works in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Our proposed combination of a three-operator splitting architecture, called DYS-Net, and Jacobian-Free Backpropagation (JFB) is presented in Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Numerical examples are provided in Section 5 and concluding remarks in Section 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 2 The Predict-and-Optimize Paradigm 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1 LP reformulation We begin by observing that the combinatorial optimization problem (2) is equivalent to a linear program: x⋆(d) = argmin x∈X w(d)⊤x = argmin x∈Conv(X) w(d)⊤x, (4) 5Note [PPM+19] considers the vertex-weighted variant of the shortest problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In our experiments we consider the edge-weighted version.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We discuss this further in Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 6See https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='com/mines-opt-ml/SPO_with_DYS 2 where Conv(X) is the convex hull of X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' This is because for generic7 w(d) the solution to the linear program lies at a corner of the polytope Conv(X), which must be an element of X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Equivalently: ˆx(d) = argmin x∈X wΘ(d)⊤x = argmin x∈Conv(X) wΘ(d)⊤x (5) Henceforth we follow [WDT19, MG20] and others and focus exclusively on this LP reformulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2 The Loss and Training Data Recall that we wish to train a neural network wΘ(d) such that ˆx(d) ≈ x⋆(d), where ˆx(d) is defined in (5) and x⋆(d) is defined in (4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' What loss function, and what training data, should one use?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Several works [WXQ+06, FNP+16] suggest gathering training data of the form (d, w(d)) and then training so as to minimize the discrepancy8 between w(d) and wΘ(d) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Frequently, this is referred to as the “two-stage” approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For large or complex problems this can be sub-optimal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Intuitively [Ben97] this is because small errors in the approximation wΘ(d) ≈ w(d) in areas crucial to the optimization problem (4) can yield wildly different minimizers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Instead, one should choose a loss aligned with the end goal, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' one that seeks to minimize the discrepancy between x⋆ Θ(d) and x(d) [WDT19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We discuss two such losses here.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Regret Loss To define this loss,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' also known as the “Smart Predict-then Optimize” (SPO) loss [EG22] or task loss [MG20],' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' we first consider the regret associated to solving (2),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' given a particular d,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' with ˆx(d) instead of x⋆(d): R(Θ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w) := w(d)⊤ˆxΘ(d) − w(d)⊤x⋆(d) (6) As the second term is independent of model parameters (Θ) we may drop it and consider the loss: LR(Θ) = Ed∼D � w(d)⊤ˆxΘ(d) � (7) The required training data in this setting is context-weight pairs (d,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w(d)) where d is drawn from the distribution in question D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Note that finding a model with low task loss guarantees that cost of the model output (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w⊤ˆx(d)) is close to the true optimal cost (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' w⊤x⋆(d)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' However, it provides no guarantees as to how well ˆx(d) approximates x⋆(d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Argmin loss If w(d) is never observed directly, only x⋆(d), one can minimize the mismatch between ˆx(d) and x⋆(d): LA(Θ) = 1 2Ed∼D � ∥x⋆(d) − ˆxΘ(d)∥2� (8) A similar loss is used in [PPM+19], except that the Hamming distance between x⋆(d) and ˆx(d), not the ℓ2 distance, is computed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The required training data for this loss is context-solution pairs (d, x⋆(d)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='3 Argmin differentiation In practice we replace the true risks for either loss by empirical ones: LR(Θ) = m � i=1 w⊤ i ˆxΘ(di) (9) LA(Θ) = 1 2 m � i=1 ∥x⋆(di) − ˆxΘ(di)∥2 (10) 7For certain w(d) it may happen that an entire face of Conv(X) is a solution to the linear program.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In this case, (?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=') is no longer strictly correct;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' rather the left-hand side is a subset of the right-hand side.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In the interest of clarity we ignore this edge case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 8for example the least-square discrepancy ∥w(d) − wΘ(d)∥ 3 where wi are the observed/true weights corresponding to di.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In order to use gradient based methods, one has to compute the derivative with respect to Θ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Appealing to the chain rule: dLR dΘ = m � i=1 w⊤ i ∂ˆxi,Θ dwi,Θ dwi,Θ dΘ (11) dLA dΘ = m � i=1 (x⋆(di) − ˆxΘ(di))⊤ ∂ˆxi,Θ dwi,Θ dwi,Θ dΘ , (12) where wi,Θ := wΘ(di) is the neural network approximation evaluated at di.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For both LR and LA, the key difficulty is making sense of dˆxΘ/dw.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' As discussed in Section 1, ˆxΘ(d) is piecewise constant as a function of w, and this remains true for the LP reformulation (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' So, for all w either dx⋆ Θ(d)/dw = 0 or dx⋆ Θ(d)/dw is undefined—both cases do not yield an informative gradient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' To remedy this, [WDT19, MG20] propose adding a small amount of regularization to the objective function in (5) so that the objective function becomes strongly convex, whence ˆxΘ(d) becomes a continuously differentiable function of w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We shall follow [WDT19] and add a small quadratic regularizer, so that (5) becomes: ˆxΘ(d) = argmin x∈Conv(X) wΘ(d)⊤x + γ∥x∥2 2 � �� � ≜fΘ(x;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='d) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (13) Thus, for the rest of this paper we shall consider the problem ˆxΘ(d) = argmin x∈C fΘ(x;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) (14) where C is a polytope (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' the convex hull of a finite set) and fΘ(x;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) is strongly convex with respect to x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Our goal is to solve (14) and simultaneously computing the derivative9 dˆxΘ/dΘ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' this problem is frequently referred to as argmin differentiation as has received much attention lately [AK17, AAB+19b, ABB+19, BKK19, FHL+22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Crucially, we note that every polytope may be expressed in the standardized form10: C = {x ∈ Rn : Ax = b and x ≥ 0} , (15) where the inequality x ≥ 0 is interpreted componentwise, see [Zie12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 3 Prior Work The most common approach to computing dˆxΘ/dΘ, proposed in [AK17] and used in [WDT19, MG20, FWDT20], starts with the KKT conditions for constrained optimality: ∂fΘ ∂x (ˆxΘ) + A⊤ˆλ + ˆν = 0 Aˆx − b = 0 D(ˆν)ˆxΘ = 0 where ˆλ and ˆν ≥ 0 are Lagrange multipliers associated to the optimal solution ˆxΘ [Ber97] and D(ˆν) is a matrix with ˆν along its diagonal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Differentiating these equations with respect to Θ and rearranging one obtains: � � ∂2fΘ ∂x2 A I A⊤ 0 0 D(ˆν) 0 D(ˆxΘ) � � � � dˆxΘ dΘ dˆλ dΘ dˆν dΘ � � = � � ∂2fΘ ∂x∂θ 0 0 � � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (16) Solving (16) then yields dˆxΘ dΘ (as well as dˆλ dΘ and dˆν dΘ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The computational bottleneck in this approach is computing ˆλ and ˆν in addition to ˆxΘ for (14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' If dim(x) = n and A ∈ Rm×n this can be done with a primal-dual interior point method at a cost of O � max{n, m}3� [AK17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In principle it is possible 9For notational convenience we suppress the dependence on d 10The standardization process may add dummy variables, and so increase the dimension of the problem n 4 to exploit sparsity in A or ∂fΘ ∂x (ˆxΘ) to solve (16) faster, but in practice we observe the state-of-the-art implementation of this approach, cvxpylayers [AAB+19b], struggles with problems containing more than 100 variables, see Section 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Another approach, proposed for deep equilibrium models in [BKK19] and adapted to constrained optimization layers in [CDB+21, BBC+21] is to re-formulate (14) as a fixed point problem: Find ˆxΘ such that ˆxΘ = PC � ˆxΘ − α∂fΘ ∂x (ˆxΘ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) � (17) and then apply the implicit function theorem to obtain an explicit formula for dˆxΘ/dΘ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' However, the cost of computing PC can be prohibitive, see Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Finally, many works use a perturbation-based approach to define a continuously differentiable proxy for the solution to the unregularized optimization problem (5), which we rewrite here as g(w) = min x∈Conv(X) w⊤x, (18) omitting the dependence of w on d for notational clarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For example, [PPM+19] define a piecewise- affine interpolant to g(w).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The gradients of gλ(w) are strikingly easy to compute, requiring just one additional solve of (18) with perturbed cost w′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In [BBT+20] a stochastic perturbation is considered: gε(w) = EZ � min x∈Conv(X) (w + εZ)⊤ x � , (19) analogous to Nesterov-Spokoiny smoothing [NS17] in zeroth-order optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' By Danskin’s theorem [Dan66] the gradients of gε(w) are also easy to compute: ∇wgε(w) = EZ � argmin x∈Conv(X) (w + εZ)⊤ x � ≈ 1 m m � i=1 argmin x∈Conv(X) (w + εZi)⊤ x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (20) We implement this approach as PertOpt-net in Section 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We highlight that the advantage of such approaches is they easily wrap around existing combinatorial solvers (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Dijkstra for the shortest path problem), as only repeated solves of (18) are required for computing gradients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The disadvantage is that such solvers are usually run on CPU.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Thus, data needs to be shuttled between CPU and GPU when training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 4 DYS-Net We now describe our proposed architecture, DYS-Net.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We use this term to refer to the neural architecture (see Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1) and the custom backprop procedure (see Section 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' DYS-Net models may be trained using either the regret loss or the argmin loss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1 The Forward Pass We wish to compute ˆxΘ and dˆxΘ/dΘ where ˆxΘ = argmin x∈C⊂Rn fΘ(x;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (21) As we are interested in the case where n is very large, we are led to consider first-order methods (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' gradient descent) over second-order methods (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Newton’s method).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For example, one could consider Projected Gradient Descent (PGD): xk+1 = PC � xk − α∂fΘ ∂x (xk;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) � for k = 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' , K − 1 ˆxΘ(d) ≈ xK (22) This works for simple C for which there exists an explicit form of PC, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' when C is the probability simplex [DSSSC08, Con16, LMY21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' However, for general polytopes C no such form exists, and so one must resort to a second iterative procedure, run at every k, to compute PC(xk) as the solution to PC(xk) = argmin y∈C ∥y − xk∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (23) 5 Our core idea is to solve (21) using Davis-Yin splitting (DYS) [DY17] to avoid computation of PC in the forward pass.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' To this end, we recall the standardized representation of polytopes (15) and write: C = {x : Ax = b and x ≥ 0} = {x : Ax = b} � �� � ≜C1 ∩ {x : x ≥ 0} � �� � ≜C2 = C1 ∩ C2, (24) where we note PC is hard to compute but PC1 and PC2 can be computed cheaply in many settings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='11 The following theorem formulates (21) as a fixed point problem involving only PC1 and PC2, not PC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Suppose C = C1 ∩ C2 for convex C1 and C2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Suppose both Ci are polyhedral or have relative interiors with a point in common and fΘ(x;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) be strongly convex and L-smooth in x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For any α > 0 define: TΘ(x) ≜ x−PC1(x) + PC2 (2PC1(x)−x−α∇fΘ(PC1(x))) , (25) Then ˆxΘ solves (21) if and only if ˆxΘ = PC1(ˆzΘ) where ˆzΘ = TΘ(ˆzΘ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (26) Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' This follows from [HML+21, Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2] by first noting ˆxΘ solves (21) if and only if it is the solution to the variational inequality ∇xfΘ(ˆxΘ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d)⊤ (x − ˆxΘ) ≥ 0 for all x ∈ C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (27) As the operator ∇xfΘ(·;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) is L-Lipschitz continuous, it is 1/L-cocoercive by the Baillon-Haddad theorem [BH77, BC09].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The next result shows that the simple fixed point iteration method, applied with TΘ, will converge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Corollary 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Suppose fΘ is L-smooth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Take α ∈ (0, 2/L), and define the sequence zk+1 = TΘ(zk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (28) Then xk := PC1(zk) → ˆxΘ with rate O(1/k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' See [RY22, Sec 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1] The upshot is that we can compute ˆxΘ using only PC1 and PC2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Finally we note that if fΘ(·;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' d) is strongly convex with respect to x then x⋆ Θ(d) is unique for fixed d and Θ, and the mapping Θ �→ x⋆ Θ(d) is differentiable [Daf88, FP03, HML+21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2 Efficient Backpropagation Differentiating both sides of (26), we obtain dˆzΘ dΘ = ∂TΘ ∂Θ + ∂TΘ ∂z dˆzΘ dΘ (29) ⇒ dˆzΘ dΘ = J −1 Θ ∂TΘ ∂Θ where JΘ(z) = I − ∂TΘ ∂z (30) Thus, one may compute the gradient of the loss using the chain rule as follows: dℓ dΘ = dℓ dx dˆxΘ dΘ = dℓ dx �dPC1 dz dˆzΘ dΘ � = dℓ dx dPC1 dz J −1 Θ ∂TΘ ∂Θ (31) This approach requires solving a linear system with JΘ which becomes particularly expensive when dim(ˆxΘ) is large.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Instead, we use the recently introduced Jacobian-Free Backpropagation (JFB) in which the Jacobian JΘ is replaced with the identity matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' This leads to an approximation of the true gradient dℓ dΘ using pΘ = � ∂ℓ ∂x dPC1 dz ∂TΘ ∂Θ � Θ,x=x⋆ Θ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' (32) In [FHL+22] it is shown pΘ is a descent direction, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' � pΘ, dℓ dΘ � > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Moreover, JFB has been found to be effective in a wide variety of application domains imaging [HML+21, HWFGY21, HF22, RMB+22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Using JFB yields a massive computational saving, as we no longer need to solve a linear system with JΘ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 11Further splitting of C1 can yield simpler projections if a singular value decomposition of A is unavailable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 6 Algorithm 1 DYS-Net Trained with JFB 1: Input: A and b defining C, fΘ 2: Initialize Θ0 at random 3: Precompute SVD for PC1 4: for m = 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' , M − 1 do 5: Compute xK = PC1(zK) ≈ ˆxΘm using iteration (28) with TΘm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 6: Compute pΘ(xK) ≈ pΘ(ˆxΘ) using (32).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 7: Θm+1 = Θm − ηpΘm(xK).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 8: end for 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='3 Implementation The combination of DYS for the forward pass and JFB for backpropagation makes the algorithm simple to implement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' First, we note that PC1 is a projection onto the intersection of hyperplanes which is given by the following lemma [HML+21]: Lemma 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Let C1 = {x : Ax = b} where A ∈ Rm×n and b ∈ Rm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Then the projection onto C1 is given by PC1(z) = argmin x∈C1 1 2∥x − z∥2 = z − A†(Az − b) (33) where A† = UΣ−1V ⊤ and UΣV ⊤ is the compact singular value decomposition of A such that U and V have orthonormal columns and Σ is invertible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Thus, to efficiently apply PC1, a singular value decomposition of A is first computed off-line.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Also, once a parameterization is chosen for fΘ, standard autodifferentiation tools, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' PyTorch [PGM+19], can be used to compute its gradient with respect to its inputs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Moreover, backpropagation can also be simply implemented as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We first compute ˆxΘ without tracking its gradients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Next, since JFB is equivalent to backpropagating through the last layer of the fixed point iteration [FHL+22], we apply TΘ from (25) once more to ˆxΘ to track the gradients and backpropagate through the last application of TΘ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Thus, given ˆxΘ, we backpropagate in four lines of code as shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Backpropagation through x⋆ Θ y = apply_T(x_star) loss = criterion(y) loss.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='backward() optimizer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='step() Figure 2: Sample PyTorch code for backpropagation through DYS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The entire training process can thus be described in Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 5 Numerical Experiments We evaluate the performance of DYS-Net on a shortest path prediction problem inspired by [PPM+19] (see the discussion of Section1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1 Problem Setup In [PPM+19] a shortest path prediction problem is introduced for k-by-k grid graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The graph is constructed by partitioning a 8k-by-8k pixel image representing a randomly generated terrain from the computer game Warcraft II into 8-by-8 squares.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Each square represents a vertex12, and the time 12Note [PPM+19] considers the vertex-weighted variant of the shortest problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In our experiments we consider the edge-weighted version.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We discuss this further in Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 7 10-by-10 20-by-20 30-by-30 Test MSE Loss 0 25 50 75 100 epochs 10 1 DYS CVX PertOpt 0 25 50 75 100 epochs 10 2 10 1 DYS CVX PertOpt 0 25 50 75 100 epochs 10 2 10 1 DYS CVX PertOpt Training Time 0 25 50 75 100 epochs 0 20 40 60 80 DYS CVX PertOpt 0 25 50 75 100 epochs 0 100 200 300 DYS CVX PertOpt 0 25 50 75 100 epochs 0 200 400 600 800 DYS CVX PertOpt Figure 3: Comparison of of DYS-Net,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' cvxpylayers ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' and PertOptNet [BBT+20] for three different grid sizes: 10 × 10 (first column),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 20 × 20 (second column),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' and 30 × 30 (third column).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The first row shows the MSE loss vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' epochs of the testing dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The second row shows the training time vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' epochs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' grid size number of variables network size 5-by-5 40 500 10-by-10 180 2040 20-by-20 760 8420 30-by-30 1740 19200 50-by-50 4900 53960 100-by-100 19800 217860 Table 1: Number of variables (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' number of edges) per grid size for the shortest path problem described in Section 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Third column: number of parameters for all three models used: DYS-Net, cvxpyayers and PertOpt-Net taken to traverse this vertex depends on the visual content of the square (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' a square containing water takes longer to traverse than a square containing land).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The context d ∈ R8k×8k is then the visual information encoded by all these squares, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' an input image.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In [PPM+19] problem instances are considered for k ∈ {12, 18, 24, 30}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Data Generation We extend this experiment to larger graphs while keeping the intrinsic learning complexity of the task fixed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We do so by constructing a series of k × k grid graphs (see Fig 4) for k ∈ {5, 10, 20, 30, 50, 100} where the weights are an unknown function of a context parameter d ∈ R5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We do not allow diagonal edges thus each vertex (excepting those at corners and on sides) has four edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In all experiments we draw d uniformly at random from the unit cube [0, 1]5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We assume the true edge weights are given by w(d) = Wd where W ∈ R|E|×5, and E is the set of edges in the graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' While more complicated relationships between w and d can certainly be considered13, we deliberately choose a simple relationship between context and edge weights so that the forward and backward pass through the combinatorial solver is the bottleneck in training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Further details are presented in Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Models We test three different differential combinatorial solvers, based on cvxpylayers [AAB+19a], the perturbed optimization approach of Berthet et al [BBT+20], as well as the proposed 13and are easily implementable using our code: https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='com/mines-opt-ml/SPO_with_DYS 8 True Path DYS-net CVX-net PertOpt-net 10-by-10 20-by-20 30-by-30 Figure 4: True paths (column 1), paths predicted by DYS-net (column 2), CVX-net (column 3), and PertOpt-net (column 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Samples are taken from different grid sizes: 10-by-10 (row 1), 20-by-20 (row 2), and 30-by-30 (row 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' DYS-Net.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We use the exact same neural network architecture for wΘ(d) in all three approaches;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' a two layer fully connected neural network with leaky ReLU activation functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Network sizes can be seen in Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We refer to the network equipped with the various solvers as CVX-net, PertOpt-net, and DYS-net respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' CVX-net and DYS-net use the regularized, linear program form (13) of the shortest path problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We tuned the hyperparameters for each architecture to the best of our ability on the smallest problem (5-by-5 grid graphs) and then used these hyperparameter values for all other graph sizes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We train all three architectures for 100 epochs total on each problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2 Results In Figure 3, we show the test loss and training time per epoch for all three architectures: DYS-net, CVX-net, and PertOpt-net for 10-by-10, 20-by-20, and 30-by-30 grids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In terms of MSE loss, CVX-net and DYS-net lead to comparable performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In the second row of Figure 3, we observe the benefits of combining the three-operator splitting with JFB [FHL+22];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' in particular, DYS-Net trains much faster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Figure 4 shows some randomly selected outputs for the three architectures once fully trained.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Interpreting the outputs of DYS-net and CVX-net as (unnormalized) probabilities over the grid, one can use a greedy decoder to determine the most probable path from top-left to bottom-right.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For small grids, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 5-by-5, this most probable path coincides exactly with the true path for most d (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For larger grids, we find there are often slight differences between the predicted and true paths.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' This is not surprising, as the number of possible paths grows exponentially with k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Thus, the number of “almost shortest paths” grows too, making exact prediction of x⋆ d an unreasonable task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Hence, we use the regret (see (6)) of the trained model, averaged over the test set, 9 Wa) Test MSE Loss b) Training Time (in minutes) c) Regret Values 25 50 75 100 grid size 10 2 10 1 DYS CVX PertOpt 25 50 75 100 grid size 103 104 105 106 DYS CVX PertOpt 25 50 75 100 grid size 10 3 10 2 10 1 100 101 DYS CVX PertOpt Figure 5: a) Test MSE loss (left), b) training time in minutes (middle), and c) regret values (right) vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' gridsize for three methods: DYS-Net, cvxpylayers[AAB+19b], and PertOpt [BBT+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The grid sizes are chosen to be 5-by-5, 10-by-10, 20-by-20, 30-by-30, 50-by-50, and 100-by-100.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' All three algorithms are shown up to gridsize 30-by-30, however, CVX is unable to load or run problems with gridsize over 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Indeed, this is because the optimization variable x is too large.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Dimensions of the variables can be found in Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' PertOpt [BBT+20] can be trained on the larger problems but takes a substantial amount to train.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' as a measure of accuracy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 5 shows that while CVX-net and PertOpt-net achieve low regret for small grids, DYS-net model achieves a low regret for all grids.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In addition to training faster, DYS-net can also be trained for much larger problems, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=', 100-by-100 grids, as shown in Figure 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We found that CVX-net could not handle grids larger than 30-by-30, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' , problems with more than 1740 variables14 (see Table 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Importantly, PertOpt-net takes close to a week to train for the 100-by-100 problem, whereas DYS-net takes about a day (see right Figure 5b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For details on the training setup, see Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 6 Conclusions Predicted Path Accuracy % 0 25 50 75 100 epochs 0 25 50 75 100 DYS CVX PertOpt Figure 6: Accuracy (in percentage) of predicted paths on 5-by-5 grid during training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We present a new data-driven method for solv- ing combinatorial optimization problems where parameters are unknown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The core idea is to use an implicit learn-to-optimize approach and learn an optimization algorithm based on three- operator-splitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We call this model DYS-net.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Importantly, DYS-net is trained using Jacobian- Free Backpropagation (JFB), thus avoiding com- puationally expensive backpropagation through projections on to the feasible set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Our experiments shows DYS-net leads to better accuracy with substantially lower training times.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Our approach is also easy to implement as a consequence of JFB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' While this work focused on linear objective functions, DYS-Net can be applied to more general objective functions with uniformly-Lipschitz gradients [RY21].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Moreover, as the dimensions of problems increase, this problem becomes more akin to using deep learning for optimal control problem [ONL+21, ONL+22], where the aim is to find an optimal path that minimizes an energy functional.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Investigating these connections is a direction of future work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' References [AAB+19a] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Agrawal, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Amos, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Barratt, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Boyd, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Diamond, and Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Kolter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Differentiable convex optimization layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Advances in Neural Information Processing Systems, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 14This is to be expected, as discussed in in [AK17, AAB+19a] 10 [AAB+19b] Akshay Agrawal, Brandon Amos, Shane Barratt, Stephen Boyd, Steven Diamond, and J Zico Kolter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Differentiable convex optimization layers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Advances in neural information processing systems, 32, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [ABB+19] Akshay Agrawal, Shane Barratt, Stephen Boyd, Enzo Busseti, and Walaa M Moursi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Differentiating through a cone program.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv preprint arXiv:1904.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='09043, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [AK17] Brandon Amos and J Zico Kolter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Optnet: Differentiable optimization as a layer in neural networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In International Conference on Machine Learning, pages 136–145.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' PMLR, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [BBC+21] Mathieu Blondel, Quentin Berthet, Marco Cuturi, Roy Frostig, Stephan Hoyer, Felipe Llinares-López, Fabian Pedregosa, and Jean-Philippe Vert.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Efficient and modular implicit differentiation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv preprint arXiv:2105.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='15183, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [BBT+20] Quentin Berthet, Mathieu Blondel, Olivier Teboul, Marco Cuturi, Jean-Philippe Vert, and Francis Bach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Learning with differentiable pertubed optimizers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Advances in neural information processing systems, 33:9508–9519, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [BC09] Heinz H Bauschke and Patrick L Combettes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The baillon-haddad theorem revisited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv preprint arXiv:0906.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='0807, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Ben97] Yoshua Bengio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Using a financial training criterion rather than a prediction criterion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' International journal of neural systems, 8(04):433–443, 1997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Ber97] Dimitri P Bertsekas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Nonlinear programming.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Journal of the Operational Research Society, 48(3):334–334, 1997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [BH77] Jean-Bernard Baillon and Georges Haddad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Quelques propriétés des opérateurs angle- bornés et n-cycliquement monotones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Israel Journal of Mathematics, 26:137–150, 1977.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [BKK19] Shaojie Bai, J Zico Kolter, and Vladlen Koltun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Deep equilibrium models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Advances in Neural Information Processing Systems, 32, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [CDB+21] Bingqing Chen, Priya L Donti, Kyri Baker, J Zico Kolter, and Mario Bergés.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Enforcing policy feasibility constraints through differentiable projection for energy optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Proceedings of the Twelfth ACM International Conference on Future Energy Systems, pages 199–210, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Con16] Laurent Condat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Fast projection onto the simplex and the ℓ1 ball.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Mathematical Programming, 158(1):575–585, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Daf88] Stella Dafermos.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Sensitivity analysis in variational inequalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Mathematics of Operations Research, 13(3):421–434, 1988.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Dan66] John M Danskin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' The theory of max-min, with applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' SIAM Journal on Applied Mathematics, 14(4):641–664, 1966.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [DSSSC08] John Duchi, Shai Shalev-Shwartz, Yoram Singer, and Tushar Chandra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Efficient projections onto the ℓ1-ball for learning in high dimensions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Proceedings of the 25th international conference on Machine learning, pages 272–279, 2008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [DY17] Damek Davis and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' A three-operator splitting scheme and its optimization applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Set-valued and variational analysis, 25(4):829–858, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [EG22] Adam N Elmachtoub and Paul Grigas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Smart “predict, then optimize”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Management Science, 68(1):9–26, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [FHL+22] Samy Wu Fung, Howard Heaton, Qiuwei Li, Daniel McKenzie, Stanley Osher, and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Jfb: Jacobian-free backpropagation for implicit networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Proceedings of the AAAI Conference on Artificial Intelligence, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 11 [FNP+16] Fei Fang, Thanh Hong Nguyen, Rob Pickles, Wai Y Lam, Gopalasamy R Clements, Bo An, Amandeep Singh, Milind Tambe, Andrew Lemieux, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Deploying paws: Field optimization of the protection assistant for wildlife security.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In AAAI, volume 16, pages 3966–3973, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [FP03] Francisco Facchinei and Jong-Shi Pang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Finite-dimensional variational inequalities and complementarity problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Springer, 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [FWDT20] Aaron Ferber, Bryan Wilder, Bistra Dilkina, and Milind Tambe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Mipaal: Mixed integer program as a layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 1504–1511, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Guy] Jean Guyomarch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Warcraft ii open-source map editor, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' URL http://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' com/war2/war2edit.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [HF22] Howard Heaton and Samy Wu Fung.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Explainable ai via learning to optimize.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv preprint arXiv:2204.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='14174, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [HML+21] Howard Heaton, Daniel McKenzie, Qiuwei Li, Samy Wu Fung, Stanley Osher, and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Learn to predict equilibria via fixed point networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv preprint arXiv:2106.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='00906, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [HWFGY21] Howard Heaton, Samy Wu Fung, Aviv Gibali, and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Feasibility-based fixed point networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Fixed Point Theory and Algorithms for Sciences and Engineering, 2021(1):1–19, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Kar72] Richard M Karp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Reducibility among combinatorial problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Complexity of computer computations, pages 85–103.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Springer, 1972.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [KKM21] Imed Kacem, Hans Kellerer, and A Ridha Mahjoub.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Preface: New trends on combi- natorial optimization for network and logistical applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Annals of Operations Research, 298(1):1–5, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [LMY21] Qiuwei Li, Daniel McKenzie, and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' From the simplex to the sphere: Faster constrained optimization using the hadamard parametrization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' arXiv preprint arXiv:2112.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='05273, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [MG20] Jayanta Mandi and Tias Guns.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Interior point solving for lp-based prediction+ optimi- sation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Advances in Neural Information Processing Systems, 33:7272–7282, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [NS17] Yurii Nesterov and Vladimir Spokoiny.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Random gradient-free minimization of convex functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Foundations of Computational Mathematics, 17:527–566, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [ONL+21] Derek Onken, Levon Nurbekyan, Xingjian Li, Samy Wu Fung, Stanley Osher, and Lars Ruthotto.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' A neural network approach applied to multi-agent optimal control.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In 2021 European Control Conference (ECC), pages 1036–1041.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' IEEE, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [ONL+22] Derek Onken, Levon Nurbekyan, Xingjian Li, Samy Wu Fung, Stanley Osher, and Lars Ruthotto.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' A neural network approach for high-dimensional optimal control applied to multiagent path finding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' IEEE Transactions on Control Systems Technology, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [PGM+19] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Pytorch: An imperative style, high-performance deep learning library.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Advances in neural information processing systems, 32, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [PPM+19] Marin Vlastelica Poganˇci´c, Anselm Paulus, Vit Musil, Georg Martius, and Michal Ro- linek.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Differentiation of blackbox combinatorial solvers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In International Conference on Learning Representations, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [RMB+22] Zaccharie Ramzi, Florian Mannel, Shaojie Bai, Jean-Luc Starck, Philippe Ciuciu, and Thomas Moreau.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Shine: Sharing the inverse estimate from the forward pass for bi-level optimization and implicit models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In ICLR 2022-International Conference on Learning Representations, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 12 [RY21] Ernest K Ryu and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Large-scale convex optimization via monotone operators, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [RY22] Ernest Ryu and Wotao Yin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Large-Scale Convex Optimization: Algorithm Designs via Monotone Operators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Cambridge University Press, Cambridge, England, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [SE10] Abdelkader Sbihi and Richard W Eglese.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Combinatorial optimization and green logistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Annals of Operations Research, 175(1):159–175, 2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [WDT19] Bryan Wilder, Bistra Dilkina, and Milind Tambe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 1658–1665, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [WT21] Qi Wang and Chunlei Tang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Deep reinforcement learning for transportation network combinatorial optimization: A survey.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Knowledge-Based Systems, 233:107526, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [WXQ+06] Hao Wang, Haiyong Xie, Lili Qiu, Yang Richard Yang, Yin Zhang, and Albert Green- berg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Cope: Traffic engineering in dynamic networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' In Proceedings of the 2006 conference on Applications, technologies, architectures, and protocols for computer communications, pages 99–110, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [Zie12] Günter M Ziegler.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Lectures on polytopes, volume 152.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Springer Science & Business Media, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' [ZT21] Liwei Zhong and Guochun Tang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Preface: Combinatorial optimization drives the future of health care.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Journal of Combinatorial Optimization, 42(4):675–676, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' A Experimental Details A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='1 Additional Model Details Our implementation of PertOpt-net used a PyTorch implementation15 of the original TensorFlow code16 associated to the paper [BBT+20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We train PertOpt-net using the argmin loss (see Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2), also referred to as MSE loss in the text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We do so for consistency with the other two models tested.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' We experimented with various hyperparameter settings for 5-by-5 grids and found setting the number of samples equal to 3, the temperature (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' ε) to 1 and using Gumbel noise to work best, so we used these values for all other experiments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='2 Additional Training Details To train DYS-Net and cvxpylayers, we use an initial learning rate of 10−2 and use a scheduler that reduces whenever the loss plateaus - we found this to perform the best for these two models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' For PertOpt-Net, however, we found that using a fixed learning rate of 10−2 performed the best.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' All networks were trained using a AMD Threadripper Pro 3955WX: 16 cores, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='90 GHz, 64 MB cache, PCIe 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='0 CPU and an NVIDIA RTX A6000 GPU.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' Training times can be seen in Figure 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content=' 15See code at github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='com/tuero/perturbations-differential-pytorch 16See code at github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} +page_content='com/google-research/google-research/tree/master/perturbations 13' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/vdFQT4oBgHgl3EQfuzZ_/content/2301.13395v1.pdf'} diff --git a/wtAyT4oBgHgl3EQfnfhR/content/tmp_files/2301.00489v1.pdf.txt b/wtAyT4oBgHgl3EQfnfhR/content/tmp_files/2301.00489v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..a589395432db5e6d7f19a3ef181ab41f7c27de36 --- /dev/null +++ b/wtAyT4oBgHgl3EQfnfhR/content/tmp_files/2301.00489v1.pdf.txt @@ -0,0 +1,1757 @@ +Federated Learning with Client-Exclusive Classes +Jiayun Zhang1, Xiyuan Zhang1, Xinyang Zhang2, Dezhi Hong3†, Rajesh K. Gupta1, Jingbo Shang1 +1University of California, San Diego, La Jolla, CA, USA +2University of Illinois at Urbana-Champaign, Champaign, IL, USA 3Amazon +{jiz069,xiyuanzh,gupta,jshang}@ucsd.edu,xz43@illinois.edu,hondezhi@amazon.com +ABSTRACT +Existing federated classification algorithms typically assume the +local annotations at every client cover the same set of classes. In this +paper, we aim to lift such an assumption and focus on a more gen- +eral yet practical non-IID setting where every client can work on +non-identical and even disjoint sets of classes (i.e., client-exclusive +classes), and the clients have a common goal which is to build a +global classification model to identify the union of these classes. +Such heterogeneity in client class sets poses a new challenge: how +to ensure different clients are operating in the same latent space so +as to avoid the drift after aggregation? We observe that the classes +can be described in natural languages (i.e., class names) and these +names are typically safe to share with all parties. Thus, we formu- +late the classification problem as a matching process between data +representations and class representations and break the classifica- +tion model into a data encoder and a label encoder. We leverage +the natural-language class names as the common ground to anchor +the class representations in the label encoder. In each iteration, the +label encoder updates the class representations and regulates the +data representations through matching. We further use the updated +class representations at each round to annotate data samples for +locally-unaware classes according to similarity and distill knowl- +edge to local models. Extensive experiments on four real-world +datasets show that the proposed method can outperform various +classical and state-of-the-art federated learning methods designed +for learning with non-IID data. +KEYWORDS +Federated Learning, non-IID, Label Name Sharing, Knowledge Dis- +tillation +1 +INTRODUCTION +Federated learning [30] emerges as a distributed learning paradigm +that allows multiple parties to collaboratively learn a global model +effective for all parties while ensuring the privacy of local data. +This benefits applications in a wide range of domains, such as +recommendation [23, 27, 47], ubiquitous sensing [10, 16, 17, 39] +and mobile computing [15, 19, 46]. +Existing federated classification methods, even those designed +for non-IID clients [9, 20, 21, 25, 28, 43, 44, 50], typically assume +that the local annotations at each client follow the same set of +classes; however, this assumption does not hold true in many real- +world applications. For example, a smartwatch company wants to +build a human activity classifier for all activity types, as shown +in Figure 1(a). Although their smartwatch users as clients could +experience almost all types of daily activities, each user may only +†Work unrelated to Amazon. +Posture: lying down, sitting. +Activity: sleeping. +Posture: standing, walking. +Activity: sleeping, cleaning. +Environment: home, gym. +Posture: standing, walking, +sitting, lying down. +Posture: standing, lying down. +Activity: sleeping, cleaning. +Environment: home, gym. +Learning +Training client’s class set +Testing client’s class set +Inference +Server (w/ global shared model) +Posture: lying down, sitting. +Activity: sleeping. +Posture: standing, walking. +Activity: sleeping, cleaning. +Environment: home, gym. +Posture: standing, walking, +sitting, lying down. +identified classes +Posture +Activity +Environment +(full class set) +Client +Server +unaware classes +Posture: sitting +Activity: Internet surfing +Environment: restaurant +Posture: standing, walking +Client A +Client C +Server aggregation +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +run sit sleep eat home gym +Client local data +sleep +sit +run +eat +home +gym +universal set +of classes +global model +Server +√ ground truth +data +data rep. (identified classes) +class rep. +Client A +Client B +Server aggregation +data rep. (unaware classes) +(optimal) +(classes) +run +sit +gym +home +eat +data +Representations: +Client +A +B +C +(a) The problem setting of non-identical client class sets. +Posture: lying down, sitting. +Activity: sleeping. +Posture: standing, walking. +Activity: sleeping, cleaning. +Environment: home, gym. +Posture: standing, walking, +sitting, lying down. +Posture: standing, lying down. +Activity: sleeping, cleaning. +Environment: home, gym. +Learning +Training client’s class set +Testing client’s class set +Inference +Server (w/ global shared model) +Posture: lying down, sitting. +Activity: sleeping. +Posture: standing, walking. +Activity: sleeping, cleaning. +Environment: home, gym. +Posture: standing, walking, +sitting, lying down. +identified classes +Posture +Activity +Environment +(full class set) +Client +Server +unaware classes +Posture: sitting +Activity: Internet surfing +Environment: restaurant +Posture: standing, walking +Client A +Client C +Server aggregation +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +√ +run sit sleep eat home gym +Client local data +sleep +sit +run +eat +home +gym +universal set +of classes +global model +Server +√ ground truth +data +data rep. (identified classes) +class rep. +Client A +Client B +Server aggregation +data rep. (unaware classes) +(optimal) +(classes) +run +sit +gym +home +eat +data +Representations: +Client +A +B +C +(b) Unique challenge. +Figure 1: (a) illustrates our problem setting using a behav- +ioral context recognition system where users have different +preferences in reporting (i.e., annotating) labels. (b) demon- +strates the unique challenge here — local models at different +clients may operate in different and even independent latent +spaces, making it hard to aggregate at the server. +opt to report (i.e., annotate) a subset of activities. Another example +is a federated medical diagnosis system, which attempts to infer +all types of diseases of a patient for comprehensive health screen- +ing. Physicians and specialist groups with different expertise can +participate in this federated learning system as clients. As one can +see here, different specialists will only offer disease annotations +within their domains, even if a patient may have several types of +diseases at the same time. This makes the class sets at many clients +non-identical and even non-overlapping. +We aim to lift this assumption and work on a new and rather prac- +tical federated learning setting, client-exclusive classes, where +the local annotations at each client cover different and even non- +overlapping sets of classes. We denote the classes that are not +covered in the local annotations as locally-unaware classes at each +client. Each client can hold local data without any annotations, no +matter whether their true labels are among the locally-unaware +classes or not. Also, the classification task here can be either single- +label or multi-label. When it is multi-label, the local data might be +only partially labeled due to the locally-unaware classes. There- +fore, this new setting is more general and challenging than the +missing class scenario [22], which is only applicable to single-label +classification. +The non-identical client class sets pose a significant challenge of +huge variance in local training across different clients. As shown +in Figure 1(b), one can view classification as a matching process +between data representations and label representations in a latent +space. Because of the non-identical client class sets, locally trained +classifiers are more likely to operate in drastically different latent +spaces. Moreover, when the class sets are non-overlapping, it is +arXiv:2301.00489v1 [cs.LG] 1 Jan 2023 + +山山possible that the latent spaces at different clients are completely in- +dependent. This would result in inaccurate classification boundaries +after aggregation at the server, making our setting more challenging +than non-IID clients with identical client class sets. +We propose a novel federated learning framework FedAlign to +better align the latent spaces across clients from both label and data +perspectives as follows: +(1) Anchor the label representations using label names. We observe +that the natural-language class names (i.e., label names) often +carry valuable information for understanding label semantics, +and more importantly, they are typically safe to share with all +parties. Therefore, we break the classification model into a data +encoder and a label encoder as shown in Figure 2, and then lever- +age the label names as the common ground for label encoders. +The server initializes the label encoder with pre-trained text +representations, such as word embedding. The label encoder +will be then distributed to different clients and updated alter- +natively with data encoders during local training and global +aggregation, mutually regulating the latent space. +(2) Connect the data representations via locally-unaware classes. The +representations given by the label encoder are representative +of the classes. Their distances to the data samples in the latent +space can tell how likely a sample belongs to a class. Therefore, +we introduce regularization in a knowledge distillation man- +ner. Specifically, as shown in Figure 2, at each client, we first +annotate local data based on their similarities with the repre- +sentations of the locally-unaware classes, and then add another +cross-entropy loss between the pseudo-annotations and the +predictions by local models. Such regularization connects data +encoders across different clients, so they have a better chance +to be in the same latent space. +We conduct experiments on four real-world applications, includ- +ing the most challenging scenario of multi-label classification with +non-overlapping client class sets. We show that FedAlign can out- +perform various state-of-the-art federated learning algorithms. Our +contributions are as follows: +• We lift the common assumption of identical client class sets in +federated learning and study a more general yet practical setting, +non-identical client class sets. Such heterogeneity in client class +sets poses a new challenge — local models at different clients +may operate in different and even independent latent spaces. +• We propose a novel framework FedAlign to better align the +latent spaces across clients by anchoring the label representations +using label names and connecting data representations via locally- +unaware classes. +• We conduct extensive experiments on four real-world datasets +of different tasks and show that FedAlign can further improve +various state-of-the-art federated learning algorithms. +2 +PRELIMINARIES +Problem Formulation. We aim to generate a global classification +model using federated learning with non-identical class sets, where +each client only identifies part of the classes from its dataset. Given +the universal set of classes as C, the set of classes that are identified +on client 𝑚 is C𝑚, and the set of locally-unaware classes is C𝑚, +where C𝑚 ∪ C𝑚 = C. The objective is to learn a global model +𝑔 : 𝑋 → C that given input feature 𝑋, all positive labels from the +class set 𝐶 can be inferred. +The training set on client 𝑚 is denoted as D𝑚 = {(𝑥𝑖,𝑦𝑖)}𝑁 +𝑖=1, +where 𝑥𝑖 is the data feature and 𝑦𝑖 = (𝑦𝑖,𝑐1, . . . ,𝑦𝑖,𝑐|C|) is a vector +showing the labels of each class. If𝑐𝑗 ∈ 𝐶𝑚,𝑦𝑖,𝑐 𝑗 ∈ {0, 1}. If𝑐𝑗 ∈ C𝑚, +𝑦𝑖,𝑐 𝑗 is unknown. It is possible that some data samples 𝑥𝑖 ∈ 𝐷𝑚 do +not belong to any of the classes in C𝑚, i.e., ∀𝑐 ∈ C𝑚 : 𝑦𝑖,𝑐 = 0. +Backbone Classification Model. Generally, the classification model +can be decomposed into a data encoder and a classification layer. +The data encoder takes the input data 𝑥𝑖 and generates a data +representation. Then, a linear layer (i.e., classifier) transforms the +representation into prediction logits. We discuss two types of clas- +sification tasks as follows. +Single-label multi-class classification. In this setting, each sam- +ple is associated with only one positive class. In other words, the +classes are mutually exclusive. We use the softmax activation func- +tion to get the prediction probability. Denote 𝑔𝑚(·) as the classifica- +tion model at client 𝑚 and the prediction probability of 𝑥𝑖 on class +𝑐 as 𝑔(𝑥𝑖)𝑐. The class with the maximum probabilities is predicted +as positive, i.e., ˜𝑦𝑖 = 𝑎𝑟𝑔𝑚𝑎𝑥 +𝑐 +(𝑔𝑚(𝑥𝑖)𝑐). The loss function at client +𝑚 is: +LC𝑚 = − 1 +𝑁𝑚 +𝑁𝑚 +∑︁ +𝑖=1 +∑︁ +𝑐 ∈𝐶𝑚 +𝑦𝑖,𝑐 log𝑔𝑚(𝑥𝑖)𝑐 +Multi-label classification. In this setting, each sample may be +associated with a set of positive classes. For example, a person +may have both diabetes and hypertension. The sigmoid function is +applied to get the prediction probability, each element of which is +the probability that the input data 𝑥𝑖 is associated with each class. +The final predictions are achieved by thresholding the probabilities +at 0.5. If 𝑔(𝑥𝑖)𝑐 > 0.5, sample 𝑥𝑖 is predicted to be in class 𝑐, i.e., +˜𝑦𝑖,𝑐 = 1. The learning objective is to minimize the binary cross- +entropy loss over the identified classes: +LC𝑚 = − 1 +𝑁𝑚 +𝑁𝑚 +∑︁ +𝑖=1 +∑︁ +𝑐 ∈C𝑚 +[𝑦𝑖,𝑐 log𝑔𝑚(𝑥𝑖)𝑐+(1−𝑦𝑖,𝑐) log(1−𝑔𝑚(𝑥𝑖)𝑐)]. +3 +THE FEDALIGN FRAMEWORK +3.1 +Overview +The pseudo code of FedAlign can be found in Algorithm 1. Learning +with FedAlign framework consists of the following steps: +(1) Label name sharing and label encoder initialization: Be- +fore training, the server collects the natural language label +names from the clients. The server initializes the label encoder +𝜃0 via pre-trained text representations, such as word embedding. +We expect more advanced techniques like pre-trained neural +language models could make the learning converge even faster, +but we leave it as future work. The data encoder 𝜔0 is randomly +initialized. +(2) Client selection and model communication: At 𝑡-th round, +the server randomly selects a subset of clients 𝑆𝑡 and sends the +global model parameters to them. +(3) Local training: Client 𝑚 ∈ 𝑆𝑡 independently trains its local +model and returns the parameters. +(4) Model aggregation: The server aggregates the parameters of +client models into global parameters. +2 + +… +run, sit, +stand… +sleep, +sing… +gym, +home… +Server +Server +pretrain +initialize +round 0: initialization +round #: training & communication +Communication rounds +Client $ … & +KD loss +CE loss +!' +( +"' +! +features +label +names +label +encoder '* +data +encoder (* +. +logits +label +form distillation set +train data encoder +train label encoder +CE loss +!' +( +"' +label +aggregation +data encoder (+ +label encoder '+ +data encoder (+,- +label encoder '+,- +data encoder (+ +label encoder '+ +transfer parameters +global model +label name sharing +class rep. +data rep. +(pseudo-annotated) +. +. +freeze +freeze +Figure 2: Overview of FedAlign framework. The label names are leveraged as a common ground for label encoders to anchor +class representations. During local training, the two encoders perform alternating training to mutually regulate the latent +spaces. The labels about locally-unaware classes are assigned to the local samples based on the similarities between the data +representations and class representations and the knowledge is transferred to local models via knowledge distillation objec- +tives. +Algorithm 1: FedAlign Framework +Input +:Communication rounds 𝑇, number of selected +clients each round |𝑆𝑡 |, local training epochs 𝐸. +Output:The final global model 𝑔(𝑥;𝜃𝑇,𝜔𝑇 ). +1 Server executes: +2 Collect label names from clients and pre-train text +representations to initialize label encoder 𝜃0 ; +// Step (1) +3 Randomly initialize data encoder 𝜔0; +4 for 𝑡 = 0, 1, ...,𝑇 − 1 do +5 +Select a subset 𝑆𝑡 of clients at random ; +// Step (2) +6 +for 𝑚 ∈ 𝑆𝑡 do +7 +𝜃 (𝑡+1) +𝑚 +,𝜔 (𝑡+1) +𝑚 +← ClientUpdate(𝑚,𝜃 (𝑡),𝜔 (𝑡)); +// Step (3) +8 +𝜔 (𝑡+1) ← +1 +|𝑆𝑡 | +� +𝑚∈𝑆𝑡 +𝜔 (𝑡+1) +𝑚 +; +// Step (4) +9 +𝜃 (𝑡+1) +𝑐 +← +� +𝑚∈𝑆𝑡 ,𝑐∈C𝑚 +𝜃 (𝑡+1) +𝑚,𝑐 +|{𝑚|𝑚∈𝑆𝑡,𝑐 ∈C𝑚 }| ; +// Step (4) +10 return 𝜃𝑇,𝜔𝑇 +11 ClientUpdate(𝑚,𝜃 (𝑡),𝜔 (𝑡)): +12 𝜃 (𝑡) +𝑚 ,𝜔 (𝑡) +𝑚 ← 𝜃 (𝑡),𝜔 (𝑡); +13 for 𝑒𝑝𝑜𝑐ℎ ∈ {1, 2, ..., 𝐸} do +14 +Calculate similarities of data and class representations +and form distillation set ˜D (𝑡) +𝑚 ; +// Step (a) +15 +Freeze 𝜃 (𝑡) +𝑚 and update 𝜔 (𝑡+1) +𝑚 +; +// Step (b) +16 +Freeze 𝜔 (𝑡+1) +𝑚 +and update 𝜃 (𝑡+1) +𝑚 +; +// Step (c) +17 return 𝜃 (𝑡+1) +𝑚 +, and 𝜔 (𝑡+1) +𝑚 +to server ; +// Step (d) +Pre-training text representations and label encoder initialization in +(1) are conducted only once at the beginning. Steps (2)-(4) repeat for +𝑇 rounds until the global model converges. During local training +in (3), each client 𝑚 ∈ 𝑆𝑡 conducts the following steps: +(a) Form distillation dataset for locally-unaware classes: Client +𝑚 forms a distillation set ˜D (𝑡+1) +𝑚 +for locally-unaware classes +C𝑚 by using the latest class representations. +(b) Train data encoder: Client 𝑚 freezes the label encoder pa- +rameters and trains the data encoder. +(c) Train label encoder: Client 𝑚 freezes the parameters in the +data encoder and trains the label encoder. +(d) Model communication after local updates: Client𝑚 sends +the updated model parameters to the server. +3.2 +Semantic Label Name Sharing +The vanilla classification model in Section 2 learns latent label +spaces merely based on the local training data with numerical +label IDs. With non-identical client class sets, local models at dif- +ferent clients are likely to form different and even independent +label spaces, making the classification boundaries aggregated at +the server inaccurate. To better align the label spaces, we turn to +the semantics of label names as a common ground to anchor class +representations. The natural language label names carry valuable +information for understanding the label correlations. For example, +in behavioral context recognition, the activity of “lying down" is +likely to indicate the person is “sleeping", and the possible location +of the activity is “at home". Such knowledge about label correlations +not only exists in the datasets to investigate, but also can be mined +through analyzing the semantics of the label names. +Incorporating Label Encoder to Classification Model. We re- +place the classification layer in a conventional classification model +with a label encoder as shown in Figure 2. The label encoder +3 + +𝑓𝜃 : 𝑤𝑐 → 𝑟𝑐 ∈ 𝑅𝑑 takes the natural language label names 𝑤𝑐 as +the inputs and maps them into representations 𝑟𝑐. Prior knowledge +about label semantics can be inserted into the label encoder by ini- +tializing it with pre-trained label embeddings. Inspired by existing +works that learn semantic word embeddings based on word-word +co-occurrence and point-wise mutual information (PMI) [1, 14, 34], +we use an external text corpus related to the domain of the classifica- +tion task to extract knowledge of label co-occurrence and pre-train +the label embeddings for initializing the label encoder. The details +about label embedding pre-training can be found in Appendix. +For each input data 𝑥𝑖, the model calculates the matching scores +of 𝑥𝑖 and every class by taking the dot product of the data repre- +sentation ℎ𝜔 : 𝑥𝑖 → 𝑧𝑖 ∈ R𝑑 output from the data encoder and +the class representations output by the label encoder. This way, +it calculates the similarity between the representations of input +data and classes. Then, an activation function is applied to the dot +product to get the prediction probabilities of 𝑥𝑖 belonging to each +class. Same as described in Section 2, for multi-label classification, +the sigmoid function is used and the final prediction is achieved by +thresholding the probabilities at 0.5. For single-label classification, +the softmax function is used and the final prediction is the class +with the maximum probabilities. +Alternating Training of Encoders. The data encoder and label +encoder are two branches in the classification model. We want the +representations obtained by one encoder to regulate the training +of the other while preventing mutual interference. Thus, at each +learning iteration, we first freeze the parameters in the label encoder +𝜃 and update the parameters of the data encoder 𝜔. Then, we freeze +the parameters in the data encoder and update the label encoder +parameters. +3.3 +Knowledge Distillation for +Locally-unaware Classes +Due to the lack of label information of certain classes to support +supervised training, the training at each client is biased toward the +identified classes. To mitigate such drift, we design a knowledge +distillation strategy for unaware classes at each client by further +utilizing the class representations. The class representations are +updated with the training of the label encoder at each round of +communication. As the classification problem is formed as a match- +ing between the embeddings of data and label names in Section 3.2, +the distance between the label names and the data samples in the +latent space can tell how likely a sample belongs to a class. Thus, +we use the class representations to annotate samples for the locally- +unaware classes and distill knowledge to local models. +Forming Local Distillation Dataset. When the client receives +the parameters of the data encoder 𝜔 (𝑡) and label encoder 𝜃 (𝑡) at +𝑡-th round, it uses the label encoder to get the updated class repre- +sentations {𝑓 (𝑡) +𝜃 +(𝑤𝑐)|𝑐 ∈ C} and uses the data encoder to generate +the data representations of its local data {ℎ(𝑡) +𝜔 (𝑥𝑖)|(𝑥𝑖,𝑦𝑖) ∈ D𝑚}. +Then, for each locally-unaware class �𝑐, the client calculates its co- +sine similarities to the data: +𝑠 (𝑡) +𝑖,�𝑐 = +𝑓 (𝑡) +𝜃 +(𝑤 �𝑐) · ℎ(𝑡) +𝜔 (𝑥𝑖) +∥𝑓 (𝑡) +𝜃 +(𝑤 �𝑐)∥ · ∥ℎ(𝑡) +𝜔 (𝑥𝑖)∥ +. +Samples are annotated according to their similarities to the class. +The samples with the highest similarities with the class representa- +tion 𝑓 (𝑡) +𝜃 +(𝑤 �𝑐) are annotated as positive samples of class �𝑐. The client +also annotates the negative samples for the locally-unaware classes. +The sample with the least similarities with 𝑓 (𝑡) +𝜃 +(𝑤 �𝑐) is annotated +as a negative sample of �𝑐. +The number of samples to be annotated depends on the per- +centile of similarity. We set the thresholds ˆ𝜏 (𝑡) +�𝑐 +and ˇ𝜏 (𝑡) +�𝑐 +as the 𝑞1-th +and 𝑞2-th percentile of the similarities over all samples for annotat- +ing positive samples and negative samples respectively. The client +annotates a subset of the samples whose similarities are higher than +ˆ𝜏 (𝑡) +�𝑐 +as the positive samples in class �𝑐 (i.e., ˜𝑦(𝑡) +𝑖,�𝑐 = 1) and those with +selection scores lower than ˇ𝜏 (𝑡) +�𝑐 +as the negative samples in class �𝑐 +(i.e., ˜𝑦(𝑡) +𝑖,�𝑐 = 0). The distillation dataset after the 𝑡-th round is: +˜D (𝑡) +𝑚 ← {(𝑥𝑖, ˜𝑦(𝑡) +𝑖 +)|𝑠 (𝑡) +𝑖,�𝑐 > ˆ𝜏 (𝑡) +�𝑐 +or 𝑠 (𝑡) +𝑖,�𝑐 < ˇ𝜏 (𝑡) +�𝑐 +, �𝑐 ∈ C𝑚} +In single-label classification, there is another constraint that the +unlabeled sample can be annotated as a positive sample of class �𝑐 +only if the class is the closest to the data representation among all +classes, i.e., �𝑐 = 𝑎𝑟𝑔𝑚𝑎𝑥 +𝑐 ∈C +(𝑠 (𝑡) +𝑖,𝑐 ). +Knowledge Distillation. The pseudo annotations about the locally- +unaware classes C𝑚 are then used as knowledge to transfer to +client 𝑚’s local model. Let 𝑁 (𝑡) +�𝑐 +be the number of samples in ˜D (𝑡) +𝑚 , +𝑔(𝑡+1) +𝑚 +(·) be the local model with parameters being updated. We +use binary cross-entropy loss for multi-label classification as the +distillation loss: +L(𝑡) +C𝑚 = − 1 +𝑁 +′ +𝑚 +𝑁 +′ +𝑚 +∑︁ +𝑖=1 +∑︁ +�𝑐 ∈C𝑚 +[ ˜𝑦(𝑡) +𝑖,𝑐 log𝑔(𝑡+1) +𝑚 +(𝑥𝑖)�𝑐+ +(1 − ˜𝑦(𝑡) +𝑖,𝑐 ) log(1 − 𝑔(𝑡+1) +𝑚 +(𝑥𝑖)�𝑐)], +where 𝑁 +′ +𝑚 is the number of samples in the distillation set ˜D (𝑡) +𝑚 . +For single-label classification, we use cross-entropy loss as dis- +tillation loss: +L(𝑡) +C𝑚 = − 1 +𝑁 +′ +𝑚 +𝑁 +′ +𝑚 +∑︁ +𝑖=1 +∑︁ +𝑐 ∈C +˜𝑦(𝑡) +𝑖,𝑐 log𝑔(𝑡+1) +𝑚 +(𝑥𝑖)𝑐. +We use a parameter 𝛼 to control the weight of the distillation +loss term. The learning objective of the (𝑡)-th round on client 𝑚 is +to minimize the following loss term: +L(𝑡) +𝑚 = L(𝑡) +C𝑚 + 𝛼L(𝑡) +C𝑚 +3.4 +Model Aggregation +After the clients finish the local training, they send the updated +parameters to the server. After the server collects the parameters +from all participating clients 𝑚 ∈ 𝑆𝑡, it conducts model aggregation. +For the data encoder, the parameters are averaged across all client +models: +𝜔 (𝑡+1) ← +1 +|𝑆𝑡 | +∑︁ +𝑚∈𝑆𝑡 +𝜔 (𝑡+1) +𝑚 +. +4 + +For the label encoder, the representations of class 𝑐 are aver- +aged only among the client models that have class 𝑐 as its locally- +identified class: +𝜃 (𝑡+1) +𝑐 +← +� +𝑚∈𝑆𝑡,𝑐 ∈C𝑚 +𝜃 (𝑡+1) +𝑚,𝑐 +|{𝑚|𝑚 ∈ 𝑆𝑡,𝑐 ∈ C𝑚}| . +4 +EXPERIMENTS +We first introduce the applications and the datasets we use. Then, +we describe the compared methods and experimental setup. We +show the experimental results with different settings and conduct +an ablation study and to show the effectiveness of the designs in +FedAlign. +4.1 +Datasets +We conduct experiments on 6 datasets covering 4 different applica- +tion scenarios and both single-label and multi-label classification +problems. Table 1 offers an overview and the details are as follows. +(1) Behavioral Context Recognition. The task is to infer the +context of human activity, e.g., what the person is doing, where +and with whom the person is. ExtraSensory [40] is a bench- +mark dataset for this application scenario. It contains 51 classes +in total which can be partitioned into 5 main categories: posture, +location, activity, and companion. Based on ExtraSensory, we +construct 3 datasets with non-overlapping client class sets as +follows. +• ES-5. In this dataset, we have 5 clients and every client only +has annotations from a different main category (i.e., one main +category to one client). Training samples are then assigned to +clients according to their associated classes. Since ExtraSen- +sory is a multi-label dataset, the same training sample may +have multiple classes from different main categories. In this +case, we assign the training sample based on the most infre- +quent class among these multiple labels to ensure that each +locally-identified class will have at least one positive sample. +To make this dataset more realistic, we always assign all the +data of each human subject to the same client. +• ES-15 and ES-25. We also increase the number of clients to +15 and 25 to further challenge the compared methods. We +start with the 5 class groups as ES-5 and iteratively split the +class groups until the number of class groups is the same as +the number of clients. For every split, we select the group +with the most number of classes and randomly separate it +into two sub-groups. Every class group is visible and only +visible to one client. One can then apply a similar process as +ES-5 to assign training samples to different clients. +(2) Medical Code Prediction. Medical codes describe whether +a patient has a specific medical condition or is at risk of de- +velopment. The task aims at automatically annotating medical +codes from clinical notes that record information about what +happened during a patient’s hospitalization. We start with the +MIMIC-III database [7] and follow the preprocessing proce- +dure in [32] to form the benchmark MIMIC-III 50-label dataset +that contains the 50 most frequent labels. These classes can be +Table 1: Our datasets cover application scenarios of behav- +ioral context recognition, clinical phenotype classification, +human activity recognition, and text classification. The im- +balance factor of a dataset refers to the ratio of its smallest +class size to the largest class size. +Dataset +|C| +# of clients +Avg. 𝑁𝑚 +Avg. |C𝑚| +Imbalance +Remarks +ES-5 +51 +5 +5,446 +10.2 +0.0013 +Multi-label, +non- +overlapping +client class +sets +ES-15 +15 +1,769 +3.4 +ES-25 +25 +1,073 +2.04 +MIMIC-III-10 +50 +10 +807 +5 +0.1157 +PAMAP2-9 +18 +9 +1,287 +5 +0.2049 +Single-label +R8-8 +8 +8 +617 +3 +0.0130 +grouped into 10 categories according to the ICD-9 taxonomy1. +We construct the MIMIC-III-10 dataset by partitioning the +dataset into 10 clients following the same strategy as that in +ES-5. +(3) Human Activity Recognition. The task aims at identifying +the movement or action of a person based on sensor data. We +start with the PAMAP2 [38] dataset, which contains data of +18 different physical activities collected from 9 subjects. We +construct the PAMAP2-9 dataset by using the data from each +subject as a client, which results in 9 clients. For each client, we +randomly select 5 classes to be its locally-identified classes. +(4) Text Classification. The task is to categorize text into orga- +nized groups. We use the Reuters-21578 R8 dataset [2], which is +a collection of news articles. We construct the R8-8 dataset by +setting the number of clients as 8, randomly dividing the data +into 8 partitions, and assigning one partition to each client. For +each client, we randomly select 3 classes to be the identified +classes. +More details about data preprocessing are described in Appendix. +4.2 +Compared Methods +We compare FedAlign with various classical [30] and state-of-the- +art federated learning methods [9, 20, 21] designed for non-IID data +problems as follows. +• FedAvg [30] is a classical federated learning method. The weights +of local models uploaded by clients in each round are averaged to +get the global model. Then, the server sends the updated model +to each client. +• FedProx [21] deals with the heterogeneity in the client models +by enforcing a 𝐿2 regularization term in local optimization to +limit the distance between local model and global model. +• MOON [20] adds a contrastive loss term to maximize the consis- +tency of representations learned by the local model and that by +the global model and minimize the consistency between repre- +sentations learned by the local models of consecutive rounds. +• Scaffold [9] addresses the client-variance problem by maintain- +ing control variates to estimate the update directions of the global +model and the client model. The drift of local training is approx- +imated by the difference between the local and global update +directions and added to the local updates to mitigate drift. +1the International Statistical Classification of Diseases and Related Health Problems +(ICD): ftp://ftp.cdc.gov/pub/Health_Statistics/NCHS/Publications/ICD-9/ucod.txt +5 + +Table 2: Main experimental results (%) on the six datasets. ExtraSensory (ES-5, ES-15, ES-25) and MIMIC-III are multi-label +datasets where class sets of different clients are non-overlapping. PAMAP2 and R8 are single-label datasets where the class +sets of different clients overlap. The results are averaged over 5 runs. +Method +ES-5 +ES-15 +ES-25 +MIMIC-III-10 +PAMAP2-9 +R8-8 +F1 +Acc +F1 +Acc +F1 +Acc +F1 +Acc +F1 +Acc +F1 +Acc +FedAvg [30] +28.77 +80.79 +22.71 +62.44 +19.52 +50.42 +35.04 +67.07 +68.89 +71.45 +78.51 +92.76 +FedProx (𝜇 = 0.001) [21] +29.26 +80.67 +22.42 +61.91 +19.48 +51.16 +34.75 +65.98 +69.70 +73.63 +79.05 +92.61 +FedProx (𝜇 = 0.0001) [21] +28.53 +79.14 +22.52 +61.95 +19.05 +52.33 +33.73 +64.31 +69.38 +71.78 +75.98 +91.78 +MOON [20] +29.12 +81.00 +22.84 +62.53 +19.52 +50.24 +34.62 +66.34 +71.70 +74.25 +79.26 +93.07 +Scaffold [9] +28.14 +77.13 +23.15 +61.69 +19.73 +48.81 +33.58 +62.84 +73.57 +75.60 +82.83 +94.43 +FedRS (𝛼 = 0.5) [22] +28.01 +78.72 +22.50 +62.09 +19.44 +51.41 +34.82 +66.80 +68.70 +71.42 +80.10 +92.74 +FedRS (𝛼 = 0.9) [22] +28.25 +79.25 +22.55 +62.17 +19.40 +50.87 +35.44 +67.45 +71.81 +74.44 +76.68 +91.81 +FedPU [26] +- +- +- +- +- +- +- +- +85.39 +87.59 +83.17 +94.23 +FedAlign +29.69 +83.92 +23.36 +73.61 +20.78 +62.11 +37.87 +71.13 +87.21 +88.14 +83.76 +94.92 +10 +20 +30 +40 +50 +0.22 +0.24 +0.26 +0.28 +0.3 +0.32 +F1 +ES-5 +10 +20 +30 +40 +50 +0.14 +0.16 +0.18 +0.2 +0.22 +0.24 +ES-15 +10 +20 +30 +40 +50 +0.12 +0.14 +0.16 +0.18 +0.2 +0.22 +ES-25 +20 +40 +60 +80 +100 +0.1 +0.2 +0.3 +0.4 +MIMIC-III-10 +10 +20 +30 +40 +50 +0 +0.25 +0.5 +0.75 +1 +R8-8 +10 +20 +30 +40 +50 +0.6 +0.7 +0.8 +Accuracy +10 +20 +30 +40 +50 +0.5 +0.6 +0.7 +0.8 +10 +20 +30 +40 +50 +0.3 +0.4 +0.5 +0.6 +0.7 +0.8 +20 +40 +60 +80 +100 +0.5 +0.6 +0.7 +10 +20 +30 +40 +50 +0.25 +0.5 +0.75 +1 +FedAlign +FedAvg +MOON +FedProx +Scaffold +FedPU +FedRS +10 +20 +30 +40 +50 +0.5 +0.75 +1 +10 +20 +30 +40 +50 +0.25 +0.5 +0.75 +1 +PAMAP2-9 +Figure 3: Performance w.r.t. Communication Rounds on Six Datasets. The results are averaged over 5 runs. +We also compare two state-of-the-art federated learning meth- +ods [22, 26] designed for classification problems with missing classes. +Specifically, +• FedRS [22] is designed for federated learning with missing classes +where each client only owns data of part of the classes (i.e., locally- +identified classes in our terminology). It restricts the weight up- +date of the missing classes in the classifier by adding scaling +factors to the softmax operation. +• FedPU [26] is designed for the scenario where each client only +labels a small part of their dataset and there exists unlabeled data +from both locally-identified classes and locally-unaware classes. +It utilizes the labeled data at each client to estimate the misclassi- +fication loss between the unaware classes of the other clients and +adds the estimated loss to the local optimization objective. As the +calculation of the expected risk is based on single-label classifica- +tion, where each sample is associated with only one of the classes, +it does not hold true in multi-label problems. While it is difficult +to generalize FedPU for multi-label classification problems, we +compare it with FedAlign in the single-label applications. +4.3 +Experiment Setup +Base Neural Network Model. For a fair comparison, we use the +same network setting for all compared methods. The data encoder +is a Transformer Encoder [41]. We use one encoder layer in the +data encoder, where the number of heads in multi-head attention +is 4 and the dimension of the feed-forward network is 64. The label +encoder is a single hidden layer neural network. The dimension of +data and class representations 𝑑 is 256. +Evaluation Metrics. Due to label imbalance, we adopt accuracy +and F1-score to evaluate the performance. They are often used as +benchmark metrics for the datasets and tasks in our experiments [32, +35, 38, 40]. We calculate the metrics for each class and report the +macro-average. +Train/Test Split. For MIMIC-III and R8, we use the data split pro- +vided by the dataset. For the other datasets, we use the data from +80% of the data for training and 20% of the data for testing. +Learning Setting. For ExtraSensory (ES-5, ES-15, ES-25), PAMAP2- +9, R8-8, we run 𝑇 = 50 communication rounds. For MIMIC-III-10, +we run 𝑇 = 100 rounds as it takes longer to converge. The number +of selected clients per round is |𝑆𝑡 | = 5 and the local epochs 𝐸 = 5. +Optimizer and Hyperparameters. For compared methods, we +try different values for hyperparameters 𝜇 (in FedProx and MOON) +6 + +Table 3: Results (% Averaged Over 5 Runs) of Ablation Study +Method +PAMAP2-9 +R8-8 +F1 +Acc +F1 +Acc +FedAvg +68.89 +71.45 +77.27 +90.59 +FedAlign w/o semantic +83.39 +85.40 +82.37 +94.13 +FedAlign w/o distillation +70.87 +74.59 +79.67 +91.06 +FedAlign w/o alternation +86.01 +87.50 +83.22 +94.63 +FedAlign +87.21 +88.14 +83.76 +94.92 +and 𝛼 (in FedRS) that are often adopted in the previous papers [20– +22]. For MOON, we set 𝜇 = 0.001 for all datasets. For FedProx, we +report the performance with 𝜇 = 0.001 and 𝜇 = 10−5. For FedRS, +we report the performance with 𝛼 = 0.5 and 𝛼 = 0.9. +4.4 +Main Results and Analysis +Single-Label, Non-identical but Overlapping Client Class Sets. +Table 2 shows the results. For both PAMAP2 and R8 datasets, +FedAlign performs better than the baseline methods. The non-IID +problems that FedRS and FedPU aim to tackle are slightly different +from ours. Although they show improvements over the classical +algorithm FedAvg and the methods designed for the typical non- +IID setting (FedProx, MOON, and Scaffold), FedAlign shows better +performance compared with FedRS and FedPU in the problem of +non-identical client class sets. +Multi-Label, Non-overlapping Client Class Sets. As one can +clearly see, FedAlign always yields better performance than the +baseline methods. Remarkably, with non-identical client class sets, +the three state-of-the-art algorithms designed to deal with non-IID +data (i.e., FedProx, MOON, and Scaffold) do not guarantee improve- +ment over FedAvg (e.g., Scaffold loses to FedAvg on ES-5). FedRS +does not show improvement over its base algorithm FedAvg. Al- +though it is designed for federated learning with missing classes, +its mechanism is specifically based on single-label (softmax) classi- +fication. In multi-label (sigmoid) classification, the weight update +of class 𝑐 is only affected by features from class 𝑐. Adding scaling +factors to missing classes does not affect the weight update of the +identified classes, thus, does not solve the non-IID problem. +For the ExtraSensory dataset, as the number of partitions in- +creases (5, 15, 25), the performance of the methods declines. This is +due to two reasons: first, the label sets are more widely distributed +so the training samples with identified classes on each client be- +come fewer; second, the label space of different clients becomes +more diverse, making it difficult to learn and aggregate information. +However, we observe a growing trend in the improvement of test- +ing accuracy when increasing the number of partitions. With more +widely distributed classes and fewer training samples, FedAlign +show greater advantages. +Performance vs. Communication Rounds. In Figure 3, we show +the test performance with respect to the number of communication +rounds. We see that FedAlign outperforms the baselines since a +few rounds of communication. There is a clear performance gap +between the baseline methods and FedAlign. +10 +20 +30 +40 +50 +0.2 +0.4 +0.6 +0.8 +1 +F1 +PAMAP2-9 +10 +20 +30 +40 +50 +0.4 +0.6 +0.8 +1 +Accuracy +10 +20 +30 +40 +50 +0 +0.2 +0.4 +0.6 +0.8 +1 +F1 +R8-8 +10 +20 +30 +40 +50 +0.2 +0.4 +0.6 +0.8 +1 +Accuracy +FedAlign +FedAlign w/o distillation +FedAlign w/o semantic +FedAvg +FedAlign w/o alternation +Figure 4: Performance (averaged over 5 runs) w.r.t. commu- +nication rounds in ablation study. +4.5 +Ablation Study +We conduct an ablation study to evaluate the contribution of each +design in FedAlign on the PAMAP and R8 datasets. First, we evalu- +ate the performance of the method without knowledge distillation +(denoted as FedAlign w/o distillation). The classification model +consists of a label encoder and a label encoder and the framework +conducts alternating training of the two modules. Second, we eval- +uate the performance of the method without the semantic label +name sharing (denoted as FedAlign w/o semantic). The distil- +lation dataset is formed by annotating the samples according to +their prediction confidence given by the latest global model. Highly +confident predictions are used as pseudo-labels and the confidence +threshold is decided by the same percentile value as in FedAlign. +Third, We evaluate the performance of the method without alter- +nating training (denoted as FedAlign w/o alternation) which +updates the label encoder and data encoder simultaneously. Note +that the model aggregation method in FedAlign is based on Fe- +dAvg (i.e., averaging the model parameters), thus we also compare +FedAvg as the baseline method. +Table 3 shows the performance at the end of 50 communication +rounds and Figure 4 shows the performance with respect to the num- +ber of communication rounds. We notice the performance drops +when removing any of the designs. The methods with complete +features can work better than using only part of the designs. This +shows the key designs (semantic label name sharing, knowledge +distillation, and alternating training) in FedAlign all contribute to +performance improvement, and combining them can produce the +best performance. +5 +RELATED WORK +We first review the important topic related to the problem we +discuss, that is, federated learning with non-IID data. Then, we +review two lines of related work that motivate our designs in the +framework: label semantics modeling and knowledge distillation +7 + +in federated learning. +5.1 +Federated Learning with non-IID data +One of the fundamental challenges in federated learning is the +presence of non-IID data [8]. The reasons and solutions to this +challenge are being actively explored. One common solution is to +mitigate client variance during local training by adding regulariza- +tion [20, 21], or estimating global update directions and removing +drift [9]. Another type of solution improves model aggregation at +the server [25, 43, 44], such as assigning aggregation weights accord- +ing to local optimization steps [44] or doing layer-wise aggregation +by matching hidden elements with similar feature extraction signa- +tures [43]. Other works propose solutions from the data perspective. +They leverage public datasets to conduct distillation for model fu- +sion [25] or calibrate classifiers using synthesized features [28, 50]. +Personalization is another type of solution [16, 39], which enables +different clients to have different model parameters with regard +to their own data distribution. These methods tackle more relaxed +non-IID problems that assume clients have the same set of classes. +We study the problem of non-identical class sets in federated +learning, which is a more general case of non-IID label distribution +skew. Some recent works [22, 26] consider the problem that clients +can only access part of the whole class set. FedRS [22] deals with +the case where each client only owns the data of certain classes. +It restricted the weight update of the missing classes during the +local procedure by adding a scaling factor in the softmax function. +FedPU [26] focuses on a scenario where each client only labels +a small part of their dataset and there exists unlabeled data from +both positive classes (i.e., identified classes in ours) and negative +(i.e., locally-unaware in ours) classes. It utilizes the labeled data +at each client to estimate the misclassification loss between the +negative classes of the other clients and adds the estimated loss +to local optimization objective. The problem settings in the two +related works are slightly different from ours and applying them to +our problem shows less improvement compared to our proposed +method. +5.2 +Label Semantics Modeling +For tasks where some of the label co-occurrence patterns can not +be directly observed from the training dataset, such as zero-shot +learning [11], it is hard for the model to understand label corre- +lations. To deal with the problem, several methods are proposed +to leverage prior knowledge such as knowledge graphs [42] or +model semantic label embedding from textual information about +classes [13, 29, 36, 45, 48]. For example, Ba et al. [13] derived em- +bedding features for classes from natural language descriptions +and learned a mapping to transform the text features of classes to +the visual image feature space. Radford et al [36] used contrastive +pre-training to jointly train an image encoder and a text encoder +and predict the correct pairings of image and text caption, which +helps to produce high-quality image representations. Matsuki et +al [29] and Wu et al [45] incorporate word embeddings for zero-shot +learning in human activity recognition. These methods show the +potential of allowing models to use semantic relationships between +labels to make predictions for classes not observed in the training +set, which motivates our design of semantic label name sharing. +5.3 +Knowledge Distillation in Federated +Learning +Knowledge distillation [5] is originally used for transferring knowl- +edge from a large model or ensemble of models to a single smaller +model. The large pre-trained models are regarded as the teachers +and the smaller model (i.e., student) mimics the outputs or inter- +mediate features of the teachers. In federated learning, knowledge +distillation has been used to deal with statistical heterogeneity [50] +and system heterogeneity (i.e., different client models) [18, 25]. +A public dataset is usually required, or otherwise, a generator is +trained to get synthetic data [50]. Recently, semi-supervised meth- +ods are also applied in federated learning for cases where labeled +training data is limited [6, 24]. Our work is different from the pre- +vious works in that the non-identical class sets setting we focus on +has no labeled data for the unaware classes at each client. We use +the class representations to annotate data samples and add regular- +ization in a knowledge distillation manner to transfer knowledge +about unaware classes to client models. +6 +CONCLUSIONS AND FUTURE WORK +We studied the problem of federated classification with non-identical +class sets. We propose the FedAlign framework and demonstrated +its use in federated learning for various applications. FedAlign +breaks the classification model into a data encoder and a label en- +coder. Semantic label learning is conducted by leveraging a domain- +related corpus and shared label names. The pre-trained semantic +label embeddings contain the knowledge of label correlations and +are then used to guide the training of data encoder. Moreover, +the knowledge distillation strategy complements the training of +unaware classes at each client. These two modules are a key to mit- +igating client variance in FedAlign, which addresses the challenge +of non-identical class sets. We show that FedAlign improves the +baseline algorithms for federated learning with non-IID data and +achieves new state-of-the-art. +For future directions, we consider more general system hetero- +geneity where the participants have different network architectures, +different training processes, and different tasks at the same time. +We plan to extend our study to make federated learning compatible +with such heterogeneity. +REFERENCES +[1] John A Bullinaria and Joseph P Levy. 2007. Extracting semantic representations +from word co-occurrence statistics: A computational study. Behavior research +methods 39, 3 (2007), 510–526. +[2] Ana Cardoso-Cachopo. 2007. Improving Methods for Single-label Text Cate- +gorization. PdD Thesis, Instituto Superior Tecnico, Universidade Tecnica de +Lisboa. +[3] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: +Pre-training of deep bidirectional transformers for language understanding. arXiv +preprint arXiv:1810.04805 (2018). +[4] Felix Hamborg, Norman Meuschke, Corinna Breitinger, and Bela Gipp. 2017. +news-please: A Generic News Crawler and Extractor. In Proceedings of the 15th +International Symposium of Information Science (Berlin). 218–223. https://doi. +org/10.5281/zenodo.4120316 +[5] Geoffrey Hinton, Oriol Vinyals, Jeff Dean, et al. 2015. Distilling the knowledge +in a neural network. arXiv preprint arXiv:1503.02531 2, 7 (2015). +[6] Wonyong Jeong, Jaehong Yoon, Eunho Yang, and Sung Ju Hwang. 2020. Federated +semi-supervised learning with inter-client consistency & disjoint learning. arXiv +preprint arXiv:2006.12097 (2020). +[7] Alistair EW Johnson, Tom J Pollard, Lu Shen, Li-wei H Lehman, Mengling Feng, +Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and +8 + +Roger G Mark. 2016. MIMIC-III, a freely accessible critical care database. Scientific +data 3, 1 (2016), 1–9. +[8] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Ben- +nis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, +Rachel Cummings, et al. 2021. Advances and open problems in federated learning. +Foundations and Trends® in Machine Learning 14, 1–2 (2021), 1–210. +[9] Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebas- +tian Stich, and Ananda Theertha Suresh. 2020. Scaffold: Stochastic controlled +averaging for federated learning. In International Conference on Machine Learning. +PMLR, 5132–5143. +[10] Stefanos Laskaridis, Dimitris Spathis, and Mario Almeida. 2021. Federated mobile +sensing for activity recognition. In Proceedings of the 27th Annual International +Conference on Mobile Computing and Networking. 858–859. +[11] Chung-Wei Lee, Wei Fang, Chih-Kuan Yeh, and Yu-Chiang Frank Wang. 2018. +Multi-label zero-shot learning with structured knowledge graphs. In Proceedings +of the IEEE conference on computer vision and pattern recognition. 1576–1585. +[12] Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, +Chan Ho So, and Jaewoo Kang. 2020. BioBERT: a pre-trained biomedical language +representation model for biomedical text mining. Bioinformatics 36, 4 (2020), +1234–1240. +[13] Jimmy Lei Ba, Kevin Swersky, Sanja Fidler, et al. 2015. Predicting deep zero-shot +convolutional neural networks using textual descriptions. In Proceedings of the +IEEE international conference on computer vision. 4247–4255. +[14] Omer Levy and Yoav Goldberg. 2014. Linguistic regularities in sparse and explicit +word representations. In Proceedings of the eighteenth conference on computational +natural language learning. 171–180. +[15] Ang Li, Jingwei Sun, Pengcheng Li, Yu Pu, Hai Li, and Yiran Chen. 2021. Hermes: +an efficient federated learning framework for heterogeneous mobile clients. In +Proceedings of the 27th Annual International Conference on Mobile Computing and +Networking. 420–437. +[16] Ang Li, Jingwei Sun, Xiao Zeng, Mi Zhang, Hai Li, and Yiran Chen. 2021. Fed- +Mask: Joint Computation and Communication-Efficient Personalized Federated +Learning via Heterogeneous Masking. In Proceedings of the 19th ACM Conference +on Embedded Networked Sensor Systems. 42–55. +[17] Chenglin Li, Di Niu, Bei Jiang, Xiao Zuo, and Jianming Yang. 2021. Meta-har: +Federated representation learning for human activity recognition. In Proceedings +of the Web Conference 2021. 912–922. +[18] Daliang Li and Junpu Wang. 2019. Fedmd: Heterogenous federated learning via +model distillation. arXiv preprint arXiv:1910.03581 (2019). +[19] Liang Li, Dian Shi, Ronghui Hou, Hui Li, Miao Pan, and Zhu Han. 2021. To talk +or to work: Flexible communication compression for energy efficient federated +learning over heterogeneous mobile edge devices. In IEEE INFOCOM 2021-IEEE +Conference on Computer Communications. IEEE, 1–10. +[20] Qinbin Li, Bingsheng He, and Dawn Song. 2021. Model-contrastive federated +learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and +Pattern Recognition. 10713–10722. +[21] Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, +and Virginia Smith. 2020. Federated optimization in heterogeneous networks. +Proceedings of Machine Learning and Systems 2 (2020), 429–450. +[22] Xin-Chun Li and De-Chuan Zhan. 2021. Fedrs: Federated learning with restricted +softmax for label distribution non-iid data. In Proceedings of the 27th ACM SIGKDD +Conference on Knowledge Discovery & Data Mining. 995–1005. +[23] Feng Liang, Weike Pan, and Zhong Ming. 2021. Fedrec++: Lossless federated +recommendation with explicit feedback. In Proceedings of the AAAI conference on +artificial intelligence, Vol. 35. 4224–4231. +[24] Haowen Lin, Jian Lou, Li Xiong, and Cyrus Shahabi. 2021. +Semifed: Semi- +supervised federated learning with consistency and pseudo-labeling. arXiv +preprint arXiv:2108.09412 (2021). +[25] Tao Lin, Lingjing Kong, Sebastian U Stich, and Martin Jaggi. 2020. Ensemble +distillation for robust model fusion in federated learning. Advances in Neural +Information Processing Systems 33 (2020), 2351–2363. +[26] Xinyang Lin, Hanting Chen, Yixing Xu, Chao Xu, Xiaolin Gui, Yiping Deng, and +Yunhe Wang. 2022. Federated Learning with Positive and Unlabeled Data. In +International Conference on Machine Learning. PMLR, 13344–13355. +[27] Shuchang Liu, Shuyuan Xu, Wenhui Yu, Zuohui Fu, Yongfeng Zhang, and Amelie +Marian. 2021. FedCT: Federated collaborative transfer for recommendation. +In Proceedings of the 44th international ACM SIGIR conference on research and +development in information retrieval. 716–725. +[28] Mi Luo, Fei Chen, Dapeng Hu, Yifan Zhang, Jian Liang, and Jiashi Feng. 2021. No +fear of heterogeneity: Classifier calibration for federated learning with non-iid +data. Advances in Neural Information Processing Systems 34 (2021). +[29] Moe Matsuki, Paula Lago, and Sozo Inoue. 2019. Characterizing word embeddings +for zero-shot sensor-based human activity recognition. Sensors 19, 22 (2019), +5043. +[30] Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and +Blaise Aguera y Arcas. 2017. Communication-efficient learning of deep net- +works from decentralized data. In Artificial intelligence and statistics. PMLR, +1273–1282. +[31] Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. +Distributed representations of words and phrases and their compositionality. +Advances in neural information processing systems 26 (2013). +[32] James Mullenbach, Sarah Wiegreffe, Jon Duke, Jimeng Sun, and Jacob Eisenstein. +2018. Explainable Prediction of Medical Codes from Clinical Text. In Proceedings +of the 2018 Conference of the North American Chapter of the Association for Com- +putational Linguistics: Human Language Technologies, Volume 1 (Long Papers). +1101–1111. +[33] Bethesda (MD): National Library of Medicine. 2003. PMC Open Access Subset +[Internet]. https://www.ncbi.nlm.nih.gov/pmc/tools/openftlist/ [accessed 7-June- +2022]. +[34] Jeffrey Pennington, Richard Socher, and Christopher D Manning. 2014. Glove: +Global vectors for word representation. In Proceedings of the 2014 conference on +empirical methods in natural language processing (EMNLP). 1532–1543. +[35] Yinhua Piao, Sangseon Lee, Dohoon Lee, and Sun Kim. 2022. Sparse Structure +Learning via Graph Neural Networks for Inductive Document Classification. In +Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 36. 11165–11173. +[36] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, +Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, +et al. 2021. Learning transferable visual models from natural language supervision. +In International Conference on Machine Learning. PMLR, 8748–8763. +[37] Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. 2018. Im- +proving language understanding by generative pre-training. (2018). +[38] Attila Reiss and Didier Stricker. 2012. Introducing a new benchmarked dataset for +activity monitoring. In 2012 16th international symposium on wearable computers. +IEEE, 108–109. +[39] Linlin Tu, Xiaomin Ouyang, Jiayu Zhou, Yuze He, and Guoliang Xing. 2021. +FedDL: Federated Learning via Dynamic Layer Sharing for Human Activity +Recognition. In Proceedings of the 19th ACM Conference on Embedded Networked +Sensor Systems. 15–28. +[40] Yonatan Vaizman, Katherine Ellis, and Gert Lanckriet. 2017. Recognizing detailed +human context in the wild from smartphones and smartwatches. IEEE pervasive +computing 16, 4 (2017), 62–74. +[41] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, +Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all +you need. Advances in neural information processing systems 30 (2017). +[42] Laura Von Rueden, Sebastian Mayer, Katharina Beckh, Bogdan Georgiev, Sven +Giesselbach, Raoul Heese, Birgit Kirsch, Julius Pfrommer, Annika Pick, Rajkumar +Ramamurthy, et al. 2019. Informed Machine Learning–A Taxonomy and Survey +of Integrating Knowledge into Learning Systems. arXiv preprint arXiv:1903.12394 +(2019). +[43] Hongyi Wang, Mikhail Yurochkin, Yuekai Sun, Dimitris Papailiopoulos, and +Yasaman Khazaeni. 2020. Federated learning with matched averaging. arXiv +preprint arXiv:2002.06440 (2020). +[44] Jianyu Wang, Qinghua Liu, Hao Liang, Gauri Joshi, and H Vincent Poor. 2020. +Tackling the objective inconsistency problem in heterogeneous federated opti- +mization. Advances in neural information processing systems 33 (2020), 7611–7623. +[45] Tong Wu, Yiqiang Chen, Yang Gu, Jiwei Wang, Siyu Zhang, and Zhanghu +Zhechen. 2020. Multi-layer cross loss model for zero-shot human activity recogni- +tion. In Pacific-Asia Conference on Knowledge Discovery and Data Mining. 210–221. +[46] Chengxu Yang, Qipeng Wang, Mengwei Xu, Zhenpeng Chen, Kaigui Bian, Yunxin +Liu, and Xuanzhe Liu. 2021. Characterizing impacts of heterogeneity in federated +learning upon large-scale smartphone data. In Proceedings of the Web Conference +2021. 935–946. +[47] Jingwei Yi, Fangzhao Wu, Chuhan Wu, Ruixuan Liu, Guangzhong Sun, and +Xing Xie. 2021. Efficient-FedRec: Efficient Federated Learning Framework for +Privacy-Preserving News Recommendation. In Proceedings of the 2021 Conference +on Empirical Methods in Natural Language Processing. 2814–2824. +[48] Fengtao Zhou, Sheng Huang, and Yun Xing. 2021. Deep semantic dictionary +learning for multi-label image classification. In Proceedings of the AAAI Conference +on Artificial Intelligence, Vol. 35. 3572–3580. +[49] Yukun Zhu, Ryan Kiros, Rich Zemel, Ruslan Salakhutdinov, Raquel Urtasun, +Antonio Torralba, and Sanja Fidler. 2015. Aligning books and movies: Towards +story-like visual explanations by watching movies and reading books. In Proceed- +ings of the IEEE international conference on computer vision. 19–27. +[50] Zhuangdi Zhu, Junyuan Hong, and Jiayu Zhou. 2021. Data-free knowledge +distillation for heterogeneous federated learning. In International Conference on +Machine Learning. PMLR, 12878–12889. +9 + +A +APPENDIX +A.1 +Semantic Label Embedding Pre-training. +Before collaborative training, each client shares the natural lan- +guage names of its locally-identified classes with the server. The +server then searches for the label names in a large text corpus re- +lated to the domain of the classification task. We count the times +that each label name 𝑐𝑖 appears in each text segment (e.g., sentence +or paragraph). The label names can be phrases that contain mul- +tiple words and the order of the words may change in different +text segments while representing the same meaning, such as “colon +cancer” and “cancer of the colon”. To match phrases, we arrange the +label names into sets of words and apply a sliding window of length +𝐿𝑤 on each text segment. If the set of words in the label name is +covered by the words in the sliding window, we mark that the label +name appears in the text segment. The length of the sliding window +𝐿𝑤 is varied per the label name to search. The co-occurrence of a +pair of label names 𝑐𝑖 and 𝑐𝑗 is calculated as the point-wise mutual +information (PMI): +PMI(𝑐𝑖,𝑐𝑗) = log 𝑝(𝑐𝑖,𝑐𝑗) +𝑝(𝑐𝑖)𝑝(𝑐𝑗) , +where 𝑝(𝑐𝑖) and 𝑝(𝑐𝑗) are the individual distributions and 𝑝(𝑐𝑖,𝑐𝑗) +is the the joint distribution. The higher the PMI(𝑐𝑖,𝑐𝑗), the stronger +the association between the two label names 𝑐𝑖 and 𝑐𝑗. The server +learns semantic label embeddings based on the co-occurrence. The +principle of label embedding learning is to learn a mapping function +from labels to representations 𝑓 : C → R𝑑, which enforces labels +with similar semantics to have similar representations. To achieve +this, the server builds a label co-occurrence graph 𝐺 = ⟨𝑉, 𝐸⟩, +where the nodes 𝑉 represent the class label names and the edges 𝐸 +represent the co-occurrence relationship between the nodes. The +PMI values are zero-centered by subtracting the mean and are used +as the edge weights between label names. Edges with negative +weights are removed from the graph. For every source node 𝑐𝑜 ∈ 𝑉 , +we define 𝑁𝑠 (𝑐𝑜) ⊂ 𝑉 as its network neighborhood generated +through simulating fixed-length random walks starting from 𝑐𝑜. +The transition probability from node 𝑢 to node 𝑣 for random walk +simulation is calculated by normalizing the edge weight 𝜋𝑢,𝑣 = +𝑤𝑢,𝑣/� +𝑣′∈𝑉 𝑤𝑢,𝑣′. The objective function of label embedding is: +max +𝑓 +∑︁ +𝑐𝑜 ∈C +(− log𝑍𝑐𝑜 + +∑︁ +𝑐𝑖 ∈𝑁𝑠 (𝑐𝑜) +𝑓 (𝑐𝑖) · 𝑓 (𝑐𝑜)), +where 𝑍𝑐𝑜 = � +𝑣∈𝑉 exp(𝑓 (𝑣) · 𝑓 (𝑐𝑜)) and is approximated using +negative sampling [31]. The mapping function 𝑓 is achieved by a +single hidden layer feedforward neural network and the objective +is optimized using stochastic gradient descent. +A.2 +Text Corpus +Domain-related raw corpus is often easily accessible. For example, +novel books that describe a variety of daily scenes can be used +for understanding human activities. Academic journals are good +studying resources for understanding concepts in different domains. +We use the following corpora for applications in the experiments: +BookCorpus. [49] is a large collection of free novel books col- +lected from the web. It is a popular text corpus in the domain of +Natural Language Processing and has helped the training of many +influential language models such as BERT [3] and GPT [37]. +PubMed Open-Access (OA) subset. [33] is a text archive of jour- +nal articles and preprints in biomedical and life sciences. It has been +widely used for biomedical text mining [12]. +CommonCrawl (CC) News dataset. [4] is an English-language +news corpus collecting news articles published between 2017 to +2019 from news sites worldwide. +We use PubMed-OA as the text corpus for clinic phenotype +classification and BookCorpus for behavioral context recognition +and human activity recognition, and CC News for text classification. +We regard a sentence as a text segment in BookCorpus, an article +as a text segment in CC News, and a paragraph as a text segment +in PubMed-OA. +A.3 +Details About Datasets and Preprocessing +ExtraSensory dataset. [40]: collects time-series sensory measure- +ments of 60 participants in their natural behavior, using smart- +phones and smartwatches. Relevant context labels of the time-series +measurements are annotated. There are in total 51 classes, which +can be grouped into five main categories: posture (sitting, lying +down, etc.), location (gym, beach, etc.), activity (eating, sleeping, +etc.), companion (with friends, with co-workers). Each sample is +associated with 3.6 classes on average. During data preprocessing, +the time-series data of a subject is partitioned into several pieces ac- +cording to its labels to ensure each piece of the resulting time-series +data is in the same behavioral context from the beginning to the +end. Each piece of data sample is a 10-minute time-series data. Data +sample of less than 10 minutes is padded with zeros. The features +at each timestamp represent the measurements taken within one +minute. The feature dimension is 225. +MIMIC-III dataset. [7]: contains inpatient data of over 40, 000 +patients. We follow the preprocessing instruction in [32] to derive +the benchmark MIMIC-III 50-label dataset. The data of each patient +is the clinical notes that record information about what happened +during a patient’s hospitalization. The vocabulary size of the clinical +notes is 4,896. The notes has an average length of 1,512 words. The +dataset contains 50 most frequent classes which can be grouped +into 10 categories according to ICD, such as circulatory systems, +digestive systems, etc. Each data sample is associated with 5.69 +classes on average. +PAMAP2. [38]: collects sensory measurements of 9 subjects per- +forming 18 different physical activities (e.g., sitting, ascending stairs, +etc.). Data are collected by inertial measurement units (IMU) sen- +sors and a heart rate monitor. The feature dimension is 52. We +organize the data into time series with a window size of 100. +Reuters-21578 R8. [2]: is a collection of news articles including 8 +classes (e.g. trade, grain, earn, etc.). For preprocessing, words with +no less than 100 occurrences are kept. The vocabulary size is 534. +Each piece of text data has an average length of 66 words. +10 + diff --git a/wtAyT4oBgHgl3EQfnfhR/content/tmp_files/load_file.txt b/wtAyT4oBgHgl3EQfnfhR/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..fa2d10ab88031003f1a0c9cd5d1e48a8f1affbc7 --- /dev/null +++ b/wtAyT4oBgHgl3EQfnfhR/content/tmp_files/load_file.txt @@ -0,0 +1,968 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf,len=967 +page_content='Federated Learning with Client-Exclusive Classes Jiayun Zhang1, Xiyuan Zhang1, Xinyang Zhang2, Dezhi Hong3†, Rajesh K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Gupta1, Jingbo Shang1 1University of California, San Diego, La Jolla, CA, USA 2University of Illinois at Urbana-Champaign, Champaign, IL, USA 3Amazon {jiz069,xiyuanzh,gupta,jshang}@ucsd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='edu,xz43@illinois.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='edu,hondezhi@amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='com ABSTRACT Existing federated classification algorithms typically assume the local annotations at every client cover the same set of classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In this paper, we aim to lift such an assumption and focus on a more gen- eral yet practical non-IID setting where every client can work on non-identical and even disjoint sets of classes (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', client-exclusive classes), and the clients have a common goal which is to build a global classification model to identify the union of these classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Such heterogeneity in client class sets poses a new challenge: how to ensure different clients are operating in the same latent space so as to avoid the drift after aggregation?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We observe that the classes can be described in natural languages (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', class names) and these names are typically safe to share with all parties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Thus, we formu- late the classification problem as a matching process between data representations and class representations and break the classifica- tion model into a data encoder and a label encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We leverage the natural-language class names as the common ground to anchor the class representations in the label encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In each iteration, the label encoder updates the class representations and regulates the data representations through matching.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We further use the updated class representations at each round to annotate data samples for locally-unaware classes according to similarity and distill knowl- edge to local models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Extensive experiments on four real-world datasets show that the proposed method can outperform various classical and state-of-the-art federated learning methods designed for learning with non-IID data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' KEYWORDS Federated Learning, non-IID, Label Name Sharing, Knowledge Dis- tillation 1 INTRODUCTION Federated learning [30] emerges as a distributed learning paradigm that allows multiple parties to collaboratively learn a global model effective for all parties while ensuring the privacy of local data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' This benefits applications in a wide range of domains, such as recommendation [23, 27, 47], ubiquitous sensing [10, 16, 17, 39] and mobile computing [15, 19, 46].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Existing federated classification methods, even those designed for non-IID clients [9, 20, 21, 25, 28, 43, 44, 50], typically assume that the local annotations at each client follow the same set of classes;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' however, this assumption does not hold true in many real- world applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For example, a smartwatch company wants to build a human activity classifier for all activity types, as shown in Figure 1(a).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Although their smartwatch users as clients could experience almost all types of daily activities, each user may only †Work unrelated to Amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: lying down, sitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping, cleaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Environment: home, gym.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking, sitting, lying down.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, lying down.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping, cleaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Environment: home, gym.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Learning Training client’s class set Testing client’s class set Inference Server (w/ global shared model) Posture: lying down, sitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping, cleaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Environment: home, gym.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking, sitting, lying down.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' identified classes Posture Activity Environment (full class set) Client Server unaware classes Posture: sitting Activity: Internet surfing Environment: restaurant Posture: standing, walking Client A Client C Server aggregation √ √ √ √ √ √ √ √ √ √ √ √ √ run sit sleep eat home gym Client local data sleep sit run eat home gym universal set of classes global model Server √ ground truth data data rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (identified classes) class rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Client A Client B Server aggregation data rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (unaware classes) (optimal) (classes) run sit gym home eat data Representations: Client A B C (a) The problem setting of non-identical client class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: lying down, sitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping, cleaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Environment: home, gym.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking, sitting, lying down.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, lying down.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping, cleaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Environment: home, gym.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Learning Training client’s class set Testing client’s class set Inference Server (w/ global shared model) Posture: lying down, sitting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Activity: sleeping, cleaning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Environment: home, gym.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Posture: standing, walking, sitting, lying down.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' identified classes Posture Activity Environment (full class set) Client Server unaware classes Posture: sitting Activity: Internet surfing Environment: restaurant Posture: standing, walking Client A Client C Server aggregation √ √ √ √ √ √ √ √ √ √ √ √ √ run sit sleep eat home gym Client local data sleep sit run eat home gym universal set of classes global model Server √ ground truth data data rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (identified classes) class rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Client A Client B Server aggregation data rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (unaware classes) (optimal) (classes) run sit gym home eat data Representations: Client A B C (b) Unique challenge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Figure 1: (a) illustrates our problem setting using a behav- ioral context recognition system where users have different preferences in reporting (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', annotating) labels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (b) demon- strates the unique challenge here — local models at different clients may operate in different and even independent latent spaces, making it hard to aggregate at the server.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' opt to report (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', annotate) a subset of activities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Another example is a federated medical diagnosis system, which attempts to infer all types of diseases of a patient for comprehensive health screen- ing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Physicians and specialist groups with different expertise can participate in this federated learning system as clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' As one can see here, different specialists will only offer disease annotations within their domains, even if a patient may have several types of diseases at the same time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' This makes the class sets at many clients non-identical and even non-overlapping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We aim to lift this assumption and work on a new and rather prac- tical federated learning setting, client-exclusive classes, where the local annotations at each client cover different and even non- overlapping sets of classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We denote the classes that are not covered in the local annotations as locally-unaware classes at each client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Each client can hold local data without any annotations, no matter whether their true labels are among the locally-unaware classes or not.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Also, the classification task here can be either single- label or multi-label.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' When it is multi-label, the local data might be only partially labeled due to the locally-unaware classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' There- fore, this new setting is more general and challenging than the missing class scenario [22], which is only applicable to single-label classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The non-identical client class sets pose a significant challenge of huge variance in local training across different clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' As shown in Figure 1(b), one can view classification as a matching process between data representations and label representations in a latent space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Because of the non-identical client class sets, locally trained classifiers are more likely to operate in drastically different latent spaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Moreover, when the class sets are non-overlapping, it is arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='00489v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='LG] 1 Jan 2023 山山possible that the latent spaces at different clients are completely in- dependent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' This would result in inaccurate classification boundaries after aggregation at the server, making our setting more challenging than non-IID clients with identical client class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We propose a novel federated learning framework FedAlign to better align the latent spaces across clients from both label and data perspectives as follows: (1) Anchor the label representations using label names.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We observe that the natural-language class names (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', label names) often carry valuable information for understanding label semantics, and more importantly, they are typically safe to share with all parties.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Therefore, we break the classification model into a data encoder and a label encoder as shown in Figure 2, and then lever- age the label names as the common ground for label encoders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The server initializes the label encoder with pre-trained text representations, such as word embedding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The label encoder will be then distributed to different clients and updated alter- natively with data encoders during local training and global aggregation, mutually regulating the latent space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (2) Connect the data representations via locally-unaware classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The representations given by the label encoder are representative of the classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Their distances to the data samples in the latent space can tell how likely a sample belongs to a class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Therefore, we introduce regularization in a knowledge distillation man- ner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Specifically, as shown in Figure 2, at each client, we first annotate local data based on their similarities with the repre- sentations of the locally-unaware classes, and then add another cross-entropy loss between the pseudo-annotations and the predictions by local models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Such regularization connects data encoders across different clients, so they have a better chance to be in the same latent space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We conduct experiments on four real-world applications, includ- ing the most challenging scenario of multi-label classification with non-overlapping client class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We show that FedAlign can out- perform various state-of-the-art federated learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Our contributions are as follows: We lift the common assumption of identical client class sets in federated learning and study a more general yet practical setting, non-identical client class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Such heterogeneity in client class sets poses a new challenge — local models at different clients may operate in different and even independent latent spaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We propose a novel framework FedAlign to better align the latent spaces across clients by anchoring the label representations using label names and connecting data representations via locally- unaware classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We conduct extensive experiments on four real-world datasets of different tasks and show that FedAlign can further improve various state-of-the-art federated learning algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2 PRELIMINARIES Problem Formulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We aim to generate a global classification model using federated learning with non-identical class sets, where each client only identifies part of the classes from its dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Given the universal set of classes as C, the set of classes that are identified on client 𝑚 is C𝑚, and the set of locally-unaware classes is C𝑚, where C𝑚 ∪ C𝑚 = C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The objective is to learn a global model 𝑔 : 𝑋 → C that given input feature 𝑋, all positive labels from the class set 𝐶 can be inferred.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The training set on client 𝑚 is denoted as D𝑚 = {(𝑥𝑖,𝑦𝑖)}𝑁 𝑖=1, where 𝑥𝑖 is the data feature and 𝑦𝑖 = (𝑦𝑖,𝑐1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ,𝑦𝑖,𝑐|C|) is a vector showing the labels of each class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' If𝑐𝑗 ∈ 𝐶𝑚,𝑦𝑖,𝑐 𝑗 ∈ {0, 1}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' If𝑐𝑗 ∈ C𝑚, 𝑦𝑖,𝑐 𝑗 is unknown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It is possible that some data samples 𝑥𝑖 ∈ 𝐷𝑚 do not belong to any of the classes in C𝑚, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', ∀𝑐 ∈ C𝑚 : 𝑦𝑖,𝑐 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Backbone Classification Model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Generally, the classification model can be decomposed into a data encoder and a classification layer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The data encoder takes the input data 𝑥𝑖 and generates a data representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, a linear layer (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', classifier) transforms the representation into prediction logits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We discuss two types of clas- sification tasks as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Single-label multi-class classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In this setting, each sam- ple is associated with only one positive class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In other words, the classes are mutually exclusive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use the softmax activation func- tion to get the prediction probability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Denote 𝑔𝑚(·) as the classifica- tion model at client 𝑚 and the prediction probability of 𝑥𝑖 on class 𝑐 as 𝑔(𝑥𝑖)𝑐.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The class with the maximum probabilities is predicted as positive, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', ˜𝑦𝑖 = 𝑎𝑟𝑔𝑚𝑎𝑥 𝑐 (𝑔𝑚(𝑥𝑖)𝑐).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The loss function at client 𝑚 is: LC𝑚 = − 1 𝑁𝑚 𝑁𝑚 ∑︁ 𝑖=1 ∑︁ 𝑐 ∈𝐶𝑚 𝑦𝑖,𝑐 log𝑔𝑚(𝑥𝑖)𝑐 Multi-label classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In this setting, each sample may be associated with a set of positive classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For example, a person may have both diabetes and hypertension.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The sigmoid function is applied to get the prediction probability, each element of which is the probability that the input data 𝑥𝑖 is associated with each class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The final predictions are achieved by thresholding the probabilities at 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' If 𝑔(𝑥𝑖)𝑐 > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5, sample 𝑥𝑖 is predicted to be in class 𝑐, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', ˜𝑦𝑖,𝑐 = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The learning objective is to minimize the binary cross- entropy loss over the identified classes: LC𝑚 = − 1 𝑁𝑚 𝑁𝑚 ∑︁ 𝑖=1 ∑︁ 𝑐 ∈C𝑚 [𝑦𝑖,𝑐 log𝑔𝑚(𝑥𝑖)𝑐+(1−𝑦𝑖,𝑐) log(1−𝑔𝑚(𝑥𝑖)𝑐)].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 3 THE FEDALIGN FRAMEWORK 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='1 Overview The pseudo code of FedAlign can be found in Algorithm 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Learning with FedAlign framework consists of the following steps: (1) Label name sharing and label encoder initialization: Be- fore training, the server collects the natural language label names from the clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The server initializes the label encoder 𝜃0 via pre-trained text representations, such as word embedding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We expect more advanced techniques like pre-trained neural language models could make the learning converge even faster, but we leave it as future work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The data encoder 𝜔0 is randomly initialized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (2) Client selection and model communication: At 𝑡-th round, the server randomly selects a subset of clients 𝑆𝑡 and sends the global model parameters to them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (3) Local training: Client 𝑚 ∈ 𝑆𝑡 independently trains its local model and returns the parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (4) Model aggregation: The server aggregates the parameters of client models into global parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2 … run, sit, stand… sleep, sing… gym, home… Server Server pretrain initialize round 0: initialization round #: training & communication Communication rounds Client $ … & KD loss CE loss !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='\' ( "\' !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=" features label names label encoder '* data encoder (* ." metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' logits label form distillation set train data encoder train label encoder CE loss !' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='\' ( "\' label aggregation data encoder (+ label encoder \'+ data encoder (+,- label encoder \'+,- data encoder (+ label encoder \'+ transfer parameters global model label name sharing class rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' data rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (pseudo-annotated) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' freeze freeze Figure 2: Overview of FedAlign framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The label names are leveraged as a common ground for label encoders to anchor class representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' During local training, the two encoders perform alternating training to mutually regulate the latent spaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The labels about locally-unaware classes are assigned to the local samples based on the similarities between the data representations and class representations and the knowledge is transferred to local models via knowledge distillation objec- tives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Algorithm 1: FedAlign Framework Input :Communication rounds 𝑇, number of selected clients each round |𝑆𝑡 |, local training epochs 𝐸.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Output:The final global model 𝑔(𝑥;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='𝜃𝑇,𝜔𝑇 ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 1 Server executes: 2 Collect label names from clients and pre-train text representations to initialize label encoder 𝜃0 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (1) 3 Randomly initialize data encoder 𝜔0;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4 for 𝑡 = 0, 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=',𝑇 − 1 do 5 Select a subset 𝑆𝑡 of clients at random ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (2) 6 for 𝑚 ∈ 𝑆𝑡 do 7 𝜃 (𝑡+1) 𝑚 ,𝜔 (𝑡+1) 𝑚 ← ClientUpdate(𝑚,𝜃 (𝑡),𝜔 (𝑡));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (3) 8 𝜔 (𝑡+1) ← 1 |𝑆𝑡 | � 𝑚∈𝑆𝑡 𝜔 (𝑡+1) 𝑚 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (4) 9 𝜃 (𝑡+1) 𝑐 ← � 𝑚∈𝑆𝑡 ,𝑐∈C𝑚 𝜃 (𝑡+1) 𝑚,𝑐 |{𝑚|𝑚∈𝑆𝑡,𝑐 ∈C𝑚 }| ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (4) 10 return 𝜃𝑇,𝜔𝑇 11 ClientUpdate(𝑚,𝜃 (𝑡),𝜔 (𝑡)): 12 𝜃 (𝑡) 𝑚 ,𝜔 (𝑡) 𝑚 ← 𝜃 (𝑡),𝜔 (𝑡);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 13 for 𝑒𝑝𝑜𝑐ℎ ∈ {1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', 𝐸} do 14 Calculate similarities of data and class representations and form distillation set ˜D (𝑡) 𝑚 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (a) 15 Freeze 𝜃 (𝑡) 𝑚 and update 𝜔 (𝑡+1) 𝑚 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (b) 16 Freeze 𝜔 (𝑡+1) 𝑚 and update 𝜃 (𝑡+1) 𝑚 ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (c) 17 return 𝜃 (𝑡+1) 𝑚 , and 𝜔 (𝑡+1) 𝑚 to server ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' // Step (d) Pre-training text representations and label encoder initialization in (1) are conducted only once at the beginning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Steps (2)-(4) repeat for 𝑇 rounds until the global model converges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' During local training in (3), each client 𝑚 ∈ 𝑆𝑡 conducts the following steps: (a) Form distillation dataset for locally-unaware classes: Client 𝑚 forms a distillation set ˜D (𝑡+1) 𝑚 for locally-unaware classes C𝑚 by using the latest class representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (b) Train data encoder: Client 𝑚 freezes the label encoder pa- rameters and trains the data encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (c) Train label encoder: Client 𝑚 freezes the parameters in the data encoder and trains the label encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (d) Model communication after local updates: Client𝑚 sends the updated model parameters to the server.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 Semantic Label Name Sharing The vanilla classification model in Section 2 learns latent label spaces merely based on the local training data with numerical label IDs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' With non-identical client class sets, local models at dif- ferent clients are likely to form different and even independent label spaces, making the classification boundaries aggregated at the server inaccurate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To better align the label spaces, we turn to the semantics of label names as a common ground to anchor class representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The natural language label names carry valuable information for understanding the label correlations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For example, in behavioral context recognition, the activity of “lying down" is likely to indicate the person is “sleeping", and the possible location of the activity is “at home".' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Such knowledge about label correlations not only exists in the datasets to investigate, but also can be mined through analyzing the semantics of the label names.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Incorporating Label Encoder to Classification Model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We re- place the classification layer in a conventional classification model with a label encoder as shown in Figure 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The label encoder 3 𝑓𝜃 : 𝑤𝑐 → 𝑟𝑐 ∈ 𝑅𝑑 takes the natural language label names 𝑤𝑐 as the inputs and maps them into representations 𝑟𝑐.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Prior knowledge about label semantics can be inserted into the label encoder by ini- tializing it with pre-trained label embeddings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Inspired by existing works that learn semantic word embeddings based on word-word co-occurrence and point-wise mutual information (PMI) [1, 14, 34], we use an external text corpus related to the domain of the classifica- tion task to extract knowledge of label co-occurrence and pre-train the label embeddings for initializing the label encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The details about label embedding pre-training can be found in Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For each input data 𝑥𝑖, the model calculates the matching scores of 𝑥𝑖 and every class by taking the dot product of the data repre- sentation ℎ𝜔 : 𝑥𝑖 → 𝑧𝑖 ∈ R𝑑 output from the data encoder and the class representations output by the label encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' This way, it calculates the similarity between the representations of input data and classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, an activation function is applied to the dot product to get the prediction probabilities of 𝑥𝑖 belonging to each class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Same as described in Section 2, for multi-label classification, the sigmoid function is used and the final prediction is achieved by thresholding the probabilities at 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For single-label classification, the softmax function is used and the final prediction is the class with the maximum probabilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Alternating Training of Encoders.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The data encoder and label encoder are two branches in the classification model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We want the representations obtained by one encoder to regulate the training of the other while preventing mutual interference.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Thus, at each learning iteration, we first freeze the parameters in the label encoder 𝜃 and update the parameters of the data encoder 𝜔.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, we freeze the parameters in the data encoder and update the label encoder parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 Knowledge Distillation for Locally-unaware Classes Due to the lack of label information of certain classes to support supervised training, the training at each client is biased toward the identified classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To mitigate such drift, we design a knowledge distillation strategy for unaware classes at each client by further utilizing the class representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The class representations are updated with the training of the label encoder at each round of communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' As the classification problem is formed as a match- ing between the embeddings of data and label names in Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2, the distance between the label names and the data samples in the latent space can tell how likely a sample belongs to a class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Thus, we use the class representations to annotate samples for the locally- unaware classes and distill knowledge to local models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Forming Local Distillation Dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' When the client receives the parameters of the data encoder 𝜔 (𝑡) and label encoder 𝜃 (𝑡) at 𝑡-th round, it uses the label encoder to get the updated class repre- sentations {𝑓 (𝑡) 𝜃 (𝑤𝑐)|𝑐 ∈ C} and uses the data encoder to generate the data representations of its local data {ℎ(𝑡) 𝜔 (𝑥𝑖)|(𝑥𝑖,𝑦𝑖) ∈ D𝑚}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, for each locally-unaware class �𝑐, the client calculates its co- sine similarities to the data: 𝑠 (𝑡) 𝑖,�𝑐 = 𝑓 (𝑡) 𝜃 (𝑤 �𝑐) · ℎ(𝑡) 𝜔 (𝑥𝑖) ∥𝑓 (𝑡) 𝜃 (𝑤 �𝑐)∥ · ∥ℎ(𝑡) 𝜔 (𝑥𝑖)∥ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Samples are annotated according to their similarities to the class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The samples with the highest similarities with the class representa- tion 𝑓 (𝑡) 𝜃 (𝑤 �𝑐) are annotated as positive samples of class �𝑐.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The client also annotates the negative samples for the locally-unaware classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The sample with the least similarities with 𝑓 (𝑡) 𝜃 (𝑤 �𝑐) is annotated as a negative sample of �𝑐.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The number of samples to be annotated depends on the per- centile of similarity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We set the thresholds ˆ𝜏 (𝑡) �𝑐 and ˇ𝜏 (𝑡) �𝑐 as the 𝑞1-th and 𝑞2-th percentile of the similarities over all samples for annotat- ing positive samples and negative samples respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The client annotates a subset of the samples whose similarities are higher than ˆ𝜏 (𝑡) �𝑐 as the positive samples in class �𝑐 (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', ˜𝑦(𝑡) 𝑖,�𝑐 = 1) and those with selection scores lower than ˇ𝜏 (𝑡) �𝑐 as the negative samples in class �𝑐 (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', ˜𝑦(𝑡) 𝑖,�𝑐 = 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The distillation dataset after the 𝑡-th round is: ˜D (𝑡) 𝑚 ← {(𝑥𝑖, ˜𝑦(𝑡) 𝑖 )|𝑠 (𝑡) 𝑖,�𝑐 > ˆ𝜏 (𝑡) �𝑐 or 𝑠 (𝑡) 𝑖,�𝑐 < ˇ𝜏 (𝑡) �𝑐 , �𝑐 ∈ C𝑚} In single-label classification, there is another constraint that the unlabeled sample can be annotated as a positive sample of class �𝑐 only if the class is the closest to the data representation among all classes, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', �𝑐 = 𝑎𝑟𝑔𝑚𝑎𝑥 𝑐 ∈C (𝑠 (𝑡) 𝑖,𝑐 ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Knowledge Distillation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The pseudo annotations about the locally- unaware classes C𝑚 are then used as knowledge to transfer to client 𝑚’s local model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Let 𝑁 (𝑡) �𝑐 be the number of samples in ˜D (𝑡) 𝑚 , 𝑔(𝑡+1) 𝑚 (·) be the local model with parameters being updated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use binary cross-entropy loss for multi-label classification as the distillation loss: L(𝑡) C𝑚 = − 1 𝑁 ′ 𝑚 𝑁 ′ 𝑚 ∑︁ 𝑖=1 ∑︁ �𝑐 ∈C𝑚 [ ˜𝑦(𝑡) 𝑖,𝑐 log𝑔(𝑡+1) 𝑚 (𝑥𝑖)�𝑐+ (1 − ˜𝑦(𝑡) 𝑖,𝑐 ) log(1 − 𝑔(𝑡+1) 𝑚 (𝑥𝑖)�𝑐)], where 𝑁 ′ 𝑚 is the number of samples in the distillation set ˜D (𝑡) 𝑚 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For single-label classification, we use cross-entropy loss as dis- tillation loss: L(𝑡) C𝑚 = − 1 𝑁 ′ 𝑚 𝑁 ′ 𝑚 ∑︁ 𝑖=1 ∑︁ 𝑐 ∈C ˜𝑦(𝑡) 𝑖,𝑐 log𝑔(𝑡+1) 𝑚 (𝑥𝑖)𝑐.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use a parameter 𝛼 to control the weight of the distillation loss term.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The learning objective of the (𝑡)-th round on client 𝑚 is to minimize the following loss term: L(𝑡) 𝑚 = L(𝑡) C𝑚 + 𝛼L(𝑡) C𝑚 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 Model Aggregation After the clients finish the local training, they send the updated parameters to the server.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' After the server collects the parameters from all participating clients 𝑚 ∈ 𝑆𝑡, it conducts model aggregation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For the data encoder, the parameters are averaged across all client models: 𝜔 (𝑡+1) ← 1 |𝑆𝑡 | ∑︁ 𝑚∈𝑆𝑡 𝜔 (𝑡+1) 𝑚 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4 For the label encoder, the representations of class 𝑐 are aver- aged only among the client models that have class 𝑐 as its locally- identified class: 𝜃 (𝑡+1) 𝑐 ← � 𝑚∈𝑆𝑡,𝑐 ∈C𝑚 𝜃 (𝑡+1) 𝑚,𝑐 |{𝑚|𝑚 ∈ 𝑆𝑡,𝑐 ∈ C𝑚}| .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4 EXPERIMENTS We first introduce the applications and the datasets we use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, we describe the compared methods and experimental setup.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We show the experimental results with different settings and conduct an ablation study and to show the effectiveness of the designs in FedAlign.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='1 Datasets We conduct experiments on 6 datasets covering 4 different applica- tion scenarios and both single-label and multi-label classification problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Table 1 offers an overview and the details are as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (1) Behavioral Context Recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The task is to infer the context of human activity, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', what the person is doing, where and with whom the person is.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ExtraSensory [40] is a bench- mark dataset for this application scenario.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It contains 51 classes in total which can be partitioned into 5 main categories: posture, location, activity, and companion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Based on ExtraSensory, we construct 3 datasets with non-overlapping client class sets as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ES-5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In this dataset, we have 5 clients and every client only has annotations from a different main category (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', one main category to one client).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Training samples are then assigned to clients according to their associated classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Since ExtraSen- sory is a multi-label dataset, the same training sample may have multiple classes from different main categories.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In this case, we assign the training sample based on the most infre- quent class among these multiple labels to ensure that each locally-identified class will have at least one positive sample.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To make this dataset more realistic, we always assign all the data of each human subject to the same client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ES-15 and ES-25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We also increase the number of clients to 15 and 25 to further challenge the compared methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We start with the 5 class groups as ES-5 and iteratively split the class groups until the number of class groups is the same as the number of clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For every split, we select the group with the most number of classes and randomly separate it into two sub-groups.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Every class group is visible and only visible to one client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' One can then apply a similar process as ES-5 to assign training samples to different clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (2) Medical Code Prediction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Medical codes describe whether a patient has a specific medical condition or is at risk of de- velopment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The task aims at automatically annotating medical codes from clinical notes that record information about what happened during a patient’s hospitalization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We start with the MIMIC-III database [7] and follow the preprocessing proce- dure in [32] to form the benchmark MIMIC-III 50-label dataset that contains the 50 most frequent labels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' These classes can be Table 1: Our datasets cover application scenarios of behav- ioral context recognition, clinical phenotype classification, human activity recognition, and text classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The im- balance factor of a dataset refers to the ratio of its smallest class size to the largest class size.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Dataset |C| # of clients Avg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 𝑁𝑚 Avg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' |C𝑚| Imbalance Remarks ES-5 51 5 5,446 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='0013 Multi-label, non- overlapping client class sets ES-15 15 1,769 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 ES-25 25 1,073 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='04 MIMIC-III-10 50 10 807 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='1157 PAMAP2-9 18 9 1,287 5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2049 Single-label R8-8 8 8 617 3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='0130 grouped into 10 categories according to the ICD-9 taxonomy1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We construct the MIMIC-III-10 dataset by partitioning the dataset into 10 clients following the same strategy as that in ES-5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (3) Human Activity Recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The task aims at identifying the movement or action of a person based on sensor data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We start with the PAMAP2 [38] dataset, which contains data of 18 different physical activities collected from 9 subjects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We construct the PAMAP2-9 dataset by using the data from each subject as a client, which results in 9 clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For each client, we randomly select 5 classes to be its locally-identified classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (4) Text Classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The task is to categorize text into orga- nized groups.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use the Reuters-21578 R8 dataset [2], which is a collection of news articles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We construct the R8-8 dataset by setting the number of clients as 8, randomly dividing the data into 8 partitions, and assigning one partition to each client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For each client, we randomly select 3 classes to be the identified classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' More details about data preprocessing are described in Appendix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 Compared Methods We compare FedAlign with various classical [30] and state-of-the- art federated learning methods [9, 20, 21] designed for non-IID data problems as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedAvg [30] is a classical federated learning method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The weights of local models uploaded by clients in each round are averaged to get the global model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, the server sends the updated model to each client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedProx [21] deals with the heterogeneity in the client models by enforcing a 𝐿2 regularization term in local optimization to limit the distance between local model and global model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' MOON [20] adds a contrastive loss term to maximize the consis- tency of representations learned by the local model and that by the global model and minimize the consistency between repre- sentations learned by the local models of consecutive rounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Scaffold [9] addresses the client-variance problem by maintain- ing control variates to estimate the update directions of the global model and the client model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The drift of local training is approx- imated by the difference between the local and global update directions and added to the local updates to mitigate drift.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 1the International Statistical Classification of Diseases and Related Health Problems (ICD): ftp://ftp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='cdc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='gov/pub/Health_Statistics/NCHS/Publications/ICD-9/ucod.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='txt 5 Table 2: Main experimental results (%) on the six datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ExtraSensory (ES-5, ES-15, ES-25) and MIMIC-III are multi-label datasets where class sets of different clients are non-overlapping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PAMAP2 and R8 are single-label datasets where the class sets of different clients overlap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The results are averaged over 5 runs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Method ES-5 ES-15 ES-25 MIMIC-III-10 PAMAP2-9 R8-8 F1 Acc F1 Acc F1 Acc F1 Acc F1 Acc F1 Acc FedAvg [30] 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='77 80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='79 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='71 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='44 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='52 50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='42 35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='04 67.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='07 68.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='89 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='45 78.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='51 92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='76 FedProx (𝜇 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='001) [21] 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='26 80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='67 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='42 61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='91 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='48 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='16 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='75 65.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='98 69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='70 73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='63 79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='05 92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='61 FedProx (𝜇 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='0001) [21] 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='53 79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='14 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='52 61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='95 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='05 52.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='33 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='73 64.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='31 69.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='38 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='78 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='98 91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='78 MOON [20] 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='12 81.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='00 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='84 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='53 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='52 50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='24 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='62 66.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='34 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='70 74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='25 79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='26 93.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='07 Scaffold [9] 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='14 77.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='13 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='15 61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='69 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='73 48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='81 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='58 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='84 73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='57 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='60 82.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='83 94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='43 FedRS (𝛼 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5) [22] 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='01 78.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='72 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='50 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='09 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='44 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='41 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='82 66.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='80 68.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='70 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='42 80.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='10 92.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='74 FedRS (𝛼 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='9) [22] 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='25 79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='25 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='55 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='17 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='40 50.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='87 35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='44 67.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='45 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='81 74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='44 76.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='68 91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='81 FedPU [26] 85.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='39 87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='59 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='17 94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='23 FedAlign 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='69 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='92 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='36 73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='61 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='78 62.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='11 37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='87 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='13 87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='21 88.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='14 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='76 94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='92 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='22 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='24 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='26 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='28 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='32 F1 ES-5 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='14 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='16 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='18 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='22 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='24 ES-15 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='12 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='14 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='16 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='18 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='22 ES-25 20 40 60 80 100 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 MIMIC-III-10 10 20 30 40 50 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='75 1 R8-8 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 Accuracy 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='7 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 20 40 60 80 100 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='7 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='75 1 FedAlign FedAvg MOON FedProx Scaffold FedPU FedRS 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='75 1 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='75 1 PAMAP2-9 Figure 3: Performance w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Communication Rounds on Six Datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The results are averaged over 5 runs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We also compare two state-of-the-art federated learning meth- ods [22, 26] designed for classification problems with missing classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Specifically, FedRS [22] is designed for federated learning with missing classes where each client only owns data of part of the classes (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', locally- identified classes in our terminology).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It restricts the weight up- date of the missing classes in the classifier by adding scaling factors to the softmax operation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedPU [26] is designed for the scenario where each client only labels a small part of their dataset and there exists unlabeled data from both locally-identified classes and locally-unaware classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It utilizes the labeled data at each client to estimate the misclassi- fication loss between the unaware classes of the other clients and adds the estimated loss to the local optimization objective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' As the calculation of the expected risk is based on single-label classifica- tion, where each sample is associated with only one of the classes, it does not hold true in multi-label problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' While it is difficult to generalize FedPU for multi-label classification problems, we compare it with FedAlign in the single-label applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 Experiment Setup Base Neural Network Model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For a fair comparison, we use the same network setting for all compared methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The data encoder is a Transformer Encoder [41].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use one encoder layer in the data encoder, where the number of heads in multi-head attention is 4 and the dimension of the feed-forward network is 64.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The label encoder is a single hidden layer neural network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The dimension of data and class representations 𝑑 is 256.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Evaluation Metrics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Due to label imbalance, we adopt accuracy and F1-score to evaluate the performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' They are often used as benchmark metrics for the datasets and tasks in our experiments [32, 35, 38, 40].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We calculate the metrics for each class and report the macro-average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Train/Test Split.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For MIMIC-III and R8, we use the data split pro- vided by the dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For the other datasets, we use the data from 80% of the data for training and 20% of the data for testing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Learning Setting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For ExtraSensory (ES-5, ES-15, ES-25), PAMAP2- 9, R8-8, we run 𝑇 = 50 communication rounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For MIMIC-III-10, we run 𝑇 = 100 rounds as it takes longer to converge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The number of selected clients per round is |𝑆𝑡 | = 5 and the local epochs 𝐸 = 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Optimizer and Hyperparameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For compared methods, we try different values for hyperparameters 𝜇 (in FedProx and MOON) 6 Table 3: Results (% Averaged Over 5 Runs) of Ablation Study Method PAMAP2-9 R8-8 F1 Acc F1 Acc FedAvg 68.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='89 71.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='45 77.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='27 90.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='59 FedAlign w/o semantic 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='39 85.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='40 82.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='37 94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='13 FedAlign w/o distillation 70.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='87 74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='59 79.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='67 91.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='06 FedAlign w/o alternation 86.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='01 87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='50 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='22 94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='63 FedAlign 87.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='21 88.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='14 83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='76 94.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='92 and 𝛼 (in FedRS) that are often adopted in the previous papers [20– 22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For MOON, we set 𝜇 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='001 for all datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For FedProx, we report the performance with 𝜇 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='001 and 𝜇 = 10−5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For FedRS, we report the performance with 𝛼 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 and 𝛼 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 Main Results and Analysis Single-Label, Non-identical but Overlapping Client Class Sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Table 2 shows the results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For both PAMAP2 and R8 datasets, FedAlign performs better than the baseline methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The non-IID problems that FedRS and FedPU aim to tackle are slightly different from ours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Although they show improvements over the classical algorithm FedAvg and the methods designed for the typical non- IID setting (FedProx, MOON, and Scaffold), FedAlign shows better performance compared with FedRS and FedPU in the problem of non-identical client class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Multi-Label, Non-overlapping Client Class Sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' As one can clearly see, FedAlign always yields better performance than the baseline methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Remarkably, with non-identical client class sets, the three state-of-the-art algorithms designed to deal with non-IID data (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', FedProx, MOON, and Scaffold) do not guarantee improve- ment over FedAvg (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', Scaffold loses to FedAvg on ES-5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedRS does not show improvement over its base algorithm FedAvg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Al- though it is designed for federated learning with missing classes, its mechanism is specifically based on single-label (softmax) classi- fication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In multi-label (sigmoid) classification, the weight update of class 𝑐 is only affected by features from class 𝑐.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Adding scaling factors to missing classes does not affect the weight update of the identified classes, thus, does not solve the non-IID problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For the ExtraSensory dataset, as the number of partitions in- creases (5, 15, 25), the performance of the methods declines.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' This is due to two reasons: first, the label sets are more widely distributed so the training samples with identified classes on each client be- come fewer;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' second, the label space of different clients becomes more diverse, making it difficult to learn and aggregate information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' However, we observe a growing trend in the improvement of test- ing accuracy when increasing the number of partitions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' With more widely distributed classes and fewer training samples, FedAlign show greater advantages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Performance vs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Communication Rounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Figure 3, we show the test performance with respect to the number of communication rounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We see that FedAlign outperforms the baselines since a few rounds of communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' There is a clear performance gap between the baseline methods and FedAlign.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 1 F1 PAMAP2-9 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 1 Accuracy 10 20 30 40 50 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 1 F1 R8-8 10 20 30 40 50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='8 1 Accuracy FedAlign FedAlign w/o distillation FedAlign w/o semantic FedAvg FedAlign w/o alternation Figure 4: Performance (averaged over 5 runs) w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' commu- nication rounds in ablation study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5 Ablation Study We conduct an ablation study to evaluate the contribution of each design in FedAlign on the PAMAP and R8 datasets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' First, we evalu- ate the performance of the method without knowledge distillation (denoted as FedAlign w/o distillation).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The classification model consists of a label encoder and a label encoder and the framework conducts alternating training of the two modules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Second, we eval- uate the performance of the method without the semantic label name sharing (denoted as FedAlign w/o semantic).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The distil- lation dataset is formed by annotating the samples according to their prediction confidence given by the latest global model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Highly confident predictions are used as pseudo-labels and the confidence threshold is decided by the same percentile value as in FedAlign.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Third, We evaluate the performance of the method without alter- nating training (denoted as FedAlign w/o alternation) which updates the label encoder and data encoder simultaneously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Note that the model aggregation method in FedAlign is based on Fe- dAvg (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', averaging the model parameters), thus we also compare FedAvg as the baseline method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Table 3 shows the performance at the end of 50 communication rounds and Figure 4 shows the performance with respect to the num- ber of communication rounds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We notice the performance drops when removing any of the designs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The methods with complete features can work better than using only part of the designs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' This shows the key designs (semantic label name sharing, knowledge distillation, and alternating training) in FedAlign all contribute to performance improvement, and combining them can produce the best performance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 5 RELATED WORK We first review the important topic related to the problem we discuss, that is, federated learning with non-IID data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Then, we review two lines of related work that motivate our designs in the framework: label semantics modeling and knowledge distillation 7 in federated learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='1 Federated Learning with non-IID data One of the fundamental challenges in federated learning is the presence of non-IID data [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The reasons and solutions to this challenge are being actively explored.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' One common solution is to mitigate client variance during local training by adding regulariza- tion [20, 21], or estimating global update directions and removing drift [9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Another type of solution improves model aggregation at the server [25, 43, 44], such as assigning aggregation weights accord- ing to local optimization steps [44] or doing layer-wise aggregation by matching hidden elements with similar feature extraction signa- tures [43].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Other works propose solutions from the data perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' They leverage public datasets to conduct distillation for model fu- sion [25] or calibrate classifiers using synthesized features [28, 50].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Personalization is another type of solution [16, 39], which enables different clients to have different model parameters with regard to their own data distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' These methods tackle more relaxed non-IID problems that assume clients have the same set of classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We study the problem of non-identical class sets in federated learning, which is a more general case of non-IID label distribution skew.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Some recent works [22, 26] consider the problem that clients can only access part of the whole class set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedRS [22] deals with the case where each client only owns the data of certain classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It restricted the weight update of the missing classes during the local procedure by adding a scaling factor in the softmax function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedPU [26] focuses on a scenario where each client only labels a small part of their dataset and there exists unlabeled data from both positive classes (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', identified classes in ours) and negative (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', locally-unaware in ours) classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It utilizes the labeled data at each client to estimate the misclassification loss between the negative classes of the other clients and adds the estimated loss to local optimization objective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The problem settings in the two related works are slightly different from ours and applying them to our problem shows less improvement compared to our proposed method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 Label Semantics Modeling For tasks where some of the label co-occurrence patterns can not be directly observed from the training dataset, such as zero-shot learning [11], it is hard for the model to understand label corre- lations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To deal with the problem, several methods are proposed to leverage prior knowledge such as knowledge graphs [42] or model semantic label embedding from textual information about classes [13, 29, 36, 45, 48].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For example, Ba et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [13] derived em- bedding features for classes from natural language descriptions and learned a mapping to transform the text features of classes to the visual image feature space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Radford et al [36] used contrastive pre-training to jointly train an image encoder and a text encoder and predict the correct pairings of image and text caption, which helps to produce high-quality image representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Matsuki et al [29] and Wu et al [45] incorporate word embeddings for zero-shot learning in human activity recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' These methods show the potential of allowing models to use semantic relationships between labels to make predictions for classes not observed in the training set, which motivates our design of semantic label name sharing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 Knowledge Distillation in Federated Learning Knowledge distillation [5] is originally used for transferring knowl- edge from a large model or ensemble of models to a single smaller model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The large pre-trained models are regarded as the teachers and the smaller model (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', student) mimics the outputs or inter- mediate features of the teachers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In federated learning, knowledge distillation has been used to deal with statistical heterogeneity [50] and system heterogeneity (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', different client models) [18, 25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' A public dataset is usually required, or otherwise, a generator is trained to get synthetic data [50].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Recently, semi-supervised meth- ods are also applied in federated learning for cases where labeled training data is limited [6, 24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Our work is different from the pre- vious works in that the non-identical class sets setting we focus on has no labeled data for the unaware classes at each client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use the class representations to annotate data samples and add regular- ization in a knowledge distillation manner to transfer knowledge about unaware classes to client models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 6 CONCLUSIONS AND FUTURE WORK We studied the problem of federated classification with non-identical class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We propose the FedAlign framework and demonstrated its use in federated learning for various applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedAlign breaks the classification model into a data encoder and a label en- coder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Semantic label learning is conducted by leveraging a domain- related corpus and shared label names.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The pre-trained semantic label embeddings contain the knowledge of label correlations and are then used to guide the training of data encoder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Moreover, the knowledge distillation strategy complements the training of unaware classes at each client.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' These two modules are a key to mit- igating client variance in FedAlign, which addresses the challenge of non-identical class sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We show that FedAlign improves the baseline algorithms for federated learning with non-IID data and achieves new state-of-the-art.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For future directions, we consider more general system hetero- geneity where the participants have different network architectures, different training processes, and different tasks at the same time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We plan to extend our study to make federated learning compatible with such heterogeneity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' REFERENCES [1] John A Bullinaria and Joseph P Levy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Extracting semantic representations from word co-occurrence statistics: A computational study.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Behavior research methods 39, 3 (2007), 510–526.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [2] Ana Cardoso-Cachopo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Improving Methods for Single-label Text Cate- gorization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PdD Thesis, Instituto Superior Tecnico, Universidade Tecnica de Lisboa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [3] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Bert: Pre-training of deep bidirectional transformers for language understanding.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:1810.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='04805 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [4] Felix Hamborg, Norman Meuschke, Corinna Breitinger, and Bela Gipp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' news-please: A Generic News Crawler and Extractor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 15th International Symposium of Information Science (Berlin).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 218–223.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='5281/zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='4120316 [5] Geoffrey Hinton, Oriol Vinyals, Jeff Dean, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Distilling the knowledge in a neural network.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:1503.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='02531 2, 7 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [6] Wonyong Jeong, Jaehong Yoon, Eunho Yang, and Sung Ju Hwang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Federated semi-supervised learning with inter-client consistency & disjoint learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='12097 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [7] Alistair EW Johnson, Tom J Pollard, Lu Shen, Li-wei H Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and 8 Roger G Mark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' MIMIC-III, a freely accessible critical care database.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Scientific data 3, 1 (2016), 1–9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [8] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Ben- nis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Advances and open problems in federated learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Foundations and Trends® in Machine Learning 14, 1–2 (2021), 1–210.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [9] Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebas- tian Stich, and Ananda Theertha Suresh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Scaffold: Stochastic controlled averaging for federated learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In International Conference on Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PMLR, 5132–5143.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [10] Stefanos Laskaridis, Dimitris Spathis, and Mario Almeida.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Federated mobile sensing for activity recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 27th Annual International Conference on Mobile Computing and Networking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 858–859.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [11] Chung-Wei Lee, Wei Fang, Chih-Kuan Yeh, and Yu-Chiang Frank Wang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Multi-label zero-shot learning with structured knowledge graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the IEEE conference on computer vision and pattern recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 1576–1585.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [12] Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' BioBERT: a pre-trained biomedical language representation model for biomedical text mining.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Bioinformatics 36, 4 (2020), 1234–1240.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [13] Jimmy Lei Ba, Kevin Swersky, Sanja Fidler, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Predicting deep zero-shot convolutional neural networks using textual descriptions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the IEEE international conference on computer vision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4247–4255.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [14] Omer Levy and Yoav Goldberg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Linguistic regularities in sparse and explicit word representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the eighteenth conference on computational natural language learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 171–180.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [15] Ang Li, Jingwei Sun, Pengcheng Li, Yu Pu, Hai Li, and Yiran Chen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Hermes: an efficient federated learning framework for heterogeneous mobile clients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 27th Annual International Conference on Mobile Computing and Networking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 420–437.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [16] Ang Li, Jingwei Sun, Xiao Zeng, Mi Zhang, Hai Li, and Yiran Chen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Fed- Mask: Joint Computation and Communication-Efficient Personalized Federated Learning via Heterogeneous Masking.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 42–55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [17] Chenglin Li, Di Niu, Bei Jiang, Xiao Zuo, and Jianming Yang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Meta-har: Federated representation learning for human activity recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the Web Conference 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 912–922.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [18] Daliang Li and Junpu Wang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Fedmd: Heterogenous federated learning via model distillation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:1910.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='03581 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [19] Liang Li, Dian Shi, Ronghui Hou, Hui Li, Miao Pan, and Zhu Han.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To talk or to work: Flexible communication compression for energy efficient federated learning over heterogeneous mobile edge devices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In IEEE INFOCOM 2021-IEEE Conference on Computer Communications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' IEEE, 1–10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [20] Qinbin Li, Bingsheng He, and Dawn Song.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Model-contrastive federated learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 10713–10722.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [21] Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, and Virginia Smith.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Federated optimization in heterogeneous networks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Proceedings of Machine Learning and Systems 2 (2020), 429–450.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [22] Xin-Chun Li and De-Chuan Zhan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Fedrs: Federated learning with restricted softmax for label distribution non-iid data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 995–1005.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [23] Feng Liang, Weike Pan, and Zhong Ming.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Fedrec++: Lossless federated recommendation with explicit feedback.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the AAAI conference on artificial intelligence, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 4224–4231.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [24] Haowen Lin, Jian Lou, Li Xiong, and Cyrus Shahabi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Semifed: Semi- supervised federated learning with consistency and pseudo-labeling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:2108.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='09412 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [25] Tao Lin, Lingjing Kong, Sebastian U Stich, and Martin Jaggi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Ensemble distillation for robust model fusion in federated learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Advances in Neural Information Processing Systems 33 (2020), 2351–2363.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [26] Xinyang Lin, Hanting Chen, Yixing Xu, Chao Xu, Xiaolin Gui, Yiping Deng, and Yunhe Wang.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Federated Learning with Positive and Unlabeled Data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In International Conference on Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PMLR, 13344–13355.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [27] Shuchang Liu, Shuyuan Xu, Wenhui Yu, Zuohui Fu, Yongfeng Zhang, and Amelie Marian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedCT: Federated collaborative transfer for recommendation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 716–725.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [28] Mi Luo, Fei Chen, Dapeng Hu, Yifan Zhang, Jian Liang, and Jiashi Feng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' No fear of heterogeneity: Classifier calibration for federated learning with non-iid data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Advances in Neural Information Processing Systems 34 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [29] Moe Matsuki, Paula Lago, and Sozo Inoue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Characterizing word embeddings for zero-shot sensor-based human activity recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Sensors 19, 22 (2019), 5043.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [30] Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Communication-efficient learning of deep net- works from decentralized data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Artificial intelligence and statistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PMLR, 1273–1282.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [31] Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Distributed representations of words and phrases and their compositionality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Advances in neural information processing systems 26 (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [32] James Mullenbach, Sarah Wiegreffe, Jon Duke, Jimeng Sun, and Jacob Eisenstein.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Explainable Prediction of Medical Codes from Clinical Text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 2018 Conference of the North American Chapter of the Association for Com- putational Linguistics: Human Language Technologies, Volume 1 (Long Papers).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 1101–1111.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [33] Bethesda (MD): National Library of Medicine.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PMC Open Access Subset [Internet].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='ncbi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='nlm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='nih.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='gov/pmc/tools/openftlist/ [accessed 7-June- 2022].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [34] Jeffrey Pennington, Richard Socher, and Christopher D Manning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Glove: Global vectors for word representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 1532–1543.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [35] Yinhua Piao, Sangseon Lee, Dohoon Lee, and Sun Kim.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Sparse Structure Learning via Graph Neural Networks for Inductive Document Classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the AAAI Conference on Artificial Intelligence, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 11165–11173.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [36] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Learning transferable visual models from natural language supervision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In International Conference on Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PMLR, 8748–8763.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [37] Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Im- proving language understanding by generative pre-training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [38] Attila Reiss and Didier Stricker.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Introducing a new benchmarked dataset for activity monitoring.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In 2012 16th international symposium on wearable computers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' IEEE, 108–109.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [39] Linlin Tu, Xiaomin Ouyang, Jiayu Zhou, Yuze He, and Guoliang Xing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' FedDL: Federated Learning via Dynamic Layer Sharing for Human Activity Recognition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 15–28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [40] Yonatan Vaizman, Katherine Ellis, and Gert Lanckriet.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Recognizing detailed human context in the wild from smartphones and smartwatches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' IEEE pervasive computing 16, 4 (2017), 62–74.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [41] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Attention is all you need.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Advances in neural information processing systems 30 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [42] Laura Von Rueden, Sebastian Mayer, Katharina Beckh, Bogdan Georgiev, Sven Giesselbach, Raoul Heese, Birgit Kirsch, Julius Pfrommer, Annika Pick, Rajkumar Ramamurthy, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Informed Machine Learning–A Taxonomy and Survey of Integrating Knowledge into Learning Systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:1903.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='12394 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [43] Hongyi Wang, Mikhail Yurochkin, Yuekai Sun, Dimitris Papailiopoulos, and Yasaman Khazaeni.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Federated learning with matched averaging.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' arXiv preprint arXiv:2002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='06440 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [44] Jianyu Wang, Qinghua Liu, Hao Liang, Gauri Joshi, and H Vincent Poor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Tackling the objective inconsistency problem in heterogeneous federated opti- mization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Advances in neural information processing systems 33 (2020), 7611–7623.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [45] Tong Wu, Yiqiang Chen, Yang Gu, Jiwei Wang, Siyu Zhang, and Zhanghu Zhechen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Multi-layer cross loss model for zero-shot human activity recogni- tion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Pacific-Asia Conference on Knowledge Discovery and Data Mining.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 210–221.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [46] Chengxu Yang, Qipeng Wang, Mengwei Xu, Zhenpeng Chen, Kaigui Bian, Yunxin Liu, and Xuanzhe Liu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Characterizing impacts of heterogeneity in federated learning upon large-scale smartphone data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the Web Conference 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 935–946.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [47] Jingwei Yi, Fangzhao Wu, Chuhan Wu, Ruixuan Liu, Guangzhong Sun, and Xing Xie.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Efficient-FedRec: Efficient Federated Learning Framework for Privacy-Preserving News Recommendation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2814–2824.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [48] Fengtao Zhou, Sheng Huang, and Yun Xing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Deep semantic dictionary learning for multi-label image classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceedings of the AAAI Conference on Artificial Intelligence, Vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 3572–3580.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [49] Yukun Zhu, Ryan Kiros, Rich Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Aligning books and movies: Towards story-like visual explanations by watching movies and reading books.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In Proceed- ings of the IEEE international conference on computer vision.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 19–27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [50] Zhuangdi Zhu, Junyuan Hong, and Jiayu Zhou.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Data-free knowledge distillation for heterogeneous federated learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' In International Conference on Machine Learning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PMLR, 12878–12889.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 9 A APPENDIX A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='1 Semantic Label Embedding Pre-training.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Before collaborative training, each client shares the natural lan- guage names of its locally-identified classes with the server.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The server then searches for the label names in a large text corpus re- lated to the domain of the classification task.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We count the times that each label name 𝑐𝑖 appears in each text segment (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', sentence or paragraph).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The label names can be phrases that contain mul- tiple words and the order of the words may change in different text segments while representing the same meaning, such as “colon cancer” and “cancer of the colon”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To match phrases, we arrange the label names into sets of words and apply a sliding window of length 𝐿𝑤 on each text segment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' If the set of words in the label name is covered by the words in the sliding window, we mark that the label name appears in the text segment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The length of the sliding window 𝐿𝑤 is varied per the label name to search.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The co-occurrence of a pair of label names 𝑐𝑖 and 𝑐𝑗 is calculated as the point-wise mutual information (PMI): PMI(𝑐𝑖,𝑐𝑗) = log 𝑝(𝑐𝑖,𝑐𝑗) 𝑝(𝑐𝑖)𝑝(𝑐𝑗) , where 𝑝(𝑐𝑖) and 𝑝(𝑐𝑗) are the individual distributions and 𝑝(𝑐𝑖,𝑐𝑗) is the the joint distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The higher the PMI(𝑐𝑖,𝑐𝑗), the stronger the association between the two label names 𝑐𝑖 and 𝑐𝑗.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The server learns semantic label embeddings based on the co-occurrence.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The principle of label embedding learning is to learn a mapping function from labels to representations 𝑓 : C → R𝑑, which enforces labels with similar semantics to have similar representations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' To achieve this, the server builds a label co-occurrence graph 𝐺 = ⟨𝑉, 𝐸⟩, where the nodes 𝑉 represent the class label names and the edges 𝐸 represent the co-occurrence relationship between the nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The PMI values are zero-centered by subtracting the mean and are used as the edge weights between label names.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Edges with negative weights are removed from the graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For every source node 𝑐𝑜 ∈ 𝑉 , we define 𝑁𝑠 (𝑐𝑜) ⊂ 𝑉 as its network neighborhood generated through simulating fixed-length random walks starting from 𝑐𝑜.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The transition probability from node 𝑢 to node 𝑣 for random walk simulation is calculated by normalizing the edge weight 𝜋𝑢,𝑣 = 𝑤𝑢,𝑣/� 𝑣′∈𝑉 𝑤𝑢,𝑣′.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The objective function of label embedding is: max 𝑓 ∑︁ 𝑐𝑜 ∈C (− log𝑍𝑐𝑜 + ∑︁ 𝑐𝑖 ∈𝑁𝑠 (𝑐𝑜) 𝑓 (𝑐𝑖) · 𝑓 (𝑐𝑜)), where 𝑍𝑐𝑜 = � 𝑣∈𝑉 exp(𝑓 (𝑣) · 𝑓 (𝑐𝑜)) and is approximated using negative sampling [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The mapping function 𝑓 is achieved by a single hidden layer feedforward neural network and the objective is optimized using stochastic gradient descent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='2 Text Corpus Domain-related raw corpus is often easily accessible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For example, novel books that describe a variety of daily scenes can be used for understanding human activities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Academic journals are good studying resources for understanding concepts in different domains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use the following corpora for applications in the experiments: BookCorpus.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [49] is a large collection of free novel books col- lected from the web.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It is a popular text corpus in the domain of Natural Language Processing and has helped the training of many influential language models such as BERT [3] and GPT [37].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PubMed Open-Access (OA) subset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [33] is a text archive of jour- nal articles and preprints in biomedical and life sciences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' It has been widely used for biomedical text mining [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' CommonCrawl (CC) News dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [4] is an English-language news corpus collecting news articles published between 2017 to 2019 from news sites worldwide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We use PubMed-OA as the text corpus for clinic phenotype classification and BookCorpus for behavioral context recognition and human activity recognition, and CC News for text classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We regard a sentence as a text segment in BookCorpus, an article as a text segment in CC News, and a paragraph as a text segment in PubMed-OA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='3 Details About Datasets and Preprocessing ExtraSensory dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [40]: collects time-series sensory measure- ments of 60 participants in their natural behavior, using smart- phones and smartwatches.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Relevant context labels of the time-series measurements are annotated.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' There are in total 51 classes, which can be grouped into five main categories: posture (sitting, lying down, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ), location (gym, beach, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ), activity (eating, sleeping, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' ), companion (with friends, with co-workers).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Each sample is associated with 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='6 classes on average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' During data preprocessing, the time-series data of a subject is partitioned into several pieces ac- cording to its labels to ensure each piece of the resulting time-series data is in the same behavioral context from the beginning to the end.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Each piece of data sample is a 10-minute time-series data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Data sample of less than 10 minutes is padded with zeros.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The features at each timestamp represent the measurements taken within one minute.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The feature dimension is 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' MIMIC-III dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [7]: contains inpatient data of over 40, 000 patients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We follow the preprocessing instruction in [32] to derive the benchmark MIMIC-III 50-label dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The data of each patient is the clinical notes that record information about what happened during a patient’s hospitalization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The vocabulary size of the clinical notes is 4,896.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The notes has an average length of 1,512 words.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The dataset contains 50 most frequent classes which can be grouped into 10 categories according to ICD, such as circulatory systems, digestive systems, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Each data sample is associated with 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='69 classes on average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' PAMAP2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [38]: collects sensory measurements of 9 subjects per- forming 18 different physical activities (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=', sitting, ascending stairs, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Data are collected by inertial measurement units (IMU) sen- sors and a heart rate monitor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The feature dimension is 52.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' We organize the data into time series with a window size of 100.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Reuters-21578 R8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' [2]: is a collection of news articles including 8 classes (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' trade, grain, earn, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' For preprocessing, words with no less than 100 occurrences are kept.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' The vocabulary size is 534.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' Each piece of text data has an average length of 66 words.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} +page_content=' 10' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/wtAyT4oBgHgl3EQfnfhR/content/2301.00489v1.pdf'} diff --git a/xdE2T4oBgHgl3EQfLwbX/content/tmp_files/2301.03718v1.pdf.txt b/xdE2T4oBgHgl3EQfLwbX/content/tmp_files/2301.03718v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..f14c81403b9bc66e15a00e944e5a5ff651fdad1f --- /dev/null +++ b/xdE2T4oBgHgl3EQfLwbX/content/tmp_files/2301.03718v1.pdf.txt @@ -0,0 +1,1505 @@ +Simulating optical linear absorption for mesoscale molecular aggregates: +an adaptive hierarchy of pure states approach +Tarun Gera,1 Lipeng Chen,2 Alex Eisfeld,2 Jeffrey R. Reimers,3, 4 Elliot J. Taffet,1 and Doran I. G. B. Raccah1 +1)Department of Chemistry, Southern Methodist University, PO Box 750314, Dallas, TX, +USA +2)Max Planck Institute for the Physics of Complex Systems, Nöthnitzer Str 38, Dresden, +Germany +3)International Centre for Quantum and Molecular Structures and the School of Physics, Shanghai University, 200444, Shanghai, +China +4)School of Mathematical and Physical Sciences, University of technology Sydney, NSW 2007, +Australia +(*Electronic mail: doranb@smu.edu) +(Dated: 11 January 2023) +In this paper, we present a new method for calculating linear absorption spectra for large molecular aggregates, called +dyadic adaptive HOPS (DadHOPS). This method combines the adaptive HOPS (adHOPS) framework, which uses +locality to improve computational scaling, with the dyadic HOPS method previously developed to calculate linear and +non-linear spectroscopic signals. To construct a local representation of dyadic HOPS, we introduce an initial state +decomposition which reconstructs the linear absorption spectra from a sum over locally excited initial conditions. We +demonstrate the sum over initial conditions can be efficiently Monte Carlo sampled, that the corresponding calculations +achieve size-invariant (i.e. O(1)) scaling for sufficiently large aggregates, and that it allows for the trivial inclusion of +static disorder in the Hamiltonian. We present calculations on the photosystem I core complex to explore the behavior +of the initial state decomposition in complex molecular aggregates, and proof-of-concept DadHOPS calculations on an +artificial molecular aggregate inspired by perylene bis-imide. +I. +INTRODUCTION +Optical +spectroscopy +provides +essential +insight +into +excited-state processes of molecular aggregates ranging from +artificial assemblies in solution to photosynthetic pigment- +protein complexes.1–5 The interpretation of spectroscopic ob- +servables in the presence of both structural heterogeneity and +broad homogeneous lineshapes often requires detailed numer- +ical simulations. +However, established quantum dynamics +methods can struggle with the combination of strong electron- +vibrational coupling and spatially-extended structures charac- +teristic of molecular aggregates.6 As a result, the development +of efficient algorithms for simulating time-resolved optical +spectroscopy measurements of extended molecular aggregates +remains a persistent challenge. +From a molecular perspective, simulating optical spec- +troscopy requires solving the time-evolution of nuclear wave +packets on electronic potential energy surfaces. +A variety +of methods have been developed to solve this time evolution +within a formally exact framework such as multiconfiguration +time-dependent Hartree (MCTDH),7–9 multilayer MCTDH +(ML-MCTDH),10–13 multi-configuration Ehrenfest,14–16 and +ab initio multiple spawning.17,18 However, the simultaneous +description of electronic, intra-molecular vibrational, and en- +vironmental degrees of freedom on equal footing introduces +intractable computational scaling for large molecular aggre- +gates; a problem colloquially known as the curse of dimen- +sionality. +Open-quantum system approaches provide a powerful +coarse-graining where the relevant electronic degrees of free- +dom are time-evolved in a reduced density matrix while cou- +pled to an effective thermal environment. Most commonly, an +open-quantum system Hamiltonian incorporates a linear cou- +pling to a infinite collection of harmonic oscillators parame- +terized to mimic the influence of both the intramolecular vi- +brations and the surrounding (condensed phase) environment +within a linear response approximation. Within this frame- +work, several formally exact methods are available for mod- +eling molecular excitons including Hierarchical equations of +motion (HEOM),19,20 quasi-adibatic path integrals,21,22 time- +dependent density matrix renormalization group theory (TD- +DMRG),23,24 time-evolving density operator with orthogo- +nal polynomials,25,26 and the multi-D1 Davydov ansatz.27,28 +While these methods are capable of scaling to much larger +aggregates than are usually achievable with wave function +propagation techniques, they still suffer from effectively ex- +ponential scaling with the number of electronic states (i.e. +molecules) included in the calculations. +Recent develop- +ments in reduced scaling techniques - including modular path +integrals29,30 and tensor contraction31–35 - suggest new paths +forward may continue to extend the size of molecular aggre- +gates that are tractable with these methods. +Recently, there has been a growing interest in solving +open-quantum systems using stochastic wave functions. In +these approaches, the reduced density matrix is unravelled +into an ensemble of wave functions that can be time-evolved +independently. +In this context, the non-Markovian quan- +tum state diffusion (NMQSD) equation36,37 provides a for- +mally exact solution to the dynamics. The hierarchy of pure +states (HOPS) is a numerically convenient formulation of the +NMQSD equations38–40 which has been recently extended +into the dyadic HOPS equation to simulate both linear41 and +non-linear42 optical spectroscopy. Dyadic HOPS propagates +the bra- and ket-side of the reduced density matrix separately +arXiv:2301.03718v1 [quant-ph] 9 Jan 2023 + +2 +according to the HOPS equation and provides a clear connec- +tion to the non-linear response function formalism.5 +Dyadic HOPS is limited to small molecular aggregates +by the same poor computational scaling that plagues HOPS. +The recent development of adaptive HOPS (adHOPS), which +leverages the locality of physical wave functions to achieve +computational costs that stop increasing with system-size for +sufficiently large aggregates (i.e. O(1) scaling),43 raises the +possibility of dramatically expanding the reach of dyadic +HOPS within an adaptive framework. +In this paper, we +demonstrate dyadic adaptive HOPS (DadHOPS) calculations +of linear absorption for large molecular aggregates. +We +present an initial state decomposition for the dipole auto- +correlation function that characterizes linear absorption which +provides a local construction of the spectroscopic observable. +By combining DadHOPS with an initial state decomposition, +we are able to perform size-invariant scaling simulations of +linear absorption spectra for mesoscale molecular aggregates +and easily incorporate the influence of static disorder in the +system Hamiltonian. +This paper is organized as follows: In Sec. II, we pro- +vide a theoretical background for the readers. We discuss the +Hamiltonian for an open quantum system followed by a brief +introduction for the HOPS and dyadic HOPS methods. In +Sec. III, we describe an algorithm to connect adaptivity with +dyadic HOPS method and provide conditions for relative error +bounds that are necessary for DadHOPS. In Sec. IV, we ap- +ply the initial state decomposition method to decompose the +dipole autocorrelation function into a sum over local corre- +lation functions. We discuss an effective method for Monte +Carlo sampling over stochastic noise trajectories and initial +states, with both dyadic HOPS and DadHOPS. Using 4-site +and 12-site chains as examples we provide proof-of-concept +calculations for the applicability of the DadHOPS method. In +Sec. V, we apply the initial state decomposition to photosys- +tem I and then we apply DadHOPS to a realistic model for +artificial molecular aggregates inspired by Perylene bis-imide +(PBI). Finally, we conclude in Sec. VI with a summary and a +brief outlook. In Appendix A, we provide a detailed deriva- +tion of the dipole autocorrelation function decomposition into +a sum of local correlation functions based on a generalized +initial state decomposition. +II. +THEORETICAL BACKGROUND +A. +Hamiltonian +We model molecular aggregates consisting of N chro- +mophores with an open-quantum system Hamiltonian +ˆH = ˆHS + ˆHB + ˆHSB +(1) +where the system Hamiltonian +ˆHS = Eg|g⟩⟨g|+ +N +∑ +n=1 +En|n⟩⟨n|+ +N +∑ +n=1 +N +∑ +m̸=n +Vnm|n⟩⟨m| +(2) +is composed of a shared electronic ground state, electronic ex- +cited states with vertical excitation energy (En), and an elec- +tronic coupling between pigments Vnm. The electronic states +of each pigment are linearly coupled +ˆHSB = − +N +∑ +n=1 +ˆLn∑ +q +gnq +�ˆb† +nq + ˆbnq +� +(3) +to an independent harmonic reservoir +ˆHB = +N +∑ +n=1∑ +q +¯hωnqˆb† +nqˆbnq +(4) +with a system-bath coupling operator ˆLn = |n⟩⟨n|. The influ- +ence of the vibrational modes on the dynamics of the elec- +tronic system is described by the bath-correlation function +αn(τ) = +� ∞ +0 dω Jn(ω) +� +coth +�β ¯hω +2 +� +cos(ωτ)−isin(ωτ) +� +(5) +that contains the spectral density Jn(ω) = ∑q |gnq|2δ(ω −ωnq) +and the inverse temperature β = +1 +kBT . We decompose the bath +correlation function into a sum of exponentials indexed by jn +αn(t) = ∑ +jn +gjne−γ jnt/¯h +(6) +where gjn and γjn are, in general, complex valued. +B. +Light-Matter Interaction +The light matter interaction is described in terms of the +scalar, collective dipole moment operator +ˆµeff = +N +∑ +n=1 +(µµn ·εε)|n⟩⟨g|+h.c., +(7) +which is defined by the polarization of the incident electric +field (εε) and the transition dipole operator the individual pig- +ments (µµn). The action of the collective dipole moment oper- +ator on the ground state of the aggregate is to excite the super- +position state +|ψex⟩ = +1 +µtot +ˆµeff |g⟩ +(8) +where µtot = +� +∑N +n=1(µµn ·εε)2. +The linear absorption spectrum is given by the half-sided +Fourier transform of the dipole auto-correlation function +C(t) = Tr +� +ˆµeff e−i ˆHt/¯h� ˆµeff|g⟩⟨g|⊗ ˆρB +� +ei ˆHt/¯h� +(9) +where ˆρ0 = |g⟩⟨g|⊗ ˆρB is a factorized initial total density ma- +trix with ˆρB = e−β ˆHB/TrB +� +e−β ˆHB +� +being the density matrix +of the thermal bath. + +3 +C. +Hierarchy of Pure States (HOPS) +The Hierarchy of Pure States (HOPS) is a formally exact +solution to the open-quantum system Hamiltonian presented +in Section II A. Using HOPS, time evolution of the system +reduced density matrix, starting from a separable initial state +( ˆΦ(0) = |ψ(0)⟩⟨ψ(0)|� ˆρB), can be described in terms of an +ensemble average (E[·]) over stochastic wave functions +ρ(t) = E +� +|ψ(⃗0)(t;z∗)⟩⟨ψ(⃗0)(t;z∗)| +⟨ψ(⃗0)(t;z∗)|ψ(⃗0)(t;z∗)⟩ +� +(10) +where z∗ is a stochastic trajectory with components asso- +ciated with individual thermal environments zn,t defined by +E[zn,t] = 0, E[zn,tzn,s] = 0, and E[z∗ +n,tzn,s] = αn(t − s). The +time-evolution of the system wave functions can be calculated +using the non-linear HOPS equation +¯h∂t|ψ(⃗k)(t;z∗)⟩ = +� +−i ˆHS −⃗k ·⃗γ +∑ +n +ˆLn(z∗ +n,t +ξn,t) +� +|ψ(⃗k)(t;z∗)⟩ ++ ∑ +n, jn +kjnγ jn ˆLn|ψ(⃗k−⃗e jn)(t;z∗)⟩ +− ∑ +n, jn +(gjn +γjn +)(ˆL† +n −⟨ˆL† +n⟩t)|ψ(⃗k+⃗e jn)(t;z∗)⟩, +(11) +where the physical wave function is given by |ψ(⃗0)(t;z∗)⟩, the +auxiliary wave functions are indexed by a vector ⃗k, ⃗γ is the +vector of correlation function exponents (γn), and +ξt,n = 1 +¯h +� t +0 dsα∗ +n(t −s)⟨L† +n⟩s +(12) +is a memory term that causes a drift in the effective noise. The +expectation value of the system-bath coupling operator is +⟨ˆL† +n⟩t = ⟨ψ(⃗0)(t;z∗)|ˆL† +n|ψ(⃗0)(t;z∗)⟩ +⟨ψ(⃗0)(t;z∗)|ψ(⃗0)(t;z∗)⟩ +. +(13) +We limit the hierarchy to a finite depth kmax using the trian- +gular truncation condition where we only include auxiliary +wave functions satisfying the condition {⃗k ∈ A : ∑i ki ≤ kmax}, +though other static filtering approaches have been explored +previously.44 +D. +Dyadic HOPS +Following the introduction of the pure state decomposition +by Hartmann et al.45 and the subsequent work of Chen et al.,41 +the dipole autocorrelation function that defines the linear ab- +sorption spectrum can be written as +C(t) = µtot +∑ +η∈{±1,±i} +η +2 E[⟨vη(t;z∗)| ˆµeff |vη(t;z∗)⟩] +(14) +where |vη(t;z∗)⟩ is the initial state +|vη⟩ = 1 +√ +2 +(|ψex⟩+η |g⟩) +(15) +FIG. 1. Algorithm for adHOPS +time-evolved in the full Hilbert space and η is the index of +the four pure state initial conditions. This correlation function +reduces (accounting for cancellation between η terms), to +C(t) = µ2 +totE[⟨ψex|ψex(t;z∗)⟩]eiEgt/¯h. +(16) +As written, Eq. (16) requires the system wave functions to +be propagated in the Hilbert space containing both the elec- +tronic ground and excited state. The dyadic HOPS equation, +introduced in Ref. 41, implicitly accounts for the influence of +the ground state but only time evolves the excited states. The +dyadic HOPS equation is equivalent to Eq. (11) except that the +expectation value of a system-bath coupling operator is given +by +⟨ˆL† +n⟩t = ⟨ψ(⃗0)(t;z∗)|ˆL† +n|ψ(⃗0)(t;z∗)⟩ +⟨ψ(⃗0)(t;z∗)|ψ(⃗0)(t;z∗)⟩+1 +, +(17) +and the dipole autocorrelation function becomes +C(t) = µ2 +totE +� +⟨ψex|ψex(t;z∗)⟩ +1 +2 (||ψex(t;z∗)||2 +1) +� +eiEgt/¯h. +(18) +III. +DYADIC ADAPTIVE HOPS +Here, we present an algorithm for constructing an adap- +tive basis during a dyadic HOPS trajectory that can sub- +stantially reduce computational requirements when simulat- +ing large systems. The algorithm presented here builds on +previous work implementing adaptive HOPS for a density ma- +trix calculation.43 The basic physical picture is that HOPS tra- +jectories are localized by the presence of their thermal envi- +ronments (i.e. bath Hamiltonians) and, as a result, the time- +evolution can be described by a local basis that moves with the +trajectory. To ensure the adaptive calculations retain the for- +mally exact structure of HOPS, the local basis must guarantee +a controllable error bound on the calculation. +Previously, Ref. 43 has demonstrated that it is possible to +construct an adaptive HOPS algorithm where computational +expense does not scale with system size for sufficiently large +aggregates (i.e. O(1) scaling). Fig. 1 presents a sketch of this +algorithm, where the local basis is constructed as a direct sum +(At +�St) of set of auxiliary wave functions (At, called the +‘auxiliary basis’) and the set of molecular states (St, called + +Update +Construct +Construct +p++S +At+dt +Yes +HOPS +Current +Calculate +Calculate +State +Update +e +Step? +No4 +the ‘state basis’). Every time point, we build a new auxil- +iary and state basis that ensures the difference between the +full derivative of all wave functions (represented by ∂Φ) and +the derivative constructed in the reduced basis ˜∂ ˜Φ is less than +a user specified threshold δ (i.e. ||∂Φ− ˜∂ ˜Φ|| < δ). For con- +venience, we split the user specified derivative error bound +(δ = +� +δ 2 +A +δ 2 +S ) into two components, the auxiliary bound +(δA) and the state bound (δS), which allow us to independently +control the precision we require when constructing the auxil- +iary and state bases, respectively. +Extending the algorithm to dyadic adaptive HOPS (Dad- +HOPS) requires a generalization to account for the fact that +the norm of the physical wave function (|ψ(⃗0)(t;z∗)⟩) is not +conserved during the time evolution of the trajectory. To en- +sure a consistent accuracy across the trajectory, we use relative +error bounds +� +∆A/S(t) = δA/S ·||ψ(⃗0)(t;z∗)|| +� +that evolves in +time with the norm of the physical wave function. +All calculations reported here are run with MesoHOPS +v1.2.14647 +IV. +MONTE CARLO SAMPLING LOCAL TRAJECTORIES +Here, we present the combinations of the DadHOPS algo- +rithm with a local representation of the dipole autocorrelation +function to enable efficient simulation of absorption spectra +for mesoscale molecular aggregates. For the DadHOPS algo- +rithm to be efficient the spatial extent of delocalization must +be substantially smaller than the full size of the molecular +aggregate. Simultaneously, implemented directly, the dyadic +HOPS expression for the dipole autocorrelation function has +an initial condition that can be arbitrarily delocalized. In the +following, first, we demonstrate that the dipole autocorrela- +tion function can be decomposed into local initial conditions. +Second, we demonstrate that the total correlation function can +be reproduced by a simultaneous Monte Carlo sampling of +noise trajectories (z∗ +t ) and local initial conditions. Finally, we +combine the initial state decomposition with the DadHOPS +algorithm to demonstrate a local construction of a linear ab- +sorption spectrum that can achieve size-invariant (i.e. O(1)) +scaling. +As a model system, we consider a linear chain of Npig +molecules with parallel dipole moments. The electronic cou- +pling is assumed to be nearest neighbor (V = −100 cm−1) and +we describe the bath correlation function as +αn(t) = iλΓ− +βν e−Γ+t/¯h + −iλΓ+ +βν +e−Γ−t/¯h +(19) +which is a high-temperature approximation to the spectral +density described in Ref. 48. For these calculations, we use +λ = 35 cm−1, Γ = 50 cm−1, ν = 10 cm−1, Γ± = Γ±iν, and +T = 295 K. +A. +Initial state decomposition +We can expand the collective dipole operator +ˆµeff = ∑ +d +Ad ˆσd +(20) +into a sum over local excitation operators acting on a cluster +of pigments (d) +ˆσd = ∑ +d∈d +µd ·εε +Ad +(|d⟩⟨g|+|g⟩⟨d|) +(21) +where +Ad = +� +∑ +d∈d +(µµd ·εε)2. +(22) +These equations assume that the pigment clusters represent +a partition of the set of all pigments (i.e. they are disjoint +and their union contains all pigments), though a more general +construction is described in Appendix A. Inserting this expan- +sion into the definition of the dipole autocorrelation function +(Eq. (9)) for the first interaction with the electric field we find +C(t) = ∑ +d +AdCd(t) +(23) +where +Cd(t) = Tr +� +ˆµeff e−i ˆHt/¯h� ˆσd|g⟩⟨g|⊗ ˆρB +� +ei ˆHt/¯h� +. +(24) +In Appendix A, we prove that the local correlation function +contribution can be calculated as +Cd(t) = µtot E +� +⟨ψex|ψd(t;z∗)⟩ +1 +2(||ψd(t;z∗)||2 +1) +� +eiEgt/¯h +(25) +where ψd(t;z∗) is the initial state +|ψd⟩ = ˆσd |g⟩ +(26) +time-evolved according to the dyadic HOPS equation. In the +special case where each cluster contains a single pigment (n), +this equation reduces to +Cn(t) = µtot E +� +⟨ψex|n(t;z∗)⟩ +1 +2 (||n(t;z∗)||2 +1) +� +eiEgt/¯h +(27) +where ˆσn = (|n⟩⟨g| + |g⟩⟨n|), An = µµn · εε, and |n(t;z∗)⟩ is +a single-site initial excitation time-evolved according to the +dyadic HOPS equations. +a. +Example: +To provide insight into the initial state de- +composition and to show that it reproduces the total absorp- +tion spectrum, we perform calculations on a four site (Npig = +4) homogeneous chain (En = 0). In this case, the total correla- +tion function can be decomposed into a sum of four single-site +initial conditions (Cn(t)) +C(t) = µ0(C1(t)+C2(t)+C3(t)+C4(t)) +(28) + +5 +FIG. 2. The initial state decomposition of a linear absorption spec- +trum. The total absorption spectrum (black) calculated for the 4-site +chain agrees with the sum (grey) of the edge (green) and inner (blue) +spectral contributions. Parameters: V = −100 cm−1, λ = 35 cm−1, +Γ = 50 cm−1, ν = 10 cm−1, T = 295 K, and kmax = 5. +and µ0 = µµn · εε. The symmetry of the Hamiltonian gives rise +to only two unique initial conditions +Cedge(t) = µ0(C1(t)+C4(t)) = 2µ0C1(t) +(29) +Cinner(t) = µ0(C2(t)+C3(t)) = 2µ0C2(t) +(30) +arising from the ‘edge’ and ‘inner’ excitation calculated using +Eq. (27). Fig. 2 plots the corresponding single-site spectral +contribution +Aedge/inner(ω) = Re +�� ∞ +0 Cedge/inner(t)e−iωtdt +� +for the edge (green line) and inner (blue line) initial condi- +tions. The reconstruction of the total spectrum as the sum +of the edge and inner contributions (grey line) agrees with the +spectrum arising from the dyadic calculation of the total corre- +lation function (black line) using the four site initial condition +(Eq. (16)). Note that individual spectral contributions (i.e. +Aedge/inner(ω)) are not themselves proper absorption spectra +since initial and final state are not the same. +B. +Monte Carlo Sampling Dyadic HOPS +The total correlation function can be calculated by Monte +Carlo sampling over single-site initial conditions. The initial +state decomposition of the full correlation function introduces +an independent trace over the bath for each initial condition +C(t) = µtot∑ +d +Ad ∑ +z∗ +1 +Nd,z +� +⟨ψex|ψd(t;z∗)⟩ +1 +2(||ψd(t;z∗)||2 +1) +� +eiEgt/¯h (31) +FIG. 3. +Statistical convergence of Monte Carlo sampling over the +initial state decomposition. (a) Bootstrapped spectral error for the +total absorption spectrum calculated with single site initial condi- +tions (black triangles, Nc = 1) either without (filled) or with (open) +static disorder. Similar statistical convergence is observed for the +edge (green circles) and inner (blue squares) single site initial con- +ditions. (b) Comparison of the statistical error for the total spectrum +normalized by the square-root of the number of pigments in each +cluster (√Nc) when using single site (Nc = 1, triangles), pair (Nc = 2, +circles), and all four site (Nc = 4, squares) initial conditions. Param- +eters: V = −100 cm−1, λ = 35 cm−1, Γ = 50 cm−1, ν = 10 cm−1, +T = 295 K, and kmax = 5. +where the sum over noise trajectories (z∗) is independent of +the sum over the ND initial conditions (d). If we assume un- +biased sampling with equal sized clusters, then the number of +noise trajectories per initial condition is Nd,z = Nens/ND and +the correlation function can be calculated as +C(t) = µtot +ND +Nens ∑ +(d,z∗) +Ad +� +⟨ψex|ψd(t;z∗)⟩ +1 +2(||ψd(t;z∗)||2 +1) +� +eiEgt/¯h, (32) +where we Monte Carlo sample Nens pairs (d,z∗) to calculate +the full correlation function. +a. +Example: +Here, we perform calculations using the +same 4 site linear chain considered above. Fig. 3a compares +the bootstrapped spectral error for the total spectrum using Eq. +(32) with single site initial conditions (black filled triangles) +error = +1 +� ωmax +ωmin Aref(ω)dω +� ωmax +ωmin +��A (ω)−Aref(ω) +��dω (33) +as a function of the number of trajectories sampled from the +combined (n,z∗) space. We find that the statistical conver- +gence of the total correlation function is equivalent to that of +the edge (green circles) and inner (blue squares) spectral com- +ponents calculated independently using Eq. (27). + +Inner+Edge +Total +150 +Inner +Edge +125 +100 +A(w) +75 +50 +25 +0 +400 +-200 +0 +200 +400 +Energy (cm-1)(a) +Log1o(Mean Error) +100 +101 +102 +103 +Number of Trajectories +(b) +Log1o(VNc Error) +V +1 +100 +101 +102 +103 +Number of Trajectories6 +One advantage of Monte Carlo sampling is the trivial incor- +poration of an additional sampling over disorder in the system +Hamiltonian ( ˆHs). Fig. 3a shows the statistical convergence of +the total spectrum calculated with single-site initial conditions +in the presence of Gaussian distributed disorder on site ener- +gies (black open triangles) is almost unchanged compared to +the homogeneous case (black filled triangles). For the disor- +dered calculations, site energies form a Gaussian distribution +with mean value of zero and a standard deviation (SD) of 100 +cm−1. +Finally, we find that the statistical error of the Monte Carlo +sample for a given number of trajectories is inversely propor- +tional to the size of the clusters (Nc = Npig/ND) used for the +initial state decomposition. In Fig. 3b, we compare the nor- +malized statistical error (√Nc· Error) as a function of the num- +ber of independent trajectories (Ntraj) when using different ini- +tial conditions: single site (triangles, Nc = 1), pairs (circles, +Nc = 2), and all four sites (squares, Nc = 4). The equiva- +lence of these three results shows that increasing the size of +the clusters decreases the number of trajectories required to +reach a given statistical error. Thus, when using an equation +of motion where the delocalization extent of the initial state +does not change the computational expense, the initial state +decomposition provides no advantage. +C. +Monte Carlo Sampling Dyadic Adaptive HOPS +The local correlation functions arising from the initial state +decomposition can be efficiently simulated using DadHOPS. +Fig. 4a shows how, for a 4-site linear chain, the error of the +edge component decreases with the derivative error bound δA. +Here, we consider DadHOPS calculations converged when the +adaptive error is lower than the statistical error for the asso- +ciated ensemble of 104 trajectories (Fig. 4a, grey horizontal +line). In Fig. 4b the edge spectral contribution calculated with +converged DadHOPS parameters (thin line, δA = 10−3) re- +produces the corresponding dyadic HOPS calculation (thick +transparent line). +The DadHOPS algorithm combined with the initial state de- +composition allows us to calculate much larger system sizes +then would be possible with dyadic HOPS directly. As a first +demonstration of scaling we consider a 12-site linear chain +where the basis for the full HOPS calculation has more than +106 elements. Fig. 4c plots the total absorption spectrum cal- +culated with DadHOPS (thick line, δA = 10−3, δS = 0) using +the full initial condition (|ψ0⟩ = +1 +√ +12 ∑12 +n=1 |n⟩). We can re- +produce this spectrum by Monte Carlo sampling over single- +site initial conditions and incorporating adaptivity in both the +auxiliary and state basis (Fig. 4c, thin line, δA = δS = 10−3). +Combining adaptivity in both the auxiliary and state basis +with localized initial states can achieve size-invariant scaling +for sufficiently large aggregates. Fig. 4d plots the number of +state basis elements (bottom) and the number auxiliary wave +functions (top) required for both the full dyadic HOPS (thick +grey lines) and the DadHOPS algorithm (thin green lines) as +a function of the number of sites in the linear chain. +Here, we have demonstrated that the advantage of the local- +FIG. 4. +Comparing dyadic HOPS and DadHOPS for N-site lin- +ear chains. (a) Mean error of the edge spectrum with respect to the +auxiliary error bound (δA) for a 4-site chain. The solid grey line rep- +resents the statistical error. (b) The edge spectral component for a +4-site chain calculated with converged DadHOPS parameters (thin +line) and corresponding dyadic HOPS calculation (thick transparent +line). (c) Total absorption spectrum for a 12-site chain calculated +using DadHOPS with (thin line) and without state adaptivity (thick +line). +(d) Average number of auxiliary wave functions (top) and +state basis elements (bottom) for linear chains of different lengths re- +quired for calculations using DadHOPS (green) compared to dyadic +HOPS (thick grey). Parameters: V = −100 cm−1, λ = 35 cm−1, +Γ = 50 cm−1, ν = 10 cm−1, T = 295 K, and kmax = 5. +ized wave functions introduced by the initial state decomposi- +tion is realized by combining it with the DadHOPS equation- +of-motion that leverages locality to reduce computational ex- +pense. +V. +APPLICATION +A. +Photosystem I (PSI) +Photosystem I (PSI) is a pigment-protein complex contain- +ing 96 chlorophyll per monomer and usually found as a trimer +in higher plants and algae (Fig 5a).50 The complicated spa- +tial arrangement of chlorophyll make this an ideal test sys- +tem for exploring the behavior of the initial state decompo- +sition with respect to different definitions of clusters. Here, +we use a previously reported Hamiltonian51 where chloro- +phyll excitation energies (En) and excitonic couplings (Vn,m) +were calculated using time-dependent density function the- +ory (TDDFT) evaluated by Gaussian1652 with the CAM- +B3LYP density functional53 and the 6-31G∗ basis set;54 this +approach is known to provide quality descriptions of both + +(a) +(b) +30 - +4 site +Statistical +2 +Error +(3) +20 +10 - +Edge +0: +10-4 +10-3 +500 +0 +500 +Aux. Error Bound (SA) +Energy (cm-1) +(c) +(d) +10° +3 - +ler +12 site +103 +A(w) /100 +2 +101 +20 +state +Total +10 +N +0 +0 +500 +0 +500 +100 +102 +Energy (cm-1) +Numb. of Pigments7 +FIG. 5. +Simulating linear absorption for Photosystem I (PSI). (a) +A PSI core complex trimer is shown with a single monomer high- +lighted; the simulations are for one monomer containing 96 pig- +ments, using cyclic boundary conditions to represent the trimer. (b) +A chlorophyll a molecule with the phytyl tail truncated for clarity +is show along with the Qy transition dipole vector. (c) Comparison +of absorption spectrum simulated using HEOM (grey)49 and Dyadic +HOPS (green). (d) Bootstrapped spectral error for randomly assign- +ing pigments to clusters (black solid circles) and using clusters de- +fined by strong coupling (green open circles). Parameters: λ = 35 +cm−1, Γ = 50 cm−1, T = 300 K, Γmark = 500 cm−1, SD= 150 cm−1, +and kmax = 1. +vibrational and vibronic processes apparent from the high- +resolution absorption and emission spectra of chlorophyll- +type molecules.55 The Hamiltonian is provided in the Sup- +porting Information. The system-bath coupling is described +by a Drude-Lorentz spectral density +Jn(ω) = 2λΓω +ω2 +Γ2 +(34) +and the corresponding bath correlation function is given in a +high-temperature approximation as +αn(t) = (2λ/β −iλΓ)e−Γt/¯h +iλΓe−Γmarkt/¯h +(35) +where the second exponential is included to ensure the imagi- +nary component of αn(t) is continuous at time t = 0. To agree +with previous HEOM calculations,49 we introduce a Gaussian +distributed static disorder on the chlorophyll site energies with +a standard deviation (SD) of 150 cm−1, we approximate the +true chlorophyll Qy transition dipole moment (Fig. 5b) by +the vector defined by the position of the NA and NC nitro- +gen atoms (IUPAC notation for chlorophyll a),56 and we set +the spectral density parameters to be: λ = 35 cm−1, Γ = 50 +cm−1, T = 300 K, and Γmark = 500 cm−1. +The varied alignments of the chlorophyll Qy dipole mo- +ments throughout PSI introduces an orientation dependence +into the linear absorption spectrum. For an isotropic distri- +bution of PSI complexes, we calculate the linear absorption +spectrum from the average of the dipole autocorrelation func- +tion for the x, y, and z polarized electric fields (εε) +C(εε)(t) = µtot,εε +ND +∑ +d=1 +�Ad,εε +Nd,z +� +∑ +z∗ +� +⟨ψex|ψd(t;z∗)⟩ +1 +2(||ψd(t;z∗)||2 +1) +� +eiEgt/¯h. +(36) +Fig. 5c shows that, as expected, the HOPS calculations (green +line) reproduce previous HEOM calculations from Ref. 49 +(grey line). +To what extent does the specific choice of clusters matter +for a given cluster size? In the case of a linear chain, the +choice of clusters to be sets of adjacent pigments may appear +obvious. There is no equivalently obvious choice for the het- +erogeneously coupled pigments in PSI. Fig. 5d compares the +statistical convergence of the linear absorption spectrum for +two different definitions of clusters: the first randomly assigns +four pigments to each cluster (black solid circles), while the +second constructs clusters of 4 strongly interacting pigments +(green open circles) using an algorithm described in Appendix +B. There is no appreciable difference in the convergence be- +tween randomly assigning pigments to clusters or using clus- +ters defined by strong coupling. Combined with the results +presented in Fig. 3b showing square-root-scaling of error with +cluster size, our calculations suggest that using a cluster initial +condition only acts to increase the effective number of com- +bined (n,z∗) pairs that are being sampled. +B. +Perylene bis-imide (PBI) +Perylene bis-imide (PBI) is a family of pigments that +can form both J- and H-type aggregates in solution.57–60 J- +aggregates of PBI-1 have been the object of particular study +and show a strong vibronic progression in their linear ab- +sorption sprectra. +Previous time-dependent density func- +tional theory (TDDFT) calculations of PBI-1 have character- +ized electronic coupling between adjacent monomers (V ≈ +−500 cm−1) and 28 intramolecular vibrational modes with +appreciable Huang-Rhys factors.61 This vibrational struc- +ture has served as a starting point for different calcula- +tions of linear absorption spectra for PBI-1 aggregates us- +ing techniques such as multiconfiguration time-dependent +Hartree (MCTDH),61 multilayer MCTDH (ML-MCTDH),62 +and time-dependent density matrix renormalization group the- +ory (TD-DMRG).63 Calculations of the exciton dynamics in +aggregates ranging in size from two to 25 monomer units have +also been reported with varying degrees of complexity in their +description of the molecular vibrations.60,64,65 +We have constructed a minimal bath correlation function +composed of two modes to describe the vibrational environ- +ment of PBI. We include a vibration with a central frequency +near 1500 cm−1 to account for the group of strongly-coupled +vibrations ranging from 1370-1630 cm−1 previously reported +from TDDFT calculations.61 We also include a low-frequency + +(a) +(b) +(c) +(d) +1.0 - +Log1o(Mean Error) +Spectrum +0.5 +Abs. +0.0: +2 +14000 +15000 +101 +102 +103 +104 +Numb. of Traj +Energy (cm-1)8 +Mode +g (cm−2) +γ (cm−1) +1 +1.2×105 +50+170i +2 +1.6×106 +100+1550i +TABLE I. The exponential parameters describing the vibrational cor- +relation function of PBI. +FIG. 6. Simulating linear absorption for perylene bis-imide (PBI) J- +aggregates. (a) Compares the experimental absorption spectra from +dilute solution (0.0016 mM, grey)]59 with calculations broadened by +either SD = 300 cm−1 (green) or 400 cm−1 (blue). (b) Compares +the experimental absorption spectra from concentrated solution (0.16 +mM, grey)59 with a calculated trimer spectra (green). (c) Compares +the lineshape calculated for aggregates containing one (black), two +(green), three (blue), seven (cyan), and 1000 pigments (grey). (d) +Plots the change in the 0-0 transition peak shift (∆E, top) and the +ratio of the 0-1 peak intensity to the 0-0 peak intensity (I0,1/I0,0, +bottom) as a function of the number of molecules in the aggregate +(Npig). Parameters: kmax = 6. +vibration to account for the broad dissipative environment. +The final parameters (Table I) were selected to agree with the +available experimental data. +Our minimal description of the system-bath correlation +function is capable of reproducing the major features of both +the experimentally measured monomer PBI spectrum (Fig. +6a) and the aggregate spectrum (Fig. 6b). The monomer spec- +trum shows a broad lineshape and is better reproduced when +the Gaussian disorder of the vertical excitation energies has +a standard deviation of SD = 400 cm−1 (blue line) compared +with the SD = 300 cm−1 (green line) we use for the aggregate +spectrum. We can further improve the agreement between the +simulated and experimental monomer spectra by introducing +additional moderate frequency modes (data not shown), but +this additional complexity was not required for the reproduc- +tion of the experimental aggregate spectrum. We smooth the +aggregate spectra reported here by weighting the time-domain +dipole auto-correlation function with a cosine appodization +window66 +Θ(t) = +� +cos( π +2t/tmax) +t ≤ tmax +0 +t > tmax +(37) +that goes to zero at the last time point of the calculated tra- +jectory (tmax). The use of this window function suppresses +noise in the calculated spectra that arises from the combina- +tion of zero-padding66 and the incomplete cancellation of the +correlation function at long times due to finite sampling of the +trajectory ensembles. +The advantage of the adaptive dyadic HOPS formalism is +the ability to calculate even very large molecular aggregates +efficiently. The size of a molecular aggregate is often impor- +tant for capturing the influence of exciton delocalization on +both the 0-0 peak position and the relative magnitude of vi- +bronic side-bands.67 Fig. 6c compares the monomer (black +line), dimer (green line), trimer (blue line), and heptamer +(cyan line) lineshapes with that calculated for a 1000-site lin- +ear chain (grey line). The magnitude of the vibronic side- +band decreases rapidly from the monomer to trimer and then +changes slow with increasing chain length. A similar trend is +seen for the red-shift of the 0-0 peak. Fig. 6d quantifies this +effect by plotting the central position of the 0-0 peak (∆E, top) +and the ratio of the intensity of the 0-1 peak to the intensity +of the 0-0 peak (I0,1/I0,0, bottom) as a function of the num- +ber of pigments in the aggregate. In both cases, the grey lines +represents the asymptotic limit of a 1000-site linear chain. +Finally, let us consider the relationship between aggregate +size, locality, and the onset of size-invariant scaling in the +DadHOPS equation-of-motion. When the adaptive basis size +stops increasing with the length of the aggregate, the extent +of exciton delocalization is sufficiently small that most tra- +jectories will never sample the edge of the aggregate. This +implies that for aggregates sufficiently large that the basis is +size-invariant the spectroscopic signatures should also be size- +invariant. Fig. 7 shows that by Npig = 10 the average size of +the adaptive site basis begins to plateau before reaching a size- +invariant value of 13 for chains of 100 PBI molecules. This is +consistent with the convergence of the spectral features which +are close to the asymptotic values at Npig = 7 (Fig. 6d) and +the expectation that complete size-invariance will occur when +only a relatively small set of trajectories will sample pigments +on the edge of the chain. We suggest, then, that DadHOPS can +be thought of as introducing a dynamic separation of length- +scales where the basis size required to capture the spectral fea- +tures is solved on-the-fly rather than being asserted a priori. +As a result, DadHOPS is advantageous for simulating realistic +molecular aggregates, particularly in the presence of structural +disorder, where separations of length scales can be obscure. + +(a) +(b) +Spectrum +Spectrum +0.5 +0.5 +Abs. +Abs. +0.0 +0.0 +17500 +20000 +17500 +20000 +Energy (cm-1) +Energy (cm-1) +(c) +(p) +T- +-1000 +(cm +Spectrum +E +2000 +0.5 +lo,1 /lo,0 +Abs. +0.50 +0.25 +0.0 +-2500 +0 +2500 +3 +1 +5 +Energy (cm-1) +Number of Pigments9 +FIG. 7. +Scaling of DadHOPS basis size for PBI aggregate. Aver- +age number of auxiliary wave functions (top) and states (bottom) for +different lengths of PBI chains required for calculations using Dad- +HOPS (green) compared to dyadic HOPS (grey). +VI. +CONCLUSIONS +Here we have presented a new algorithm that combines the +dyadic adaptive HOPS (DadHOPS) equations with a initial +state decomposition and is capable of calculating optical lin- +ear absorption spectra for mesoscale molecular aggregates. +Our approach introduce an dynamic separation of length- +scales by adaptively constructing a reduced basis to describe +the time-dependence of local contributions to the dipole auto- +correlation function. The adaptive basis construction is capa- +ble of achieving size-invariant scaling (i.e. O(1)) with the +number of molecules for sufficiently large aggregates. We +have applied the initial state decomposition to the 96 chloro- +phyll photosystem I core complex and found that the specific +choice of pigment clusters is not essential to the statistical +convergence of the calculations. We calculated 1000 molecule +J-aggregate of perylene bis-imide (PBI) with DadHOPS and +characterized how the absorption spectra changes with ag- +gregate size. Because locality cannot reduce the computa- +tional expense of describing individual pigments, DadHOPS +remains limited to calculations where each thermal reservoir +has a small number of exponential modes. In the future, it may +be possible to describe more complex thermal environment by +incorporating techniques such as tensor contraction.68 +ACKNOWLEDGMENTS +AE acknowledges support from the DFG via a Heisenberg +fellowship (Grant No EI 872/10-1). +TG, EJT, and DIGB +acknowledge support from the Robert A. Welch Foundation +(Grant N-2026-20200401) and start-up funds from Southern +Methodist University. DIGB acknowledges the US National +Science Foundation CAREER Award under Grant No. CHE- +2145358. +DATA AVAILABILITY STATEMENT +The data that support the findings of this study are available +from the corresponding author upon reasonable request. +Appendix A: Dyadic HOPS with Generalized Initial State +Decomposition +The dipole autocorrelation function +C(t) = Tr +� +ˆµeff e−i ˆHt/¯h� ˆµeff|g⟩⟨g|⊗ ˆρB +� +ei ˆHt/¯h� +(A1) +depends on the time-evolution of the first-order density matrix +ˆµeff |g⟩⟨g| arising from one interaction with the electric field. +The Non-Markovian Quantum State Diffusion (NMQSD) for- +malism that gives rise to the HOPS equation, however, can +only time-evolve pure state density matrices. The pure state +decomposition introduced in Ref. 45 rewrites the initial first- +order density matrix into a sum of pure state density matrices +which can then be treated within the NMQSD (or HOPS) for- +malism. Here, we use the pure state decomposition to rewrite +the dipole autocorrelation function into a sum of locally ex- +cited correlation functions that can be efficiently simulated +using the DadHOPS equation-of-motion. +We begin by decomposing the collective transition dipole +operator +ˆµeff = +N +∑ +n +(µµn ·εε)|n⟩⟨g|+h.c., +(A2) +into an arbitrary (finite) set of interaction operators ( ˆσa) and +real-valued weights (Aa) defined to ensure that +ˆµeff = ∑ +a +Aa ˆσa +(A3) +and |ψa⟩ = ˆσa |g⟩ is a normalized wave function. The dipole +autocorrelation function can then be rewritten as +C(t) = ∑ +a +AaCa(t) +(A4) +where +Ca(t) = Tr +� +ˆµeff e−i ˆHt/¯h� ˆσa|g⟩⟨g|⊗ ˆρB +� +ei ˆHt/¯h� +. +(A5) +The decomposition expressed here is generic and does not re- +quire - for example - that the interaction operators be orthog- +onal. We will proceed in our derivation with these generic in- +teraction operators and then introduce the specific expressions +for two convenient special cases at the end. + +106 +Nhier +104 +102 +100 +100 +101 +102 +102 +Istate +101 +N +100 +100 +101 +102 +Number of Pigments10 +We can then define a set of pure states +|vη,a⟩ = 1 +√ +2 +� +η |g⟩+|ψa⟩ +� +(A6) +which can be used to reconstruct the initial first-order density +matrix associated with the correlation function for each exci- +tation operator +ρa(t = 0) = ˆσa |g⟩⟨g| = +∑ +η∈{±1,±i} +η +2 |vη,a⟩⟨vη,a|. +(A7) +To prove the second equality, we expand the summand +η +2 |vη,a⟩⟨vη,a| = 1 +4 +� +ηηη∗ |g⟩⟨g|+ηη |g⟩⟨ψa| ++ηη∗ |ψa⟩⟨g|+η |ψa⟩⟨ψa| +� +(A8) +and note that +∑ +η∈{±1,±i} +ηηη∗ = 0, +∑ +η∈{±1,±i} +ηη = 0 +∑ +η∈{±1,±i} +ηη∗ = 4, +∑ +η∈{±1,±i} +η = 0 +which reduces the right-hand side of Eq. (A7) to +∑ +η∈{±1,±i} +η +2 |vη,a⟩⟨vη,a| = |ψa⟩⟨g| = ˆσa |g⟩⟨g|. +(A9) +We can now proceed to write down the standard dyadic +HOPS expressions for the individual contributions to the +dipole autocorrelation function. We begin by rewriting the +components of the dipole autocorrelation function as +Ca(t) = +∑ +η∈{±1,±i} +Tr +� +ˆµeff e−i ˆHt/¯h�η +2 |vη,a⟩⟨vη,a|⊗ ˆρB +� +ei ˆHt/¯h� +(A10) +which is equal to +Ca(t) = +∑ +η∈{±1,±i} +E +� η +2 ⟨vη,a(t,z∗)| ˆµeff|vη,a(t,z∗)⟩ +⟨vη,a(t,z∗)|vη,a(t,z∗)⟩ +� +(A11) +where |vη,a⟩ is time-evolved according to the non-linear +HOPS equation (Eq. (11)). +As written, the time-evolution +of the pure state |vη,a⟩ is performed using a system Hamilto- +nian ( ˆHS) that contains both the ground and excited electronic +states. However, since the system Hamiltonian does not cou- +ple the electronic ground and excited states, the time evolution +of the two components of |vη,a⟩ can be decoupled giving +|vη,a(t;z∗)⟩ = 1 +√ +2 +� +η |g⟩e−iEgt/¯h +|ψa(t;z∗)⟩ +� +(A12) +where |ψa(t;z∗)⟩ is propagated using the non-linear HOPS +equation where ˆHS includes only the first excitation manifold +and the expectation value of the system-bath coupling opera- +tor in Eq. (17) is redefined as +⟨ˆL† +n⟩t = ⟨ψ(⃗0)(t;z∗)|ˆL† +n|ψ(⃗0)(t;z∗)⟩ +⟨ψ(⃗0)(t;z∗)|ψ(⃗0)(t;z∗)⟩+1 +(A13) +where the change in the denominator arises from the presence +of the ground-state component of the wave function. +We can use Eq. (A12) and the definition of the collective +dipole operator ( ˆµeff |g⟩ = µtot |ψex⟩, Eq. (7)) to rewrite the +numerator of Eq. (A11) as +η +2 ⟨vη,a(t;z∗)| ˆµeff|vη,a(t;z∗)⟩ = µtot +ηη +4 ⟨ψa(t;z∗)|ψex⟩e−iEgt/¯h ++ µtot +|η|2 +4 ⟨ψex|ψa(t;z∗)⟩eiEgt/¯h +and the denominator as +⟨vη,a(t;z∗)|vη,a(t;z∗)⟩ = 1 +2 +� +1+||ψa(t;z∗)||2� +. +(A14) +Noting that the denominator does not depend on η, the cancel- +lation in the numerator due to summation over η ∈ {±1,±i} +gives +Ca(t) = µtotE +� +⟨ψex|ψa(t;z∗)⟩ +1 +2 (||ψa(t,z∗)||2 +1) +� +eiEgt/¯h. +(A15) +This equation is equivalent to the standard dyadic HOPS ex- +pression, but now propagating a component of the total cor- +relation function defined by a decomposition of the collective +transition dipole operator into a sum of excitation operators. +a. +Cluster Decomposition +One simple decomposition of the effective collective tran- +sition dipole operator is to expand in clusters of pigments +ˆµeff = +N +∑ +n +(µµn ·εε)|n⟩⟨g|+h.c. +(A16) +=∑ +d +Ad ∑ +d∈d +µµd ·εε +Ad +(|d⟩⟨g|+|g⟩⟨d|) +(A17) +=∑ +d +Ad ˆσd +(A18) +where +ˆσd = ∑ +d∈d +µd ·εε +Ad +(|d⟩⟨g|+h.c.) +(A19) +and +Ad = +� +∑ +d∈d +(µµd ·εε)2. +(A20) +The total correlation function can then be decomposed as +C(t) = µtot∑ +d +AdE +� +⟨ψex|ψd(t;z∗)⟩ +1 +2 (||ψd(t;z∗)||2 +1) +� +eiEgt/¯h +(A21) +where |ψd(t;z∗)⟩ is the state |ψd⟩ = ˆσd |g⟩ time-evolved ac- +cording to the dyadic HOPS equation. + +11 +b. +Pigment Decomposition +If each cluster is composed of a single pigment, then +Eq. (A21) simplifies to +C(t) = µtot∑ +n +(µµn ·εε)E +� +⟨ψex|n(t;z∗)⟩ +1 +2 (||n(t;z∗)||2 +1) +� +eiEgt/¯h +(A22) +where |n(t;z∗)⟩ is the single-pigment excitation state |n⟩ time- +evolved according to the dyadic HOPS equation. +Appendix B: Defining Strongly Interacting Clusters +To explore the influence of cluster definitions, we con- +structed clusters between four strongly coupled pigments in +photosystem I (PSI). We use an iterative, three-step algorithm +designed to ensure that strongly interacting pigments are pref- +erentially included inside the same cluster. +1. Step 1: +Locate the largest coupling element (Vi, j) +among the pigments not yet assigned to a cluster. Pig- +ments (i, j) for the nucleus of a new cluster. +2. Step 2: Locate the largest coupling element to either +pigment in the new cluster to another pigment not yet +assigned to a cluster (Vm, j or Vi,m). Add this pigment to +the new cluster: (i, j,m). +3. Step 3: Locate the largest coupling element involving a +pigment in the new cluster with another pigment (n) not +yet assigned to a cluster. Add this pigment to the new +cluster: (i, j,m,n). +4. End Condition: If any pigments remain unassigned +then return to step 1 and define a new cluster. +This algorithm does not guarantee that all strong couplings +are contained within a cluster, but it does ensure that clusters +nucleate around strong coupling elements. +REFERENCES +1H. van Amerongen, L. Valk¯unas, and R. van Grondelle, Photosynthetic Ex- +citons (World Scientific, 2000). +2V. May and O. Kühn, Charge and Energy Transfer Dynamics in Molecular +Systems: A Theoretical Introduction (Wiley, 2000). +3D. Abramavicius, B. Palmieri, D. V. Voronine, F. Šanda, and S. Mukamel, +Chem. Rev. 109, 2350 (2009). +4L. Chen, P. Shenai, F. Zheng, A. Somoza, and Y. Zhao, Molecules 20, 15224 +(2015). +5S. Mukamel, Principles of Nonlinear Optical Spectroscopy, Oxford series +in optical and imaging sciences (Oxford University Press, 1995). +6M. F. Gelin, L. Chen, and W. Domcke, Chem. Rev. 122, 17339 (2022). +7M. H. Beck, A. Jäckle, G. A. Worth, and H. D. Meyer, Phys. Rep. 324, 1 +(2000). +8I. Burghardt, M. Nest, and G. A. Worth, J. Chem. Phys. 119, 5364 (2003). +9G. W. Richings, I. Polyak, K. E. Spinlove, G. A. Worth, I. Burghardt, and +B. Lasorne, Int. Rev. Phys. Chem. 34, 269 (2015). +10H. Wang and M. Thoss, J. Chem. Phys. 119, 1289 (2003). +11H. Wang, J. Phys. Chem. A 119, 7951 (2015). +12U. Manthe, J. Chem. Phys. 128, 164116 (2008). +13O. Vendrell and H.-D. Meyer, J. Chem. Phys. 134, 044135 (2011). +14D. V. Shalashilin, J. Chem. Phys. 130, 244101 (2009). +15D. V. Shalashilin, J. Chem. Phys. 132, 244111 (2010). +16L. Chen, M. F. Gelin, and D. V. Shalashilin, J. Chem. Phys. 151, 244116 +(2019). +17M. Ben-Nun and T. J. Martínez, J. Chem. Phys. 108, 7244 (1998). +18M. Ben-Nun, J. Quenneville, and T. J. Martínez, J. Phys. Chem. A 104, +5161 (2000). +19Y. Tanimura, J. Phys. Soc. Japan 75, 082001 (2006). +20Y. Tanimura, J. Chem. Phys. 153, 020901 (2020). +21N. Makri, Chem. Phys. Lett. 193, 435 (1992). +22N. Makri, Journal of Mathematical Physics 36, 2430 (1995). +23M. A. Cazalilla and J. B. Marston, Phys. Rev. Lett. 88, 256403 (2002). +24S. R. White and A. E. Feiguin, Phys. Rev. Lett. 93, 076401 (2004). +25J. Prior, A. W. Chin, S. F. Huelga, and M. B. Plenio, Physical review letters +105, 050404 (2010). +26D. Tamascelli, A. Smirne, J. Lim, S. F. Huelga, and M. B. Plenio, Phys. +Rev. Lett. 123, 090402 (2019). +27L. Chen, R. Borrelli, and Y. Zhao, J. Phys. Chem. A 121, 8757–8770 +(2017). +28L. Chen, M. Gelin, and Y. Zhao, Chem. Phys. 515, 108 (2018). +29N. Makri, J. Chem. Phys. 148, 101101 (2018). +30N. Makri, J. Chem. Phys. 149, 214108 (2018). +31R. Borrelli and S. Dolgov, J. Phys. Chem. B 125, 5397–5407 (2021). +32Y. Yan, M. Xu, T. Li, and Q. Shi, J. Chem. Phys. 154, 194104 (2021). +33A. D. Somoza, O. Marty, J. Lim, S. F. Huelga, and M. B. Plenio, Phys. Rev. +Lett. 123, 100502 (2019). +34D. Tamascelli, A. Smirne, J. Lim, S. F. Huelga, and M. B. Plenio, Phys. +Rev. Lett. 123, 090402 (2019). +35A. Strathearn, P. Kirton, D. Kilda, J. Keeling, and B. W. Lovett, Nat. Com- +mun. 9, 3322 (2018). +36L. Diósi and W. T. Strunz, Phys. Lett. 235, 569 (1997). +37L. Diósi, N. Gisin, and W. T. Strunz, Phys. Rev. A 58, 1699 (1998). +38D. Suess, A. Eisfeld, and W. T. Strunz, Phys. Rev. Lett. 113, 150403 (2014). +39R. Hartmann and W. T. Strunz, J. Chem. Theory Comput. 13, 5834 (2017). +40G. Ritschel, D. Suess, S. Möbius, W. T. Strunz, and A. Eisfeld, J. Chem. +Phys. 142, 034115 (2015). +41L. Chen, D. I. Bennett, and A. Eisfeld, J. Chem. Phys. 156, 124109 (2022). +42L. Chen, D. I. G. Bennett, and A. Eisfeld, J. Chem. Phys. 157, 114104 +(2022). +43L. Varvelo, J. K. Lynd, and D. I. Bennett, Chem. Sci. 12, 9704 (2021). +44P. P. Zhang, C. D. Bentley, and A. Eisfeld, J. Chem. Phys. 148, 134103 +(2018). +45R. Hartmann and W. T. Strunz, J. Phys. Chem. A 125, 7066 (2021). +46L. Varvelo, J. K. Lynd, B. Citty, and D. I. G. B. Raccah, Mesohops v1.2.1 +(2023), URL https://zenodo.org/record/7504694. +47Note1, the most recent version of MesoHOPS can be found here: +https://github.com/MesoscienceLab/mesohops. +48A. Ishizaki, J. Phys. Soc. Japan 89, 015001 (2020). +49T. Kramer, M. Noack, J. R. Reimers, A. Reinefeld, M. Rodríguez, and +S. Yin, Chem. Phys. 515, 262 (2018). +50P. Jordan, P. Fromme, H. T. Witt, O. Klukas, W. Saenger, and N. Krauß, +Nature 411, 909 (2001). +51S. Yin, M. G. Dahlbom, P. J. Canfield, N. S. Hush, R. Kobayashi, and J. R. +Reimers, J. Phys. Chem. B 111, 9923 (2007). +52M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, +J. R. Cheeseman, G. Scalmani, V. Barone, G. A. Petersson, H. Nakatsuji, +et al., Gaussian~16 Revision C.01 (2016), gaussian Inc. Wallingford CT. +53T. Yanai, D. P. Tew, and N. C. Handy, Chem. Phys. Lett. 393, 51 (2004). +54W. J. Hehre, R. Ditchfield, and J. A. Pople, J. Chem. Phys. 56, 2257 (1972). +55J. R. Reimers, M. Rätsep, J. M. Linnanto, and A. Freiberg, Proc. Estonian +Acad. Sci. 71, 127 (2022). +56R. Blankenship, Molecular Mechanisms of Photosynthesis (Wiley, 2021), +3rd ed. +57V. Dehm, Z. Chen, U. Baumeister, P. Prins, L. D. A. Siebbeles, and +F. Würthner, Org. Lett. 9, 1085 (2007). +58Z. Chen, V. Stepanenko, V. Dehm, P. Prins, L. Siebbeles, J. Seibt, P. Mar- +quetand, V. Engel, and F. Würthner, Chem. Eur. J. 13, 436 (2007). +59H. Marciniak, X.-Q. Li, F. Würthner, and S. Lochbrunner, J. Phys. Chem. + +12 +A 115, 648 (2010). +60M. Schröter, S. Ivanov, J. Schulze, S. Polyutov, Y. Yan, T. Pullerits, and +O. Kühn, Phys. Rep. 567, 1 (2015). +61D. Ambrosek, A. Köhn, J. Schulze, and O. Kühn, J. Phys. Chem. A 116, +11451 (2012). +62J. Seibt, T. Winkler, K. Renziehausen, V. Dehm, F. Würthner, H.-D. Meyer, +and V. Engel, J. Phys. Chem. A 113, 13475 (2009). +63J. Ren, Z. Shuai, and G. K.-L. Chan, J. Chem. Theory Comput. 14, 5027 +(2018). +64S. Kundu and N. Makri, J. Chem. Phys. 154, 114301 (2021). +65S. Kundu and N. Makri, Phys. Chem. Chem. Phys. 23, 15503 (2021). +66J. Hoch and A. Stern, NMR Data Processing (Wiley, 1996). +67R. Ghosh and F. C. Spano, Acc. Chem. Res. 53, 2201 (2020). +68X. Gao, J. Ren, A. Eisfeld, and Z. Shuai, Phys. Rev. A 105, L030202 1 +(2022). + diff --git a/xdE2T4oBgHgl3EQfLwbX/content/tmp_files/load_file.txt b/xdE2T4oBgHgl3EQfLwbX/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..4ba3edf65cb61d37ea60de0eca1ecc9a1d854bd9 --- /dev/null +++ b/xdE2T4oBgHgl3EQfLwbX/content/tmp_files/load_file.txt @@ -0,0 +1,956 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf,len=955 +page_content='Simulating optical linear absorption for mesoscale molecular aggregates: an adaptive hierarchy of pure states approach Tarun Gera,1 Lipeng Chen,2 Alex Eisfeld,2 Jeffrey R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Reimers,3, 4 Elliot J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Taffet,1 and Doran I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Raccah1 1)Department of Chemistry, Southern Methodist University, PO Box 750314, Dallas, TX, USA 2)Max Planck Institute for the Physics of Complex Systems, Nöthnitzer Str 38, Dresden, Germany 3)International Centre for Quantum and Molecular Structures and the School of Physics, Shanghai University, 200444, Shanghai, China 4)School of Mathematical and Physical Sciences, University of technology Sydney, NSW 2007, Australia (*Electronic mail: doranb@smu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='edu) (Dated: 11 January 2023) In this paper, we present a new method for calculating linear absorption spectra for large molecular aggregates, called dyadic adaptive HOPS (DadHOPS).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' This method combines the adaptive HOPS (adHOPS) framework, which uses locality to improve computational scaling, with the dyadic HOPS method previously developed to calculate linear and non-linear spectroscopic signals.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' To construct a local representation of dyadic HOPS, we introduce an initial state decomposition which reconstructs the linear absorption spectra from a sum over locally excited initial conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We demonstrate the sum over initial conditions can be efficiently Monte Carlo sampled, that the corresponding calculations achieve size-invariant (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' O(1)) scaling for sufficiently large aggregates, and that it allows for the trivial inclusion of static disorder in the Hamiltonian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We present calculations on the photosystem I core complex to explore the behavior of the initial state decomposition in complex molecular aggregates, and proof-of-concept DadHOPS calculations on an artificial molecular aggregate inspired by perylene bis-imide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' INTRODUCTION Optical spectroscopy provides essential insight into excited-state processes of molecular aggregates ranging from artificial assemblies in solution to photosynthetic pigment- protein complexes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='1–5 The interpretation of spectroscopic ob- servables in the presence of both structural heterogeneity and broad homogeneous lineshapes often requires detailed numer- ical simulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' However, established quantum dynamics methods can struggle with the combination of strong electron- vibrational coupling and spatially-extended structures charac- teristic of molecular aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='6 As a result, the development of efficient algorithms for simulating time-resolved optical spectroscopy measurements of extended molecular aggregates remains a persistent challenge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' From a molecular perspective, simulating optical spec- troscopy requires solving the time-evolution of nuclear wave packets on electronic potential energy surfaces.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A variety of methods have been developed to solve this time evolution within a formally exact framework such as multiconfiguration time-dependent Hartree (MCTDH),7–9 multilayer MCTDH (ML-MCTDH),10–13 multi-configuration Ehrenfest,14–16 and ab initio multiple spawning.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='17,18 However, the simultaneous description of electronic, intra-molecular vibrational, and en- vironmental degrees of freedom on equal footing introduces intractable computational scaling for large molecular aggre- gates;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' a problem colloquially known as the curse of dimen- sionality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Open-quantum system approaches provide a powerful coarse-graining where the relevant electronic degrees of free- dom are time-evolved in a reduced density matrix while cou- pled to an effective thermal environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Most commonly, an open-quantum system Hamiltonian incorporates a linear cou- pling to a infinite collection of harmonic oscillators parame- terized to mimic the influence of both the intramolecular vi- brations and the surrounding (condensed phase) environment within a linear response approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Within this frame- work, several formally exact methods are available for mod- eling molecular excitons including Hierarchical equations of motion (HEOM),19,20 quasi-adibatic path integrals,21,22 time- dependent density matrix renormalization group theory (TD- DMRG),23,24 time-evolving density operator with orthogo- nal polynomials,25,26 and the multi-D1 Davydov ansatz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='27,28 While these methods are capable of scaling to much larger aggregates than are usually achievable with wave function propagation techniques, they still suffer from effectively ex- ponential scaling with the number of electronic states (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' molecules) included in the calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Recent develop- ments in reduced scaling techniques - including modular path integrals29,30 and tensor contraction31–35 - suggest new paths forward may continue to extend the size of molecular aggre- gates that are tractable with these methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Recently, there has been a growing interest in solving open-quantum systems using stochastic wave functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In these approaches, the reduced density matrix is unravelled into an ensemble of wave functions that can be time-evolved independently.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In this context, the non-Markovian quan- tum state diffusion (NMQSD) equation36,37 provides a for- mally exact solution to the dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The hierarchy of pure states (HOPS) is a numerically convenient formulation of the NMQSD equations38–40 which has been recently extended into the dyadic HOPS equation to simulate both linear41 and non-linear42 optical spectroscopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dyadic HOPS propagates the bra- and ket-side of the reduced density matrix separately arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='03718v1 [quant-ph] 9 Jan 2023 2 according to the HOPS equation and provides a clear connec- tion to the non-linear response function formalism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='5 Dyadic HOPS is limited to small molecular aggregates by the same poor computational scaling that plagues HOPS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The recent development of adaptive HOPS (adHOPS), which leverages the locality of physical wave functions to achieve computational costs that stop increasing with system-size for sufficiently large aggregates (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' O(1) scaling),43 raises the possibility of dramatically expanding the reach of dyadic HOPS within an adaptive framework.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In this paper, we demonstrate dyadic adaptive HOPS (DadHOPS) calculations of linear absorption for large molecular aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We present an initial state decomposition for the dipole auto- correlation function that characterizes linear absorption which provides a local construction of the spectroscopic observable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' By combining DadHOPS with an initial state decomposition, we are able to perform size-invariant scaling simulations of linear absorption spectra for mesoscale molecular aggregates and easily incorporate the influence of static disorder in the system Hamiltonian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' This paper is organized as follows: In Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' II, we pro- vide a theoretical background for the readers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We discuss the Hamiltonian for an open quantum system followed by a brief introduction for the HOPS and dyadic HOPS methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' III, we describe an algorithm to connect adaptivity with dyadic HOPS method and provide conditions for relative error bounds that are necessary for DadHOPS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' IV, we ap- ply the initial state decomposition method to decompose the dipole autocorrelation function into a sum over local corre- lation functions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We discuss an effective method for Monte Carlo sampling over stochastic noise trajectories and initial states, with both dyadic HOPS and DadHOPS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Using 4-site and 12-site chains as examples we provide proof-of-concept calculations for the applicability of the DadHOPS method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' V, we apply the initial state decomposition to photosys- tem I and then we apply DadHOPS to a realistic model for artificial molecular aggregates inspired by Perylene bis-imide (PBI).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Finally, we conclude in Sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' VI with a summary and a brief outlook.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In Appendix A, we provide a detailed deriva- tion of the dipole autocorrelation function decomposition into a sum of local correlation functions based on a generalized initial state decomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' THEORETICAL BACKGROUND A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hamiltonian We model molecular aggregates consisting of N chro- mophores with an open-quantum system Hamiltonian ˆH = ˆHS + ˆHB + ˆHSB (1) where the system Hamiltonian ˆHS = Eg|g⟩⟨g|+ N ∑ n=1 En|n⟩⟨n|+ N ∑ n=1 N ∑ m̸=n Vnm|n⟩⟨m| (2) is composed of a shared electronic ground state, electronic ex- cited states with vertical excitation energy (En), and an elec- tronic coupling between pigments Vnm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The electronic states of each pigment are linearly coupled ˆHSB = − N ∑ n=1 ˆLn∑ q gnq �ˆb† nq + ˆbnq � (3) to an independent harmonic reservoir ˆHB = N ∑ n=1∑ q ¯hωnqˆb† nqˆbnq (4) with a system-bath coupling operator ˆLn = |n⟩⟨n|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The influ- ence of the vibrational modes on the dynamics of the elec- tronic system is described by the bath-correlation function αn(τ) = � ∞ 0 dω Jn(ω) � coth �β ¯hω 2 � cos(ωτ)−isin(ωτ) � (5) that contains the spectral density Jn(ω) = ∑q |gnq|2δ(ω −ωnq) and the inverse temperature β = 1 kBT .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We decompose the bath correlation function into a sum of exponentials indexed by jn αn(t) = ∑ jn gjne−γ jnt/¯h (6) where gjn and γjn are, in general, complex valued.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Light-Matter Interaction The light matter interaction is described in terms of the scalar, collective dipole moment operator ˆµeff = N ∑ n=1 (µµn ·εε)|n⟩⟨g|+h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=', (7) which is defined by the polarization of the incident electric field (εε) and the transition dipole operator the individual pig- ments (µµn).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The action of the collective dipole moment oper- ator on the ground state of the aggregate is to excite the super- position state |ψex⟩ = 1 µtot ˆµeff |g⟩ (8) where µtot = � ∑N n=1(µµn ·εε)2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The linear absorption spectrum is given by the half-sided Fourier transform of the dipole auto-correlation function C(t) = Tr � ˆµeff e−i ˆHt/¯h� ˆµeff|g⟩⟨g|⊗ ˆρB � ei ˆHt/¯h� (9) where ˆρ0 = |g⟩⟨g|⊗ ˆρB is a factorized initial total density ma- trix with ˆρB = e−β ˆHB/TrB � e−β ˆHB � being the density matrix of the thermal bath.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3 C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hierarchy of Pure States (HOPS) The Hierarchy of Pure States (HOPS) is a formally exact solution to the open-quantum system Hamiltonian presented in Section II A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Using HOPS, time evolution of the system reduced density matrix, starting from a separable initial state ( ˆΦ(0) = |ψ(0)⟩⟨ψ(0)|� ˆρB), can be described in terms of an ensemble average (E[·]) over stochastic wave functions ρ(t) = E � |ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)| ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ � (10) where z∗ is a stochastic trajectory with components asso- ciated with individual thermal environments zn,t defined by E[zn,t] = 0, E[zn,tzn,s] = 0, and E[z∗ n,tzn,s] = αn(t − s).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The time-evolution of the system wave functions can be calculated using the non-linear HOPS equation ¯h∂t|ψ(⃗k)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ = � −i ˆHS −⃗k ·⃗γ +∑ n ˆLn(z∗ n,t +ξn,t) � |ψ(⃗k)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ + ∑ n, jn kjnγ jn ˆLn|ψ(⃗k−⃗e jn)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ − ∑ n, jn (gjn γjn )(ˆL† n −⟨ˆL† n⟩t)|ψ(⃗k+⃗e jn)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩, (11) where the physical wave function is given by |ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩, the auxiliary wave functions are indexed by a vector ⃗k, ⃗γ is the vector of correlation function exponents (γn), and ξt,n = 1 ¯h � t 0 dsα∗ n(t −s)⟨L† n⟩s (12) is a memory term that causes a drift in the effective noise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The expectation value of the system-bath coupling operator is ⟨ˆL† n⟩t = ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ˆL† n|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (13) We limit the hierarchy to a finite depth kmax using the trian- gular truncation condition where we only include auxiliary wave functions satisfying the condition {⃗k ∈ A : ∑i ki ≤ kmax}, though other static filtering approaches have been explored previously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='44 D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dyadic HOPS Following the introduction of the pure state decomposition by Hartmann et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='45 and the subsequent work of Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=',41 the dipole autocorrelation function that defines the linear ab- sorption spectrum can be written as C(t) = µtot ∑ η∈{±1,±i} η 2 E[⟨vη(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)| ˆµeff |vη(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩] (14) where |vη(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ is the initial state |vη⟩ = 1 √ 2 (|ψex⟩+η |g⟩) (15) FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Algorithm for adHOPS time-evolved in the full Hilbert space and η is the index of the four pure state initial conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' This correlation function reduces (accounting for cancellation between η terms), to C(t) = µ2 totE[⟨ψex|ψex(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩]eiEgt/¯h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (16) As written, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (16) requires the system wave functions to be propagated in the Hilbert space containing both the elec- tronic ground and excited state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The dyadic HOPS equation, introduced in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 41, implicitly accounts for the influence of the ground state but only time evolves the excited states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The dyadic HOPS equation is equivalent to Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (11) except that the expectation value of a system-bath coupling operator is given by ⟨ˆL† n⟩t = ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ˆL† n|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩+1 , (17) and the dipole autocorrelation function becomes C(t) = µ2 totE � ⟨ψex|ψex(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2 (||ψex(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (18) III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' DYADIC ADAPTIVE HOPS Here, we present an algorithm for constructing an adap- tive basis during a dyadic HOPS trajectory that can sub- stantially reduce computational requirements when simulat- ing large systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The algorithm presented here builds on previous work implementing adaptive HOPS for a density ma- trix calculation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='43 The basic physical picture is that HOPS tra- jectories are localized by the presence of their thermal envi- ronments (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' bath Hamiltonians) and, as a result, the time- evolution can be described by a local basis that moves with the trajectory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' To ensure the adaptive calculations retain the for- mally exact structure of HOPS, the local basis must guarantee a controllable error bound on the calculation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Previously, Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 43 has demonstrated that it is possible to construct an adaptive HOPS algorithm where computational expense does not scale with system size for sufficiently large aggregates (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' O(1) scaling).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 1 presents a sketch of this algorithm, where the local basis is constructed as a direct sum (At �St) of set of auxiliary wave functions (At, called the ‘auxiliary basis’) and the set of molecular states (St, called Update Construct Construct p++S At+dt Yes HOPS Current Calculate Calculate State Update e Step?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' No4 the ‘state basis’).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Every time point, we build a new auxil- iary and state basis that ensures the difference between the full derivative of all wave functions (represented by ∂Φ) and the derivative constructed in the reduced basis ˜∂ ˜Φ is less than a user specified threshold δ (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' ||∂Φ− ˜∂ ˜Φ|| < δ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' For con- venience, we split the user specified derivative error bound (δ = � δ 2 A +δ 2 S ) into two components, the auxiliary bound (δA) and the state bound (δS), which allow us to independently control the precision we require when constructing the auxil- iary and state bases, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Extending the algorithm to dyadic adaptive HOPS (Dad- HOPS) requires a generalization to account for the fact that the norm of the physical wave function (|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩) is not conserved during the time evolution of the trajectory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' To en- sure a consistent accuracy across the trajectory, we use relative error bounds � ∆A/S(t) = δA/S ·||ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|| � that evolves in time with the norm of the physical wave function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' All calculations reported here are run with MesoHOPS v1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='14647 IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' MONTE CARLO SAMPLING LOCAL TRAJECTORIES Here, we present the combinations of the DadHOPS algo- rithm with a local representation of the dipole autocorrelation function to enable efficient simulation of absorption spectra for mesoscale molecular aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' For the DadHOPS algo- rithm to be efficient the spatial extent of delocalization must be substantially smaller than the full size of the molecular aggregate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Simultaneously, implemented directly, the dyadic HOPS expression for the dipole autocorrelation function has an initial condition that can be arbitrarily delocalized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In the following, first, we demonstrate that the dipole autocorrela- tion function can be decomposed into local initial conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Second, we demonstrate that the total correlation function can be reproduced by a simultaneous Monte Carlo sampling of noise trajectories (z∗ t ) and local initial conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Finally, we combine the initial state decomposition with the DadHOPS algorithm to demonstrate a local construction of a linear ab- sorption spectrum that can achieve size-invariant (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' O(1)) scaling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' As a model system, we consider a linear chain of Npig molecules with parallel dipole moments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The electronic cou- pling is assumed to be nearest neighbor (V = −100 cm−1) and we describe the bath correlation function as αn(t) = iλΓ− βν e−Γ+t/¯h + −iλΓ+ βν e−Γ−t/¯h (19) which is a high-temperature approximation to the spectral density described in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 48.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' For these calculations, we use λ = 35 cm−1, Γ = 50 cm−1, ν = 10 cm−1, Γ± = Γ±iν, and T = 295 K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Initial state decomposition We can expand the collective dipole operator ˆµeff = ∑ d Ad ˆσd (20) into a sum over local excitation operators acting on a cluster of pigments (d) ˆσd = ∑ d∈d µd ·εε Ad (|d⟩⟨g|+|g⟩⟨d|) (21) where Ad = � ∑ d∈d (µµd ·εε)2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (22) These equations assume that the pigment clusters represent a partition of the set of all pigments (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' they are disjoint and their union contains all pigments), though a more general construction is described in Appendix A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Inserting this expan- sion into the definition of the dipole autocorrelation function (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (9)) for the first interaction with the electric field we find C(t) = ∑ d AdCd(t) (23) where Cd(t) = Tr � ˆµeff e−i ˆHt/¯h� ˆσd|g⟩⟨g|⊗ ˆρB � ei ˆHt/¯h� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (24) In Appendix A, we prove that the local correlation function contribution can be calculated as Cd(t) = µtot E � ⟨ψex|ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2(||ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h (25) where ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗) is the initial state |ψd⟩ = ˆσd |g⟩ (26) time-evolved according to the dyadic HOPS equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In the special case where each cluster contains a single pigment (n), this equation reduces to Cn(t) = µtot E � ⟨ψex|n(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2 (||n(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h (27) where ˆσn = (|n⟩⟨g| + |g⟩⟨n|), An = µµn · εε, and |n(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ is a single-site initial excitation time-evolved according to the dyadic HOPS equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Example: To provide insight into the initial state de- composition and to show that it reproduces the total absorp- tion spectrum, we perform calculations on a four site (Npig = 4) homogeneous chain (En = 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In this case, the total correla- tion function can be decomposed into a sum of four single-site initial conditions (Cn(t)) C(t) = µ0(C1(t)+C2(t)+C3(t)+C4(t)) (28) 5 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The initial state decomposition of a linear absorption spec- trum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The total absorption spectrum (black) calculated for the 4-site chain agrees with the sum (grey) of the edge (green) and inner (blue) spectral contributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Parameters: V = −100 cm−1, λ = 35 cm−1, Γ = 50 cm−1, ν = 10 cm−1, T = 295 K, and kmax = 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' and µ0 = µµn · εε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The symmetry of the Hamiltonian gives rise to only two unique initial conditions Cedge(t) = µ0(C1(t)+C4(t)) = 2µ0C1(t) (29) Cinner(t) = µ0(C2(t)+C3(t)) = 2µ0C2(t) (30) arising from the ‘edge’ and ‘inner’ excitation calculated using Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (27).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 2 plots the corresponding single-site spectral contribution Aedge/inner(ω) = Re �� ∞ 0 Cedge/inner(t)e−iωtdt � for the edge (green line) and inner (blue line) initial condi- tions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The reconstruction of the total spectrum as the sum of the edge and inner contributions (grey line) agrees with the spectrum arising from the dyadic calculation of the total corre- lation function (black line) using the four site initial condition (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (16)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Note that individual spectral contributions (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Aedge/inner(ω)) are not themselves proper absorption spectra since initial and final state are not the same.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Monte Carlo Sampling Dyadic HOPS The total correlation function can be calculated by Monte Carlo sampling over single-site initial conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The initial state decomposition of the full correlation function introduces an independent trace over the bath for each initial condition C(t) = µtot∑ d Ad ∑ z∗ 1 Nd,z � ⟨ψex|ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2(||ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h (31) FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Statistical convergence of Monte Carlo sampling over the initial state decomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (a) Bootstrapped spectral error for the total absorption spectrum calculated with single site initial condi- tions (black triangles, Nc = 1) either without (filled) or with (open) static disorder.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Similar statistical convergence is observed for the edge (green circles) and inner (blue squares) single site initial con- ditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (b) Comparison of the statistical error for the total spectrum normalized by the square-root of the number of pigments in each cluster (√Nc) when using single site (Nc = 1, triangles), pair (Nc = 2, circles), and all four site (Nc = 4, squares) initial conditions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Param- eters: V = −100 cm−1, λ = 35 cm−1, Γ = 50 cm−1, ν = 10 cm−1, T = 295 K, and kmax = 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' where the sum over noise trajectories (z∗) is independent of the sum over the ND initial conditions (d).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' If we assume un- biased sampling with equal sized clusters, then the number of noise trajectories per initial condition is Nd,z = Nens/ND and the correlation function can be calculated as C(t) = µtot ND Nens ∑ (d,z∗) Ad � ⟨ψex|ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2(||ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h, (32) where we Monte Carlo sample Nens pairs (d,z∗) to calculate the full correlation function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Example: Here, we perform calculations using the same 4 site linear chain considered above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3a compares the bootstrapped spectral error for the total spectrum using Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (32) with single site initial conditions (black filled triangles) error = 1 � ωmax ωmin Aref(ω)dω � ωmax ωmin ��A (ω)−Aref(ω) ��dω (33) as a function of the number of trajectories sampled from the combined (n,z∗) space.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We find that the statistical conver- gence of the total correlation function is equivalent to that of the edge (green circles) and inner (blue squares) spectral com- ponents calculated independently using Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (27).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Inner+Edge Total 150 Inner Edge 125 100 A(w) 75 50 25 0 400 200 0 200 400 Energy (cm-1)(a) Log1o(Mean Error) 100 101 102 103 Number of Trajectories (b) Log1o(VNc Error) V 1 100 101 102 103 Number of Trajectories6 One advantage of Monte Carlo sampling is the trivial incor- poration of an additional sampling over disorder in the system Hamiltonian ( ˆHs).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3a shows the statistical convergence of the total spectrum calculated with single-site initial conditions in the presence of Gaussian distributed disorder on site ener- gies (black open triangles) is almost unchanged compared to the homogeneous case (black filled triangles).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' For the disor- dered calculations, site energies form a Gaussian distribution with mean value of zero and a standard deviation (SD) of 100 cm−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Finally, we find that the statistical error of the Monte Carlo sample for a given number of trajectories is inversely propor- tional to the size of the clusters (Nc = Npig/ND) used for the initial state decomposition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3b, we compare the nor- malized statistical error (√Nc· Error) as a function of the num- ber of independent trajectories (Ntraj) when using different ini- tial conditions: single site (triangles, Nc = 1), pairs (circles, Nc = 2), and all four sites (squares, Nc = 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The equiva- lence of these three results shows that increasing the size of the clusters decreases the number of trajectories required to reach a given statistical error.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Thus, when using an equation of motion where the delocalization extent of the initial state does not change the computational expense, the initial state decomposition provides no advantage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Monte Carlo Sampling Dyadic Adaptive HOPS The local correlation functions arising from the initial state decomposition can be efficiently simulated using DadHOPS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4a shows how, for a 4-site linear chain, the error of the edge component decreases with the derivative error bound δA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Here, we consider DadHOPS calculations converged when the adaptive error is lower than the statistical error for the asso- ciated ensemble of 104 trajectories (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4a, grey horizontal line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4b the edge spectral contribution calculated with converged DadHOPS parameters (thin line, δA = 10−3) re- produces the corresponding dyadic HOPS calculation (thick transparent line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The DadHOPS algorithm combined with the initial state de- composition allows us to calculate much larger system sizes then would be possible with dyadic HOPS directly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' As a first demonstration of scaling we consider a 12-site linear chain where the basis for the full HOPS calculation has more than 106 elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4c plots the total absorption spectrum cal- culated with DadHOPS (thick line, δA = 10−3, δS = 0) using the full initial condition (|ψ0⟩ = 1 √ 12 ∑12 n=1 |n⟩).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We can re- produce this spectrum by Monte Carlo sampling over single- site initial conditions and incorporating adaptivity in both the auxiliary and state basis (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4c, thin line, δA = δS = 10−3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Combining adaptivity in both the auxiliary and state basis with localized initial states can achieve size-invariant scaling for sufficiently large aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4d plots the number of state basis elements (bottom) and the number auxiliary wave functions (top) required for both the full dyadic HOPS (thick grey lines) and the DadHOPS algorithm (thin green lines) as a function of the number of sites in the linear chain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Here, we have demonstrated that the advantage of the local- FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Comparing dyadic HOPS and DadHOPS for N-site lin- ear chains.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (a) Mean error of the edge spectrum with respect to the auxiliary error bound (δA) for a 4-site chain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The solid grey line rep- resents the statistical error.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (b) The edge spectral component for a 4-site chain calculated with converged DadHOPS parameters (thin line) and corresponding dyadic HOPS calculation (thick transparent line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (c) Total absorption spectrum for a 12-site chain calculated using DadHOPS with (thin line) and without state adaptivity (thick line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (d) Average number of auxiliary wave functions (top) and state basis elements (bottom) for linear chains of different lengths re- quired for calculations using DadHOPS (green) compared to dyadic HOPS (thick grey).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Parameters: V = −100 cm−1, λ = 35 cm−1, Γ = 50 cm−1, ν = 10 cm−1, T = 295 K, and kmax = 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' ized wave functions introduced by the initial state decomposi- tion is realized by combining it with the DadHOPS equation- of-motion that leverages locality to reduce computational ex- pense.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' APPLICATION A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Photosystem I (PSI) Photosystem I (PSI) is a pigment-protein complex contain- ing 96 chlorophyll per monomer and usually found as a trimer in higher plants and algae (Fig 5a).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='50 The complicated spa- tial arrangement of chlorophyll make this an ideal test sys- tem for exploring the behavior of the initial state decompo- sition with respect to different definitions of clusters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Here, we use a previously reported Hamiltonian51 where chloro- phyll excitation energies (En) and excitonic couplings (Vn,m) were calculated using time-dependent density function the- ory (TDDFT) evaluated by Gaussian1652 with the CAM- B3LYP density functional53 and the 6-31G∗ basis set;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='54 this approach is known to provide quality descriptions of both (a) (b) 30 - 4 site Statistical 2 Error (3) 20 10 - Edge 0: 10-4 10-3 500 0 500 Aux.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Error Bound (SA) Energy (cm-1) (c) (d) 10° 3 - ler 12 site 103 A(w) /100 2 101 20 state Total 10 N 0 0 500 0 500 100 102 Energy (cm-1) Numb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' of Pigments7 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Simulating linear absorption for Photosystem I (PSI).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (a) A PSI core complex trimer is shown with a single monomer high- lighted;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' the simulations are for one monomer containing 96 pig- ments, using cyclic boundary conditions to represent the trimer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (b) A chlorophyll a molecule with the phytyl tail truncated for clarity is show along with the Qy transition dipole vector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (c) Comparison of absorption spectrum simulated using HEOM (grey)49 and Dyadic HOPS (green).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (d) Bootstrapped spectral error for randomly assign- ing pigments to clusters (black solid circles) and using clusters de- fined by strong coupling (green open circles).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Parameters: λ = 35 cm−1, Γ = 50 cm−1, T = 300 K, Γmark = 500 cm−1, SD= 150 cm−1, and kmax = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' vibrational and vibronic processes apparent from the high- resolution absorption and emission spectra of chlorophyll- type molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='55 The Hamiltonian is provided in the Sup- porting Information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The system-bath coupling is described by a Drude-Lorentz spectral density Jn(ω) = 2λΓω ω2 +Γ2 (34) and the corresponding bath correlation function is given in a high-temperature approximation as αn(t) = (2λ/β −iλΓ)e−Γt/¯h +iλΓe−Γmarkt/¯h (35) where the second exponential is included to ensure the imagi- nary component of αn(t) is continuous at time t = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' To agree with previous HEOM calculations,49 we introduce a Gaussian distributed static disorder on the chlorophyll site energies with a standard deviation (SD) of 150 cm−1, we approximate the true chlorophyll Qy transition dipole moment (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 5b) by the vector defined by the position of the NA and NC nitro- gen atoms (IUPAC notation for chlorophyll a),56 and we set the spectral density parameters to be: λ = 35 cm−1, Γ = 50 cm−1, T = 300 K, and Γmark = 500 cm−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The varied alignments of the chlorophyll Qy dipole mo- ments throughout PSI introduces an orientation dependence into the linear absorption spectrum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' For an isotropic distri- bution of PSI complexes, we calculate the linear absorption spectrum from the average of the dipole autocorrelation func- tion for the x, y, and z polarized electric fields (εε) C(εε)(t) = µtot,εε ND ∑ d=1 �Ad,εε Nd,z � ∑ z∗ � ⟨ψex|ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2(||ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (36) Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 5c shows that, as expected, the HOPS calculations (green line) reproduce previous HEOM calculations from Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 49 (grey line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' To what extent does the specific choice of clusters matter for a given cluster size?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In the case of a linear chain, the choice of clusters to be sets of adjacent pigments may appear obvious.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' There is no equivalently obvious choice for the het- erogeneously coupled pigments in PSI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 5d compares the statistical convergence of the linear absorption spectrum for two different definitions of clusters: the first randomly assigns four pigments to each cluster (black solid circles), while the second constructs clusters of 4 strongly interacting pigments (green open circles) using an algorithm described in Appendix B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' There is no appreciable difference in the convergence be- tween randomly assigning pigments to clusters or using clus- ters defined by strong coupling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Combined with the results presented in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3b showing square-root-scaling of error with cluster size, our calculations suggest that using a cluster initial condition only acts to increase the effective number of com- bined (n,z∗) pairs that are being sampled.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Perylene bis-imide (PBI) Perylene bis-imide (PBI) is a family of pigments that can form both J- and H-type aggregates in solution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='57–60 J- aggregates of PBI-1 have been the object of particular study and show a strong vibronic progression in their linear ab- sorption sprectra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Previous time-dependent density func- tional theory (TDDFT) calculations of PBI-1 have character- ized electronic coupling between adjacent monomers (V ≈ −500 cm−1) and 28 intramolecular vibrational modes with appreciable Huang-Rhys factors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='61 This vibrational struc- ture has served as a starting point for different calcula- tions of linear absorption spectra for PBI-1 aggregates us- ing techniques such as multiconfiguration time-dependent Hartree (MCTDH),61 multilayer MCTDH (ML-MCTDH),62 and time-dependent density matrix renormalization group the- ory (TD-DMRG).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='63 Calculations of the exciton dynamics in aggregates ranging in size from two to 25 monomer units have also been reported with varying degrees of complexity in their description of the molecular vibrations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='60,64,65 We have constructed a minimal bath correlation function composed of two modes to describe the vibrational environ- ment of PBI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We include a vibration with a central frequency near 1500 cm−1 to account for the group of strongly-coupled vibrations ranging from 1370-1630 cm−1 previously reported from TDDFT calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='61 We also include a low-frequency (a) (b) (c) (d) 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='0 - Log1o(Mean Error) Spectrum 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='5 Abs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='0: 2 14000 15000 101 102 103 104 Numb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' of Traj Energy (cm-1)8 Mode g (cm−2) γ (cm−1) 1 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='2×105 50+170i 2 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='6×106 100+1550i TABLE I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The exponential parameters describing the vibrational cor- relation function of PBI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Simulating linear absorption for perylene bis-imide (PBI) J- aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (a) Compares the experimental absorption spectra from dilute solution (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='0016 mM, grey)]59 with calculations broadened by either SD = 300 cm−1 (green) or 400 cm−1 (blue).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (b) Compares the experimental absorption spectra from concentrated solution (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='16 mM, grey)59 with a calculated trimer spectra (green).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (c) Compares the lineshape calculated for aggregates containing one (black), two (green), three (blue), seven (cyan), and 1000 pigments (grey).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (d) Plots the change in the 0-0 transition peak shift (∆E, top) and the ratio of the 0-1 peak intensity to the 0-0 peak intensity (I0,1/I0,0, bottom) as a function of the number of molecules in the aggregate (Npig).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Parameters: kmax = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' vibration to account for the broad dissipative environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The final parameters (Table I) were selected to agree with the available experimental data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Our minimal description of the system-bath correlation function is capable of reproducing the major features of both the experimentally measured monomer PBI spectrum (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6a) and the aggregate spectrum (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6b).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The monomer spec- trum shows a broad lineshape and is better reproduced when the Gaussian disorder of the vertical excitation energies has a standard deviation of SD = 400 cm−1 (blue line) compared with the SD = 300 cm−1 (green line) we use for the aggregate spectrum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We can further improve the agreement between the simulated and experimental monomer spectra by introducing additional moderate frequency modes (data not shown), but this additional complexity was not required for the reproduc- tion of the experimental aggregate spectrum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We smooth the aggregate spectra reported here by weighting the time-domain dipole auto-correlation function with a cosine appodization window66 Θ(t) = � cos( π 2t/tmax) t ≤ tmax 0 t > tmax (37) that goes to zero at the last time point of the calculated tra- jectory (tmax).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The use of this window function suppresses noise in the calculated spectra that arises from the combina- tion of zero-padding66 and the incomplete cancellation of the correlation function at long times due to finite sampling of the trajectory ensembles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The advantage of the adaptive dyadic HOPS formalism is the ability to calculate even very large molecular aggregates efficiently.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The size of a molecular aggregate is often impor- tant for capturing the influence of exciton delocalization on both the 0-0 peak position and the relative magnitude of vi- bronic side-bands.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='67 Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6c compares the monomer (black line), dimer (green line), trimer (blue line), and heptamer (cyan line) lineshapes with that calculated for a 1000-site lin- ear chain (grey line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The magnitude of the vibronic side- band decreases rapidly from the monomer to trimer and then changes slow with increasing chain length.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A similar trend is seen for the red-shift of the 0-0 peak.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6d quantifies this effect by plotting the central position of the 0-0 peak (∆E, top) and the ratio of the intensity of the 0-1 peak to the intensity of the 0-0 peak (I0,1/I0,0, bottom) as a function of the num- ber of pigments in the aggregate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In both cases, the grey lines represents the asymptotic limit of a 1000-site linear chain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Finally, let us consider the relationship between aggregate size, locality, and the onset of size-invariant scaling in the DadHOPS equation-of-motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' When the adaptive basis size stops increasing with the length of the aggregate, the extent of exciton delocalization is sufficiently small that most tra- jectories will never sample the edge of the aggregate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' This implies that for aggregates sufficiently large that the basis is size-invariant the spectroscopic signatures should also be size- invariant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 7 shows that by Npig = 10 the average size of the adaptive site basis begins to plateau before reaching a size- invariant value of 13 for chains of 100 PBI molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' This is consistent with the convergence of the spectral features which are close to the asymptotic values at Npig = 7 (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6d) and the expectation that complete size-invariance will occur when only a relatively small set of trajectories will sample pigments on the edge of the chain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We suggest, then, that DadHOPS can be thought of as introducing a dynamic separation of length- scales where the basis size required to capture the spectral fea- tures is solved on-the-fly rather than being asserted a priori.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' As a result, DadHOPS is advantageous for simulating realistic molecular aggregates, particularly in the presence of structural disorder, where separations of length scales can be obscure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (a) (b) Spectrum Spectrum 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='5 Abs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Abs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='0 17500 20000 17500 20000 Energy (cm-1) Energy (cm-1) (c) (p) T- 1000 (cm Spectrum E 2000 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='5 lo,1 /lo,0 Abs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='50 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='0 2500 0 2500 3 1 5 Energy (cm-1) Number of Pigments9 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Scaling of DadHOPS basis size for PBI aggregate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Aver- age number of auxiliary wave functions (top) and states (bottom) for different lengths of PBI chains required for calculations using Dad- HOPS (green) compared to dyadic HOPS (grey).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' VI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' CONCLUSIONS Here we have presented a new algorithm that combines the dyadic adaptive HOPS (DadHOPS) equations with a initial state decomposition and is capable of calculating optical lin- ear absorption spectra for mesoscale molecular aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Our approach introduce an dynamic separation of length- scales by adaptively constructing a reduced basis to describe the time-dependence of local contributions to the dipole auto- correlation function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The adaptive basis construction is capa- ble of achieving size-invariant scaling (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' O(1)) with the number of molecules for sufficiently large aggregates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We have applied the initial state decomposition to the 96 chloro- phyll photosystem I core complex and found that the specific choice of pigment clusters is not essential to the statistical convergence of the calculations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We calculated 1000 molecule J-aggregate of perylene bis-imide (PBI) with DadHOPS and characterized how the absorption spectra changes with ag- gregate size.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Because locality cannot reduce the computa- tional expense of describing individual pigments, DadHOPS remains limited to calculations where each thermal reservoir has a small number of exponential modes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' In the future, it may be possible to describe more complex thermal environment by incorporating techniques such as tensor contraction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='68 ACKNOWLEDGMENTS AE acknowledges support from the DFG via a Heisenberg fellowship (Grant No EI 872/10-1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' TG, EJT, and DIGB acknowledge support from the Robert A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Welch Foundation (Grant N-2026-20200401) and start-up funds from Southern Methodist University.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' DIGB acknowledges the US National Science Foundation CAREER Award under Grant No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' CHE- 2145358.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' DATA AVAILABILITY STATEMENT The data that support the findings of this study are available from the corresponding author upon reasonable request.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Appendix A: Dyadic HOPS with Generalized Initial State Decomposition The dipole autocorrelation function C(t) = Tr � ˆµeff e−i ˆHt/¯h� ˆµeff|g⟩⟨g|⊗ ˆρB � ei ˆHt/¯h� (A1) depends on the time-evolution of the first-order density matrix ˆµeff |g⟩⟨g| arising from one interaction with the electric field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The Non-Markovian Quantum State Diffusion (NMQSD) for- malism that gives rise to the HOPS equation, however, can only time-evolve pure state density matrices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The pure state decomposition introduced in Ref.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 45 rewrites the initial first- order density matrix into a sum of pure state density matrices which can then be treated within the NMQSD (or HOPS) for- malism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Here, we use the pure state decomposition to rewrite the dipole autocorrelation function into a sum of locally ex- cited correlation functions that can be efficiently simulated using the DadHOPS equation-of-motion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We begin by decomposing the collective transition dipole operator ˆµeff = N ∑ n (µµn ·εε)|n⟩⟨g|+h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=', (A2) into an arbitrary (finite) set of interaction operators ( ˆσa) and real-valued weights (Aa) defined to ensure that ˆµeff = ∑ a Aa ˆσa (A3) and |ψa⟩ = ˆσa |g⟩ is a normalized wave function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' The dipole autocorrelation function can then be rewritten as C(t) = ∑ a AaCa(t) (A4) where Ca(t) = Tr � ˆµeff e−i ˆHt/¯h� ˆσa|g⟩⟨g|⊗ ˆρB � ei ˆHt/¯h� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A5) The decomposition expressed here is generic and does not re- quire - for example - that the interaction operators be orthog- onal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We will proceed in our derivation with these generic in- teraction operators and then introduce the specific expressions for two convenient special cases at the end.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 106 Nhier 104 102 100 100 101 102 102 Istate 101 N 100 100 101 102 Number of Pigments10 We can then define a set of pure states |vη,a⟩ = 1 √ 2 � η |g⟩+|ψa⟩ � (A6) which can be used to reconstruct the initial first-order density matrix associated with the correlation function for each exci- tation operator ρa(t = 0) = ˆσa |g⟩⟨g| = ∑ η∈{±1,±i} η 2 |vη,a⟩⟨vη,a|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A7) To prove the second equality, we expand the summand η 2 |vη,a⟩⟨vη,a| = 1 4 � ηηη∗ |g⟩⟨g|+ηη |g⟩⟨ψa| +ηη∗ |ψa⟩⟨g|+η |ψa⟩⟨ψa| � (A8) and note that ∑ η∈{±1,±i} ηηη∗ = 0, ∑ η∈{±1,±i} ηη = 0 ∑ η∈{±1,±i} ηη∗ = 4, ∑ η∈{±1,±i} η = 0 which reduces the right-hand side of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A7) to ∑ η∈{±1,±i} η 2 |vη,a⟩⟨vη,a| = |ψa⟩⟨g| = ˆσa |g⟩⟨g|.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A9) We can now proceed to write down the standard dyadic HOPS expressions for the individual contributions to the dipole autocorrelation function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We begin by rewriting the components of the dipole autocorrelation function as Ca(t) = ∑ η∈{±1,±i} Tr � ˆµeff e−i ˆHt/¯h�η 2 |vη,a⟩⟨vη,a|⊗ ˆρB � ei ˆHt/¯h� (A10) which is equal to Ca(t) = ∑ η∈{±1,±i} E � η 2 ⟨vη,a(t,z∗)| ˆµeff|vη,a(t,z∗)⟩ ⟨vη,a(t,z∗)|vη,a(t,z∗)⟩ � (A11) where |vη,a⟩ is time-evolved according to the non-linear HOPS equation (Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (11)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' As written, the time-evolution of the pure state |vη,a⟩ is performed using a system Hamilto- nian ( ˆHS) that contains both the ground and excited electronic states.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' However, since the system Hamiltonian does not cou- ple the electronic ground and excited states, the time evolution of the two components of |vη,a⟩ can be decoupled giving |vη,a(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ = 1 √ 2 � η |g⟩e−iEgt/¯h +|ψa(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ � (A12) where |ψa(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ is propagated using the non-linear HOPS equation where ˆHS includes only the first excitation manifold and the expectation value of the system-bath coupling opera- tor in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (17) is redefined as ⟨ˆL† n⟩t = ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ˆL† n|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ ⟨ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ψ(⃗0)(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩+1 (A13) where the change in the denominator arises from the presence of the ground-state component of the wave function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We can use Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A12) and the definition of the collective dipole operator ( ˆµeff |g⟩ = µtot |ψex⟩, Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (7)) to rewrite the numerator of Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A11) as η 2 ⟨vη,a(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)| ˆµeff|vη,a(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ = µtot ηη 4 ⟨ψa(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|ψex⟩e−iEgt/¯h + µtot |η|2 4 ⟨ψex|ψa(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩eiEgt/¯h and the denominator as ⟨vη,a(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)|vη,a(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ = 1 2 � 1+||ψa(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A14) Noting that the denominator does not depend on η, the cancel- lation in the numerator due to summation over η ∈ {±1,±i} gives Ca(t) = µtotE � ⟨ψex|ψa(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2 (||ψa(t,z∗)||2 +1) � eiEgt/¯h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A15) This equation is equivalent to the standard dyadic HOPS ex- pression, but now propagating a component of the total cor- relation function defined by a decomposition of the collective transition dipole operator into a sum of excitation operators.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Cluster Decomposition One simple decomposition of the effective collective tran- sition dipole operator is to expand in clusters of pigments ˆµeff = N ∑ n (µµn ·εε)|n⟩⟨g|+h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A16) =∑ d Ad ∑ d∈d µµd ·εε Ad (|d⟩⟨g|+|g⟩⟨d|) (A17) =∑ d Ad ˆσd (A18) where ˆσd = ∑ d∈d µd ·εε Ad (|d⟩⟨g|+h.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=') (A19) and Ad = � ∑ d∈d (µµd ·εε)2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A20) The total correlation function can then be decomposed as C(t) = µtot∑ d AdE � ⟨ψex|ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2 (||ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h (A21) where |ψd(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ is the state |ψd⟩ = ˆσd |g⟩ time-evolved ac- cording to the dyadic HOPS equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 11 b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Pigment Decomposition If each cluster is composed of a single pigment, then Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' (A21) simplifies to C(t) = µtot∑ n (µµn ·εε)E � ⟨ψex|n(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ 1 2 (||n(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)||2 +1) � eiEgt/¯h (A22) where |n(t;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='z∗)⟩ is the single-pigment excitation state |n⟩ time- evolved according to the dyadic HOPS equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Appendix B: Defining Strongly Interacting Clusters To explore the influence of cluster definitions, we con- structed clusters between four strongly coupled pigments in photosystem I (PSI).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' We use an iterative, three-step algorithm designed to ensure that strongly interacting pigments are pref- erentially included inside the same cluster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Step 1: Locate the largest coupling element (Vi, j) among the pigments not yet assigned to a cluster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Pig- ments (i, j) for the nucleus of a new cluster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Step 2: Locate the largest coupling element to either pigment in the new cluster to another pigment not yet assigned to a cluster (Vm, j or Vi,m).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Add this pigment to the new cluster: (i, j,m).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Step 3: Locate the largest coupling element involving a pigment in the new cluster with another pigment (n) not yet assigned to a cluster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Add this pigment to the new cluster: (i, j,m,n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' End Condition: If any pigments remain unassigned then return to step 1 and define a new cluster.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' This algorithm does not guarantee that all strong couplings are contained within a cluster, but it does ensure that clusters nucleate around strong coupling elements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' REFERENCES 1H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' van Amerongen, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Valk¯unas, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' van Grondelle, Photosynthetic Ex- citons (World Scientific, 2000).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 2V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' May and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kühn, Charge and Energy Transfer Dynamics in Molecular Systems: A Theoretical Introduction (Wiley, 2000).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 3D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Abramavicius, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Palmieri, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Voronine, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Šanda, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Mukamel, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 109, 2350 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 4L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shenai, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Zheng, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Somoza, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Zhao, Molecules 20, 15224 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 5S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Mukamel, Principles of Nonlinear Optical Spectroscopy, Oxford series in optical and imaging sciences (Oxford University Press, 1995).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 6M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Gelin, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Domcke, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 122, 17339 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 7M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Beck, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Jäckle, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Worth, and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Meyer, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 324, 1 (2000).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 8I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Burghardt, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Nest, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Worth, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 119, 5364 (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 9G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Richings, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Polyak, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Spinlove, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Worth, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Burghardt, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lasorne, Int.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 34, 269 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 10H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Wang and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Thoss, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 119, 1289 (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 11H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Wang, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 119, 7951 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 12U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Manthe, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 128, 164116 (2008).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 13O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Vendrell and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='-D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Meyer, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 134, 044135 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 14D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shalashilin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 130, 244101 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 15D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shalashilin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 132, 244111 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 16L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Gelin, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shalashilin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 151, 244116 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 17M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ben-Nun and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Martínez, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 108, 7244 (1998).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 18M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ben-Nun, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Quenneville, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Martínez, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 104, 5161 (2000).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 19Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Tanimura, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Japan 75, 082001 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 20Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Tanimura, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 153, 020901 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 21N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Makri, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 193, 435 (1992).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 22N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Makri, Journal of Mathematical Physics 36, 2430 (1995).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 23M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Cazalilla and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Marston, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 88, 256403 (2002).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 24S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' White and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Feiguin, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 93, 076401 (2004).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 25J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Prior, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chin, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Huelga, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Plenio, Physical review letters 105, 050404 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 26D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Tamascelli, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Smirne, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Huelga, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Plenio, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 123, 090402 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 27L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Borrelli, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Zhao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 121, 8757–8770 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 28L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Gelin, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Zhao, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 515, 108 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 29N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Makri, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 148, 101101 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 30N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Makri, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 149, 214108 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 31R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Borrelli and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dolgov, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B 125, 5397–5407 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 32Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Yan, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Xu, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Li, and Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 154, 194104 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 33A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Somoza, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Marty, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Huelga, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Plenio, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 123, 100502 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 34D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Tamascelli, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Smirne, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lim, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Huelga, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Plenio, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 123, 090402 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 35A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strathearn, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kirton, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kilda, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Keeling, and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lovett, Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Com- mun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 9, 3322 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 36L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Diósi and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strunz, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 235, 569 (1997).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 37L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Diósi, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Gisin, and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strunz, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 58, 1699 (1998).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 38D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Suess, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eisfeld, and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strunz, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 113, 150403 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 39R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hartmann and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strunz, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 13, 5834 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 40G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ritschel, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Suess, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Möbius, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strunz, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eisfeld, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 142, 034115 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 41L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Bennett, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eisfeld, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 156, 124109 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 42L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Bennett, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eisfeld, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 157, 114104 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 43L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Varvelo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lynd, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Bennett, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 12, 9704 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 44P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Zhang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Bentley, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eisfeld, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 148, 134103 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 45R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hartmann and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Strunz, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 125, 7066 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 46L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Varvelo, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lynd, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Citty, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Raccah, Mesohops v1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='1 (2023), URL https://zenodo.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='org/record/7504694.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 47Note1, the most recent version of MesoHOPS can be found here: https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='com/MesoscienceLab/mesohops.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 48A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ishizaki, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Japan 89, 015001 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 49T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kramer, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Noack, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Reimers, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Reinefeld, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rodríguez, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Yin, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 515, 262 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 50P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Jordan, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Fromme, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Witt, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Klukas, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Saenger, and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Krauß, Nature 411, 909 (2001).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 51S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Yin, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dahlbom, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Canfield, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hush, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kobayashi, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Reimers, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B 111, 9923 (2007).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 52M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Frisch, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Trucks, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Schlegel, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Scuseria, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Robb, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Cheeseman, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Scalmani, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Barone, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Petersson, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Nakatsuji, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=', Gaussian~16 Revision C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='01 (2016), gaussian Inc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Wallingford CT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 53T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Yanai, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Tew, and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Handy, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 393, 51 (2004).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 54W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hehre, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ditchfield, and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Pople, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 56, 2257 (1972).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 55J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Reimers, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rätsep, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Linnanto, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Freiberg, Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Estonian Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 71, 127 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 56R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Blankenship, Molecular Mechanisms of Photosynthesis (Wiley, 2021), 3rd ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 57V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dehm, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Baumeister, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Prins, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Siebbeles, and F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Würthner, Org.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 9, 1085 (2007).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 58Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chen, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Stepanenko, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dehm, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Prins, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Siebbeles, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Seibt, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Mar- quetand, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Engel, and F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Würthner, Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 13, 436 (2007).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 59H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Marciniak, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='-Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Li, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Würthner, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Lochbrunner, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 12 A 115, 648 (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 60M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Schröter, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ivanov, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Schulze, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Polyutov, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Yan, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Pullerits, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kühn, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 567, 1 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 61D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ambrosek, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Köhn, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Schulze, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kühn, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 116, 11451 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 62J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Seibt, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Winkler, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Renziehausen, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Dehm, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Würthner, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='-D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Meyer, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Engel, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 113, 13475 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 63J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ren, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shuai, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content='-L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chan, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Theory Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 14, 5027 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 64S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kundu and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Makri, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 154, 114301 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 65S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Kundu and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Makri, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 23, 15503 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 66J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Hoch and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Stern, NMR Data Processing (Wiley, 1996).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 67R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ghosh and F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Spano, Acc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Chem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Res.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 53, 2201 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' 68X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Gao, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Ren, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Eisfeld, and Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Shuai, Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} +page_content=' A 105, L030202 1 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/xdE2T4oBgHgl3EQfLwbX/content/2301.03718v1.pdf'} diff --git a/y9E0T4oBgHgl3EQf_AIq/content/tmp_files/2301.02820v1.pdf.txt b/y9E0T4oBgHgl3EQf_AIq/content/tmp_files/2301.02820v1.pdf.txt new file mode 100644 index 0000000000000000000000000000000000000000..30367a1529c35eb27f1c4b9cc70edc241d299b17 --- /dev/null +++ b/y9E0T4oBgHgl3EQf_AIq/content/tmp_files/2301.02820v1.pdf.txt @@ -0,0 +1,2973 @@ +arXiv:2301.02820v1 [math.CO] 7 Jan 2023 +Observations on the Lov´asz θ-Function, Graph +Capacity, Eigenvalues, and Strong Products +Igal Sason +Dedicated to my friend and former teacher, Professor Emeritus Abraham (Avi) Berman, in the +occasion of his eightieth birthday +Citation: I. Sason, “Observations on the Lov´asz θ-function, graph capacity, eigenvalues, and +strong products,” Entropy, vol. 25, no. 1, paper 104, pp. 1–41, January 2023. +DOI: https://doi.org/10.3390/e25010104. +Abstract +This paper provides new observations on the Lov´asz θ-function of graphs. These include a simple +closed-form expression of that function for all strongly regular graphs, together with upper and lower +bounds on that function for all regular graphs. These bounds are expressed in terms of the second-largest +and smallest eigenvalues of the adjacency matrix of the regular graph, together with sufficient conditions +for equalities (the upper bound is due to Lov´asz, followed by a new sufficient condition for its tightness). +These results are shown to be useful in many ways, leading to the determination of the exact value of +the Shannon capacity of various graphs, eigenvalue inequalities, and bounds on the clique and chromatic +numbers of graphs. Since the Lov´asz θ-function factorizes for the strong product of graphs, the results +are also particularly useful for parameters of strong products or strong powers of graphs. Bounds on the +smallest and second-largest eigenvalues of strong products of regular graphs are consequently derived, +expressed as functions of the Lov´asz θ-function (or the smallest eigenvalue) of each factor. The resulting +lower bound on the second-largest eigenvalue of a k-fold strong power of a regular graph is compared to +the Alon–Boppana bound; under a certain condition, the new bound is superior in its exponential growth +rate (in k). Lower bounds on the chromatic number of strong products of graphs are expressed in terms +of the order and the Lov´asz θ-function of each factor. The utility of these bounds is exemplified, leading +in some cases to an exact determination of the chromatic numbers of strong products or strong powers +of graphs. The present research paper is aimed to have tutorial value as well. +Keywords: Lov´asz θ-function, Shannon capacity of a graph, strongly regular graph, strong product of +graphs, vertex- and edge-transitivity, Alon–Boppana bound, Ramanujan graph, chromatic number. +I. +INTRODUCTION +The notion of the graph capacity in Shannon’s problem of zero-error communication [86] +had a significant impact on the development of information theory and graph theory, including +the introduction of perfect graphs by C. Berge [11], strong graph products (or powers) ([7], +[84]), the introduction of the Lov´asz θ-function of a graph as a computable upper bound on its +Shannon capacity ([57], [65]), the rank-bound by Haemers [43], and other important follow-up +works that are surveyed, e.g., in [7], [9], [52], [58]. +In graph theory, there are four central sorts of graph products, each with its own applications +and theoretical interpretations. The reader is referred to the excellent handbook [46], which +presents the rich and fertile field of graph products. Strong product of graphs is one of the most +extensively studied sorts of graph products, and there exists a polynomial-time algorithm that +I. Sason is with the Viterbi Faculty of Electrical and Computer Engineering, and the Department of Mathematics +at the Technion - Israel Institute of Technology, Haifa 3200003, Israel (e-mail: eeigal@technion.ac.il). + +2 +finds the unique prime factorization of any connected graph with that type of multiplication +[31]. Strong powers of graphs are also fundamental in information theory. Their information- +theoretic significance stems from the notion of the Shannon capacity of graphs for error-free +communication [86], and the Witsenhausen rate [94] in the zero-error source coding problem +with perfect side information at the receiver. Properties of strong products and strong powers of +graphs, and bounds on their independence numbers and chromatic numbers have been extensively +studied, e.g., in [4], [7], [9], [10], [12], [13], [14], [15], [16], [17], [29], [30], [31], [36], [38], +[39], [43], [45], [46], [47], [51], [52], [54], [55], [56], [64], [65], [70], [84], [86], [87], [90], +[91], [94], [95]. +The present work continues the above paths of research. It provides some new observations on +the Lov´asz θ-function of regular graphs, calculation of the Shannon capacity of some strongly +regular graphs, bounds on eigenvalues of graphs (in particular, the second-largest and smallest +eigenvalues of the adjacency matrix, which play a key role in spectral graph theory), bounds +related to Ramanujan graphs, and strong products of graphs. The analysis in the present work +mainly relies on the notion of the Lov´asz θ-function of graphs [65]. The paper includes a thorough +review of the backgrounds relevant to this work with suitable references or explanations, which +also serve to motivate the presentation of the results in this work and to put them into perspective. +The presentation in this research paper is consequently aimed to have tutorial value as well. +The results obtained in this work are outlined as follows: +(1) A known upper bound on the Lov´asz θ-function of a regular graph is expressed in terms of +the smallest eigenvalue of its adjacency matrix [65]. A key result in this work provides a +lower bound on the Lov´asz θ-function of a regular graph, which is expressed in terms of the +second-largest eigenvalue of its adjacency matrix. New sufficient conditions for equalities +in these bounds are also obtained (Proposition 1). +(2) A simple and closed-form expression of the Lov´asz θ-function is derived for all strongly +regular graphs (Corollary 1). +(3) Eigenvalue inequalities are derived, which relate the smallest and second-largest eigenvalues +of a regular graph. They hold with equality if and only if the graph is strongly regular +(Corollaries 2 and 3). +(4) The Shannon capacity of several strongly regular graphs is determined (Section III-E). +(5) Bounds on parameters of regular graphs, and in particular of Ramanujan graphs, are derived +(Corollaries 4–6). +(6) Bounds on the smallest and the second-largest eigenvalues of strong products of regular +graphs are derived, which are expressed in terms of calculable parameters of its factors +(Proposition 2). +(7) A new lower bound on the second-largest eigenvalue of a k-fold strong power of a regular +graph is compared to the Alon–Boppana bound. Under a certain condition, the former bound +shows an improvement in its exponential growth rate as a function of k (Section III-C). +(8) Every non-complete and non-empty connected regular graph, whose Lov´asz θ-function is +below a certain value, is proved to have the property that almost all its strong powers are +highly non-Ramanujan (Proposition 3). + +3 +(9) Lower bounds on the chromatic number of strong products of graphs are expressed in terms +of the order and Lov´asz θ-function of each factor (Proposition 4). Their utility is exemplified, +while also leading to exact chromatic numbers in some cases. +The paper is structured as follows: Section II provides notation and a thorough review of +the backgrounds relevant to this work. Section III provides the results of this work, followed +by examples and discussions. It is composed of five subsections that address issues related to +the Lov´asz θ-function, Shannon capacity of graphs, Ramanujan graphs, the second-largest and +smallest eigenvalues of strong products or strong powers of graphs, and the chromatic numbers +of such graphs. Section IV proves the results in Section III. +II. +PRELIMINARIES +This section provides essential notation and preliminaries for this paper. The following standard +notation in set theory is used: N = {1, 2, . . .} is the set of natural numbers, R is the set of real +numbers, and [n] ≜ {1, . . . , n} with n ∈ N. The cardinality of a set A is a measure of its number +of elements; it is denoted by |A|, and (by definition) it is equal to its number of elements if A +is a finite set. +Let G be a graph, and let V(G) and E(G) denote, respectively, the sets of vertices and edges +in G. The order and size of a graph G are defined to be | V(G)| and | E(G)|, respectively. A +graph G is said to be finite if V(G) is a finite set. A pair of vertices are adjacent in a graph G +if these two vertices are the endpoints of an edge e ∈ E(G). +A graph is called simple if it has no loops (i.e., it has no edge with identical endpoints), and +if there are no multiple edges between any pair of adjacent vertices. A graph G is said to be +undirected if its edges have no directions; otherwise, it is a directed graph, a.k.a. a digraph. +Throughout this paper, it is assumed that the graphs under consideration are finite, undirected, +and simple. +An empty graph is an edgeless graph. A graph G is said to be an r-partite graph if its vertex +set V(G) is a disjoint union of r subsets such that every pair of vertices that are elements of an +identical subset are non-adjacent. If r = 2, then it is a bipartite graph. +A walk in a graph G is a sequence of its vertices such that (by definition) every pair of +consecutive vertices are adjacent. A path in a graph is a walk with no repeated vertices (in +other words, a path is a walk along the vertices of G such that no vertex can be visited twice). +The length of a path is defined as its number of edges. Hence, P = [v1, . . . , vℓ] is a path in a +graph G if {vi, vi+1} ∈ E(G) for all i ∈ [ℓ − 1], and all the vertices in the sequence {vi}ℓ +i=1 are +distinct; the endpoints of the path P are v1 and vℓ, and its length is equal to ℓ − 1. A cycle C +in a graph G is obtained by adding an edge to a path P such that it gives a closed walk (i.e., +a walk whose endpoints are identical). The cycle C = [v1, . . . , vℓ, v1] is of length ℓ, which is +obtained by adding the edge e = {vℓ, v1} ∈ E(G) to the above (ℓ − 1)–length path P; the two +identical endpoints of the cycle C are the vertex v1. +A graph is said to be connected if every two distinct vertices in that graph are connected by a +path (otherwise, it is a disconnected graph). The distance between a pair of distinct and connected +vertices in a graph is defined to be the length of the shortest path whose two endpoints are the +given pair of vertices. The distance between two disconnected vertices in a graph is defined to + +4 +be infinite, and the distance between a vertex and itself is set to be zero. The diameter of a +graph is the maximum distance between any pair of vertices in that graph. A connected and +finite graph has a finite diameter, and the diameter of a disconnected graph is (by definition) +infinite. +The following standard notation in graph theory is used: +(a) Kn denotes the complete graph with n ∈ N vertices, where every pair of distinct vertices +are adjacent; hence, K1 is an empty graph with a single vertex. +(b) Kn,m denotes the complete bipartite graph, which is a bipartite graph consisting of a vertex +set that is a disjoint union of two finite sets V1 and V2 of cardinalities n and m, respectively, +and a set of edges that are all the possible connections of a vertex in V1 and a vertex in V2. +(c) Pn denotes an (n − 1)–length path with n ∈ N, which is a graph with n vertices that forms +a path of length n − 1; in particular, P1 = K1. +(d) Cn denotes an n-length cycle, which is a graph with n ≥ 3 vertices that forms a cycle of +length n. +(e) K(m, r) denotes the Kneser graph with integers 1 ≤ r ≤ m. It has n = +�m +r +� +vertices, +represented by all r-subsets of [m]. Two vertices are adjacent in that graph if they are +represented by disjoint r-subsets. The graph K(m, r), provided that it has more than one +vertex, is a connected graph if and only if either m > 2r or (m, r) = (2, 1). +A subgraph is a graph that exists within another graph. More formally, F is a subgraph of +a graph G if V(F) ⊆ V(G) and E(F) ⊆ E(G). If F is a subgraph of G, we write F ⊆ G. A +spanning subgraph is obtained by edge deletions from the original graph, while its vertex set is +left unchanged. An induced subgraph is obtained by removing vertices from the original graph, +followed by the deletion of the edges that are adjacent to these removed vertices. +A clique in a graph G is a subset of pairwise adjacent vertices in G (in other words, the +induced subgraph of G on that subset is a complete subgraph). The maximum size of a clique +in G is called the clique number of G, and it is denoted by ω(G). Similarly, an independent set +(a.k.a. coclique) in G is a subset of pairwise non-adjacent vertices. The maximum size of an +independent set is called the independence number of G, and it is denoted by α(G). A proper +vertex coloring of G is a coloring of its vertices such that no pair of adjacent vertices are assigned +the same color. The smallest required number of colors for a proper coloring of the vertices in +G is called the chromatic number of G, and it is denoted by χ(G). +Let G = (V(G), E(G)) be a finite, undirected, and simple graph of order | V(G)| = n. Define +the adjacency matrix A = A(G) of the graph G to be an n-times-n symmetric matrix such that +A = (ai,j)1≤i,j≤n with ai,j = 1 if {i, j} ∈ E(G), and ai,j = 0 otherwise (hence, the diagonal +elements of A are in particular zeros). Let the eigenvalues of A (a.k.a. the eigenvalues of G) +be given in decreasing order by +λmax(G) = λ1(G) ≥ λ2(G) ≥ . . . ≥ λn(G) = λmin(G). +(1) +The spectrum of G consists of the eigenvalues of A, including their multiplicities. The terms +λ1(G), λ2(G), and λn(G) are referred to as the largest, second-largest, and smallest eigenvalues +of the graph G, respectively. The second-largest and smallest eigenvalues play a key role in +spectral graph theory, and the interested reader is referred to, e.g., [18], [25], [26], [88], [89]. + +5 +The number of edges that are incident to a vertex in a graph is called the degree of the vertex, +and a graph is said to be regular if its vertices have an identical degree. A regular graph whose +all vertices have a fixed degree d is called a d-regular graph, and it is said to have valency d. A +d-regular graph has a largest eigenvalue λ1(G) = d, with the all-ones column vector (of length +n) as an eigenvector. +Let G be a finite, simple and undirected graph. The complement of G, denoted by G, is defined +to have the same vertex set as G, and (by definition) any pair of distinct vertices in V(G) are +adjacent in G if and only if they are non-adjacent in the graph G. Hence, we have the equalities +| E(G)| + | E(G)| = +1 +2 n(n − 1), and V(G) = V(G). Furthermore, let Jn and In denote the +n-times-n all-ones and identity matrices, respectively. Then, by definition, it follows that the +adjacency matrices of the graph G and its complement G are related by the equality +A(G) = Jn − In − A(G). +(2) +Let G be a d-regular graph on n vertices, and let G be the complement graph. Then, by (2), +the spectra of G and G are related as follows [18, Section 1.3.2]: +λ1(G) = n − d − 1 = n − 1 − λ1(G), +(3) +λℓ(G) = −1 − λn+2−ℓ(G), +ℓ = 2, . . . , n. +(4) +Specifically, setting ℓ = n in (4) gives +λmin(G) = −1 − λ2(G). +(5) +A graph is called acyclic if it has no cycles, and a connected acyclic graph is called a tree. +A tree with n vertices has n − 1 edges, and for every pair of distinct vertices in a tree, there +is a unique path joining them (see, e.g., [25, Theorem 5.1.2]). A leaf in a tree is a vertex of +degree 1, and every tree contains at least two leaves (a tree is therefore not a regular graph, +unless it is the complete graph on two vertices K2). A disjoint union of trees is a forest (i.e., +a forest is an acyclic graph whose connected components are trees). In particular, a deletion of +an edge from a tree gives a forest of two trees. The interested reader is referred to, e.g., [25, +Chapter 5] for further properties and analysis related to trees and forests. +Two simple graphs G = (V(G), E(G)) and H = (V(H), E(H)) are said to be isomorphic if there +is a bijection (i.e., a one-to-one and onto mapping) f : V(G) → V(H) that preserves adjacency +and non-adjacency, i.e., {u, v} ∈ E(G) if and only if {f(u), f(v)} ∈ E(H). The notation G ∼= H +denotes that G and H are isomorphic graphs. An isomorphism from a graph to itself is called +an automorphism of the graph. +A graph is called self-complementary if G and G are isomorphic graphs. These include, e.g., +the trivial complete graph on one vertex K1, the length-3 path P4, and the 5-cycle graph C5. If G +is a self-complementary graph of order n, then the size of G is m = | E(G)| = 1 +2 +�n +2 +� += 1 +4 n(n−1). +Since only n or n − 1 can be even, either n ≡ 0 (mod 4) or n ≡ 1 (mod 4). For every such +n, there exists a recursive algorithm for constructing a self-complementary graph of order n +(see [20, Exercise 29]). More explicitly, if G is a self-complementary graph of order n, then +its disjoint union with the length-3 path P4 = [v1, v2, v3, v4], where each of the vertices v2 +and v3 in P4 is connected to all the vertices in G, gives a self-complementary graph of order + +6 +n + 4. Starting with a graph G that is equal to P4 or C5 (for graph orders of n = 4 or n = 5, +respectively) gives, by the above suggested recursive construction, a self-complementary graph +of order n for all integers n > 1 such that n ≡ 0 (mod 4) or n ≡ 1 (mod 4), respectively. +A graph G is vertex-transitive if for every two vertices of G, there is an automorphism of G +that maps one vertex to the other. Similarly, G is said to be an edge-transitive graph if for every +two edges {u1, v1} and {u2, v2} of G, there is an automorphism f : V(G) → V(G) of the graph +G, such that {u2, v2} = {f(u1), f(v1)}. A vertex-transitive graph is necessarily regular, but the +opposite does not hold in general. Unlike a vertex-transitive graph, an edge-transitive graph is +not necessarily regular. Types of graph transitivity are studied, e.g., in [37, Chapters 3–4]. +All the eigenvalues of a d-regular graph G are bounded in absolute value by d, and the largest +eigenvalue is equal to d with an eigenvector that is equal to the all-ones column vector (see, +e.g., [25, Proposition 12.1.1]). The multiplicity of the largest eigenvalue of a d-regular graph +is 1 if and only if G is connected (see, e.g., [25, Theorem 4.5.2 and Proposition 12.1.1]). A +graph G has its eigenvalues symmetric around zero (including their multiplicities) if and only +if G is bipartite (see, e.g., [25, Theorem 4.3.2]). Hence, the smallest eigenvalue of a d-regular +bipartite graph is equal to −d. Moreover, if G is d-regular and connected, then G is bipartite if +and only if −d is an eigenvalue of its adjacency matrix (see, e.g., [25, Proposition 12.1.1]). For +a d-regular graph G, let +λ(G) ≜ +max +ℓ: λℓ(G)̸=±d +��λℓ(G) +��. +(6) +A d-regular graph G is called Ramanujan if +λ(G) ≤ 2 +√ +d − 1. +(7) +The reason for the expression in the right-hand side of (7) is related to the Alon–Boppana bound, +which addresses the question of how small can the second-largest eigenvalue be for a connected +d-regular graph or for a sequence of connected d-regular graphs whose number of vertices tends +to infinity (the value of the parameter d is kept fixed). It states that for every d-regular graph G +on n vertices, with d ≥ 3, +λ2(G) ≥ 2 +√ +d − 1 − O +� +(logd n)−1� +. +(8) +A non-asymptotic version of (8) appears in [78] (see also [25, Theorem 12.2.1]). The Alon– +Boppana bound was first mentioned in [6, p. 95], and it was analyzed, e.g., in [1], [2], [3], [21], +[22], [33], [34], [35], [61], [78], [79]. Moreover, all eigenvalues of a tree, with maximum degree +d ≥ 2, are (in absolute value) at most 2 +√ +d − 1 (see [73, Theorem 1]). Examples of Ramanujan +graphs include: +(a) The complete d-regular graph Kd+1, with d ≥ 2, whose eigenvalues are equal to d with +multiplicity 1, and −1 with multiplicity d; +(b) The complete bipartite graph Kd,d, with d ≥ 2, is a d-regular graph whose two nonzero +eigenvalues are ±d (each of multiplicity 1), and its other 2d − 2 eigenvalues are zeros. +(c) The Petersen graph, which is isomorphic to the Kneser graph K(5, 2), is a Ramanujan graph +since it is 3-regular with the distinct eigenvalues 3, −1, and −2. +The interested reader is referred to [73] for a recent survey paper on Ramanujan graphs, and +to references therein (see, e.g., [69] that proves the existence of infinite families of bipartite +Ramanujan graphs of every degree greater than 2, followed by the extension of that result + +7 +to bi-regular bipartite graphs; the proof in [69] uses an original technique for controlling the +eigenvalues of some random matrices). +Let G be a d-regular graph of order n. The graph G is said to be a strongly regular graph if +there exist nonnegative integers λ and µ such that the following two conditions hold: +• +Every pair of adjacent vertices have exactly λ common neighbors; +• +Every pair of distinct and non-adjacent vertices have exactly µ common neighbors. +Such a strongly regular graph is denoted by srg(n, d, λ, µ). Some basic properties of strongly +regular graphs are next introduced, which also serve in our analysis. +(a) The complement of a strongly regular graph is also strongly regular. More explicitly, the +complement of srg(n, d, λ, µ) is given by srg(n, n − d − 1, n − 2d + µ − 2, n − 2d + λ). +(b) The four parameters of a strongly regular graph srg(n, d, λ, µ) satisfy the relation +(n − d − 1) µ = d (d − λ − 1). +(9) +(c) A strongly regular graph srg(n, d, λ, µ) has at most three distinct eigenvalues. If it is +connected, then λ1(G) = d (multiplicity 1), and the other two distinct eigenvalues are +p1,2 = 1 +2 +� +λ − µ ± +� +(λ − µ)2 + 4(d − µ) +� +, +(10) +whose respective multiplicities are given by +m1,2 = 1 +2 +� +n − 1 ∓ 2d + (n − 1)(λ − µ) +� +(λ − µ)2 + 4(d − µ) +� +. +(11) +Since multiplicities of eigenvalues must be nonnegative integers, their expressions in (11) +provide further constraints on the values of n, d, λ and µ (in addition to equality (9)). +(d) A connected regular graph with exactly three distinct eigenvalues is strongly regular. +(e) A strongly regular graph srg(n, d, λ, µ), with µ > 0, is a connected graph whose diameter +is equal to 2. This holds since two non-adjacent vertices have µ > 0 common neighbors, +so the distance between any pair of non-adjacent vertices is equal to 2. This can be also +explained by spectral graph theory since the diameter of a connected graph is strictly smaller +than the number of its distinct eigenvalues (see [25, Theorem 4.4.1]). In light of that, the +above claim about the diameter holds for all graphs that are connected and strongly regular +since these graphs only have three distinct eigenvalues. +(f) If µ = 0, the strongly regular graph is disconnected, and it is a disjoint union of equal- +sized complete graphs (i.e., a disjoint union of cliques of the same size). A disjoint union +of an arbitrary number ℓ ≥ 2 of equal-sized complete graphs, Kd+1, has the parameters +srg((d + 1)ℓ, d, d − 1, 0). In that case, d = p1 (see (10)), so the largest and second-largest +eigenvalues coincide (by (11), that common eigenvalue has multiplicity m1 + 1 = ℓ in the +graph spectrum). A strongly regular graph G is called primitive if both G and its complement +G are connected graphs. Otherwise, G is called imprimitive. An imprimitive graph is either +a disjoint union of equal-sized complete graphs or its complement, which is a complete +multipartite graph. A strongly regular graph G is imprimitive if and only if 0 or −1 is an +eigenvalue of G. + +8 +(g) Let G be a primitive strongly regular graph srg(n, d, λ, µ) with the largest eigenvalue d +(multiplicity 1), second-largest eigenvalue r = p1 (multiplicity m1), and smallest eigenvalue +s = p2 (multiplicity m2). By (3) and (4), the complement G is a primitive strongly regular +graph, having the largest eigenvalue n − d − 1 (multiplicity 1), second-largest eigenvalue +−1 − s (multiplicity m2), and smallest eigenvalue −1 − r (multiplicity m1). Each of these +primitive strongly regular graphs has three distinct eigenvalues. +The reader is referred to [19], which is focused on properties and constructions of strongly +regular graphs. +Part of this work is focused on the strong product of graphs, which is defined as follows. +Definition 1: Let G1 and G2 be two graphs. The strong product G = G1 ⊠ G2 is a graph +whose vertex set is V(G) = V(G1) × V(G2) (a Cartesian product), and distinct vertices {u1, u2} +and {v1, v2} in G are adjacent if one of the following three conditions is satisfied: +(a) u1 = v1 and {u2, v2} ∈ E(G2), +(b) {u1, v1} ∈ E(G1) and u2 = v2, +(c) {u1, v1} ∈ E(G1) and {u2, v2} ∈ E(G2). +A strong product of graphs is commutative in the sense that +G1 ⊠ G2 ∼= G2 ⊠ G1, +(12) +and it is also associative in the sense that +(G1 ⊠ G2) ⊠ G3 ∼= G1 ⊠ (G2 ⊠ G3), +(13) +for every three graphs G1, G2 and G3 (see [46, Proposition 4.1]). +This paper relies on the Lov´asz θ-function of a graph [65], which is next introduced. +Definition 2: Let G be a simple graph. An orthogonal representation of G in the d-dimensional +Euclidean space (Rd) assigns to each vertex i ∈ V(G) a vector ui ∈ Rd such that uT +i uj = 0 if +{i, j} /∈ E(G). In other words, the vertices of a simple graph are assigned vectors in Rd such that +the vectors that are assigned to any pair of distinct and non-adjacent vertices of that graph are +orthogonal. An orthonormal representation of G is an orthogonal representation of that graph +such that all representing vectors have unit length. +Remark 1: In an orthogonal representation of a graph G, non-adjacent vertices are mapped +to orthogonal vectors, although adjacent vertices are not necessarily mapped to non-orthogonal +vectors. If the latter condition is satisfied, then the orthogonal representation is said to be faithful. +Definition 3: Let G be a finite, undirected and simple graph. Its Lov´asz θ-function is given by +θ(G) ≜ min +u,c +max +i∈V(G) +1 +� +cTui +�2 , +(14) +where the minimum is taken over all orthonormal representations {ui : i ∈ V(G)} of G, and all +unit vectors c. The unit vector c is called the handle of the orthonormal representation. +By the Cauchy-Schwarz inequality, +��cTui +�� ≤ ∥c∥ ∥ui∥ = 1, so θ(G) ≥ 1 with equality if and +only if G is a complete graph. +The Lov´asz θ-function of a graph can be written as a semidefinite program, which satisfies +strong duality ([41], [57], [66] and [67, Section 11.2]). This enables to compute the value of + +9 +θ(G) in polynomial time [41]. More precisely, there is an algorithm that computes, for every +graph G and every ε > 0, a real number t such that +��θ(G) − t +�� < ε, where the running time of +the algorithm is polynomial in n ≜ | V(G)| and log +� 1 +ε +� +[67, Theorem 11.11]. +The following properties of the Lov´asz θ-function are used throughout this paper: +(a) The sandwich theorem ([41], [57], [66, Lemma 3.2.4], [67, Theorem 11.1]) is stated in the +two equivalent forms +α(G) ≤ θ(G) ≤ χ(G), +(15) +ω(G) ≤ θ(G) ≤ χ(G). +(16) +(b) [65, Theorem 7]: The Lov´asz θ-function factorizes for the strong product of graphs, i.e., +θ(G1 ⊠ G2) = θ(G1) θ(G2). +(17) +(c) [65, Corollary 2] and [65, Theorem 8]: +θ(G) θ(G) ≥ | V(G)|, +(18) +with equality in (18) if the graph G is vertex-transitive. +(d) [65, Theorem 9]: Let G be a d-regular graph of order n. Then, +θ(G) ≤ − n λn(G) +d − λn(G), +(19) +where λn(G) < 0, unless G is an empty graph. Equality holds in (19) if G is edge-transitive. +(e) Two simple observations relating the Lov´asz θ-functions of a graph and its subgraphs: +• +If F is a spanning subgraph of a graph G, then θ(F) ≥ θ(G). +• +If F is an induced subgraph of a graph G, then θ(F) ≤ θ(G). +(f) [4, Theorem 2]: Although unrelated to the analysis in this paper, another interesting property +of the Lov´asz θ-function is given by the identity +sup +H +α(G ⊠ H) +θ(G ⊠ H) = 1, +(20) +which holds for every simple, finite, and undirected graph G, where the supremum is taken +over all such graphs H. This shows that the leftmost inequality in (15) can be made arbitrarily +tight by looking at the strong product of the given graph G with a suitable graph H. +The Shannon capacity of a simple, finite and undirected graph G was introduced in [86] +to determine the maximum information rate that enables error-free communication. To that +end, a discrete memoryless communication channel is represented by a confusion graph G that +is constructed as follows. The vertices in the graph are represented by the symbols of the +input alphabet to that channel, and any two distinct vertices in that graph are adjacent if the +corresponding two input symbols are not distinguishable by the channel (in the sense that there +exists an output symbol such that the transition probabilities from each of these two input +symbols to that output symbol are strictly positive). This means that the exact values of the +positive transition probabilities of the channel, as well as the output alphabet of the channel, are +irrelevant to the construction of the confusion graph G. The rationality in doing so is the interest +to pictorially represent (by a graph) all those pairs of input symbols that are not distinguishable + +10 +by the channel. Consider a transmission of k-length strings. The k-th confusion graph of the +channel is defined as +G⊠ k ≜ G ⊠ . . . ⊠ G +� +�� +� +k−1 strong products +, +(21) +which is the k-fold strong power of G. This is because the independence number of G⊠ k is +equal to the maximum number of k-length strings at the channel input that can be transmitted +with error-free communication (indeed, a pair of non-adjacent vertices in G⊠ k represent k-length +strings that can be distinguished by the channel, as a result of having a common position in +these two input strings where the corresponding two symbols at that position are distinguishable +by the channel). Consequently, the maximum information rate per symbol that is achievable by +using input strings of length k is equal to 1 +k log α(G⊠ k) = log +k� +α(G⊠ k), for all k ∈ N (i.e., +it is the logarithm of the maximum number of k-length input strings that are distinguishable by +the channel, normalized by the length k). The Shannon capacity of a graph G is defined to be +the (exponent of the) maximum information rate per symbol that is achievable with error-free +communication, where the transmission takes place over a discrete memoryless channel whose +confusion graph is equal to G, and the length of the input strings to the channel is unlimited. It +is denoted by Θ(G) (recall that the Lov´asz θ-function is denoted by θ(G)). Taking the supremum +over k, the Shannon capacity of G is given by (see [86], and [5, Chapter 42]) +Θ(G) = sup +k∈N +k� +α(G⊠ k). +(22) +The Shannon capacity can be rarely computed exactly (see, e.g., [7], [9], [43], [51], [52], [58], +[65], [86]). Analytical observations that also explain why it is, in general, even difficult to +approximate it are addressed in [10] and [39]. Calculable upper bounds on Θ(G) were derived +by Shannon [86], Lov´asz [65], Haemers [43], and more recently by Hu et al. [51]. The Lov´asz +θ-function θ(G) is a calculable upper bound on the graph capacity Θ(G), i.e., +α(G) ≤ Θ(G) ≤ θ(G), +(23) +where the leftmost inequality in (23) follows from (22) (by setting k = 1), and the rightmost +inequality in (23) is [65, Theorem 1]. In regard to the rightmost inequality in (23), it is also +shown in [9] that the Lov´asz θ-function of a graph, θ(G), cannot be upper bounded by any +function of its Shannon capacity Θ(G). As mentioned above, the computational task of the +Lov´asz θ-function, θ(G), is in general feasible by semidefinite programming. Fundamental graph +parameters such as its Shannon capacity Θ(G), independence number α(G), clique number ω(G), +and chromatic number χ(G) are all NP-hard problems. The polynomial-time computability of +the Lov´asz θ-function of a graph makes inequalities (15), (16), and (23) very useful in obtaining +polynomial-time computable upper bounds on the independence number, clique number, and the +Shannon capacity of a graph, as well as having such a computable lower bound on the chromatic +number. + +11 +III. +THEOREMS, DISCUSSIONS AND EXAMPLES +The present section provides the results of this work, followed by examples and discussions. +It is composed of five subsections that address issues related to the Lov´asz θ-function, Shannon +capacity of graphs, Ramanujan graphs, eigenvalues, and chromatic numbers of strong graph +products or strong graph powers. +A. Bounds on Lov´asz θ-function, and an exact result for strongly regular graphs +Let G be a d-regular graph with n vertices, and let G be the complement graph of G that is an +(n−d−1)-regular graph of order n. An upper bound on θ(G) and a lower bound on θ(G) were +obtained by Lov´asz, expressed in terms of the smallest eigenvalue of the adjacency matrix of +G (see [65, Theorem 9] and [65, Corollary 3]). The novelties in the next result (Proposition 1) +are as follows: +(a) It forms a counterpart of a bound by Lov´asz [65, Theorem 9], providing a lower bound on +θ(G) and an upper bound on θ(G) that are both expressed in terms of the second-largest +eigenvalue of the adjacency matrix of G. +(b) It asserts that these two pairs of upper and lower bounds on θ(G) and θ(G) are tight for +the family of strongly regular graphs. This gives a simple closed-form expression of the +Lov´asz θ-function of a strongly regular graph srg(n, d, λ, µ) (and the complement graph) as +a function of its four parameters. +(c) Further sufficient conditions for the tightness of these bounds are provided. +Proposition 1: Let G be a d-regular graph of order n, which is a non-complete and non-empty +graph. Then, the following bounds hold for the Lov´asz θ-function of G and its complement G: +(a) +n − d + λ2(G) +1 + λ2(G) +≤ θ(G) ≤ − nλn(G) +d − λn(G). +(24) +• +Equality holds in the leftmost inequality of (24) if G is both vertex-transitive and +edge-transitive, or if G is a strongly regular graph; +• +Equality holds in the rightmost inequality of (24) if G is edge-transitive, or if G is a +strongly regular graph. +(b) +1 − +d +λn(G) ≤ θ(G) ≤ n� +1 + λ2(G)� +n − d + λ2(G). +(25) +• +Equality holds in the leftmost inequality of (25) if G is both vertex-transitive and +edge-transitive, or if G is a strongly regular graph; +• +Equality holds in the rightmost inequality of (25) if G is edge-transitive, or if G is a +strongly regular graph. +Proof: See Section IV-A1. +Remark 2: In light of the sufficient conditions for each of the four inequalities in Proposition 1 +to hold with equality, define the following subfamilies of regular graphs: +• +Let G1 be the family of graphs G such that G is both vertex-transitive and edge-transitive; + +12 +• +Let G2 be the family of regular and edge-transitive graphs; +• +Let G3 be the family of graphs G such that G is regular and edge-transitive; +• +Let G4 be the family of graphs that are both vertex-transitive and edge-transitive; +• +Let G5 be the family of the strongly regular graphs. +We next show by explicit examples, obtained by using the SageMath software [85], that none +of the families G1, G2 G3 and G4 is included in the family G5, and vice versa. +(a) The Cameron graph is a strongly regular graph srg(231, 30, 9, 3) (see [19, Section 10.54]). +Its complement is vertex-transitive (hence, regular), but not edge-transitive. This shows that +G5 ̸⊆ G3, so also G5 ̸⊆ G1. +(b) The complement of the Cameron graph is a strongly regular graph srg(231, 200, 172, 180); +it is vertex-transitive (hence, regular), but not edge-transitive. This shows that G5 ̸⊆ G2, so +also G5 ̸⊆ G4. +(c) The Foster graph is 3-regular on 90 vertices (see [19, p. 305]), which is vertex-transitive +and edge-transitive, but it is not strongly regular. This shows that G4 ̸⊆ G5, so also G2 ̸⊆ G5. +(d) The complement of the Foster graph is an 86-regular graph on 90 vertices, whose complement +(i.e., the Foster graph) is vertex-transitive and edge-transitive, but it is not strongly regular. +This shows that G1 ̸⊆ G5, so also G3 ̸⊆ G5. +The next result provides a closed-form expression of the Lov´asz θ-function for strongly regular +graphs (and their strongly regular complements). This result relies on Proposition 1 and the +closed-form expressions of the distinct eigenvalues of a strongly regular graph. +Corollary 1: Let G be a strongly regular graph with parameters srg(n, d, λ, µ). Then, +θ(G) = n (t + µ − λ) +2d + t + µ − λ, +(26) +and +θ(G) = +n +θ(G) +(27) += 1 + +2d +t + µ − λ, +(28) +where +t ≜ +� +(µ − λ)2 + 4(d − µ). +(29) +Furthermore, if 2d + (n − 1) (λ − µ) ̸= 0, then θ(G) and θ(G) are rational numbers. +Proof: See Section IV-A2. +Remark 3: By (27), if G is a strongly regular graph on n vertices, then θ(G) θ(G) = n. This +relation is also known to hold if the graph G is vertex-transitive [65, Theorem 8]. It should be +noted that not all the strongly regular graphs are necessarily vertex-transitive, so the observation +here is not implied by [65, Theorem 8]. As a counter example for strongly regular graphs that +are not vertex-transitive, consider the Chang graphs. These are three of the existing four non- +isomorphic strongly regular graphs with parameters srg(28, 12, 6, 4) [19, Section 10.11] (the +fourth such graph, denoted by T8, is the line graph of the complete graph on 8 vertices K8). The +three Chang graphs are not vertex-transitive and also not edge-transitive (in contrast to T8 that +is vertex-transitive and edge-transitive), as it can be verified by the SageMath software [85]. + +13 +Remark 4: The 5-cycle C5 is a strongly regular graph srg(5, 2, 0, 1). Its Lov´asz θ-function +coincides with its Shannon capacity, being equal to +√ +5 (see [65, Theorem 2]). Although it is an +irrational number, it is consistent with Corollary 1 since 2d+(n−1) (λ−µ) = 2·2+4(0−1) = 0. +B. Eigenvalue inequalities, strongly regular graphs, and Ramanujan graphs +The present subsection relies on Proposition 1, with the following contributions: +(a) Derivation of inequalities that relate the second-largest and smallest eigenvalues of a regular +graph. These inequalities hold with equality if and only if the graph is strongly regular. +(b) Derivation of bounds on parameters of Ramanujan graphs. +(c) A more general result is presented for a sequence of regular graphs whose degrees scale +sub-linearly with the orders of these graphs, and their orders tend to infinity. +This subsection is composed of Corollaries 2–6, which all rely on Proposition 1. It starts by +providing eigenvalue inequalities. +Corollary 2: Let G be a d-regular graph of order n, which is non-complete and non-empty. +Then, +λn(G) ≤ −d (n − d + λ2(G)) +d + (n − 1) λ2(G), +(30) +or equivalently, +λ2(G) ≥ −d (n − d + λn(G)) +d + (n − 1) λn(G). +(31) +Furthermore, (30) and (31) hold with equality if and only if G is a strongly regular graph. +Proof: See Section IV-B1. +The next result introduces, in part, Nordhaus–Gaddum type inequalities for the second-largest +and smallest eigenvalues of regular graphs, which are tight for all strongly regular graphs. +Regarding Nordhaus–Gaddum type inequalities, the interested reader is referred to [76], [77], +and [80]. +Corollary 3: Let G be a d-regular graph of order n, which is non-complete and non-empty, +and let +gℓ(G) ≜ λℓ(G) − d(n − d + λℓ(G)) +d + (n − 1)λℓ(G), +∀ ℓ ∈ [n]. +(32) +The following holds: +(a) +gn(G) ≤ −1 ≤ g2(G), +(33) +and the two inequalities in (33) hold with equality if and only if G is strongly regular. +(b) If G is a strongly regular graph, then the number of distinct values in the sequence {gℓ(G)}n +ℓ=1 +is either 2 or 3, and +• +it is equal to 2 if the multiplicities of the second-largest and smallest eigenvalues of +G are identical in the subsequence (λ2(G), . . . , λn(G)); +• +it is otherwise equal to 3. + +14 +(c) If G is self-complementary, then +λ2(G) ≥ 1 +2 +�√n − 1 +� +, +(34) +λn(G) ≤ − 1 +2 +�√n + 1� +. +(35) +(d) If G is self-complementary and strongly regular, then (34) and (35) hold with equality. +Proof: See Section IV-B2. +Example 1: Item (b) of Corollary 3 refers to the dichotomy in the number of distinct values in +the sequence {gℓ(G)}n +ℓ=1. This statement applies to all strongly regular graphs (either connected +or disconnected). We believe that the following example contributes to its clarity, in addition +to its formal proof in Section IV-B2. Let G be a disjoint union of the three complete graphs +K2, which gives a disconnected strongly regular graph. Its complement is the complete 3-partite +graph G = K2,2,2, so n = 6, and +{λℓ(G)}6 +ℓ=1 = (1, 1, 1, −1, −1, −1), +(36) +{λℓ(G)}6 +ℓ=1 = (4, 0, 0, 0, −2, −2), +(37) +{gℓ(G)}6 +ℓ=1 = (3, −1, −1, 1, −1, −1). +(38) +By (36), the multiplicities of λ2(G) and λn(G) in the spectrum of G are identical, but these +multiplicities are distinct in the subsequence {λℓ(G)}6 +ℓ=2 = (1, 1, −1, −1, −1). Hence, the fact +that the sequence {gℓ(G)}6 +ℓ=1 gets three distinct values is indeed consistent with the claim in +Item (b) of Corollary 3. +Example 2: Let G be the Hall-Janko graph, which is a strongly regular graph with parameters +srg(100, 36, 14, 12) (see Section 10.32 of [19]). As a numerical verification of Item (b) in +Corollary 3 (and its proof in Section IV-B2), the sequence {gℓ(G)}100 +ℓ=1 in (32) gets the three +distinct values: n − d − 2 = 62 (at ℓ = 1), −1 (for 2 ≤ ℓ ≤ 37 or 65 ≤ ℓ ≤ 100), and 9 +(for 38 ≤ ℓ ≤ 64). The third value (9) is attained by the sequence {gℓ(G)} twenty-seven times. +By the proof of Item (b) in Corollary 3, the multiplicity of the third value (27) is equal to the +absolute value of the difference between the multiplicities of the second-largest and smallest +eigenvalues of the graph G. The spectrum of the graph G is given by 361636(−4)63 (this can +be verified by (10) and (11)), and the above difference (in absolute value) is indeed equal +to |63 − 36| = 27. Next, let G be the 5-cycle graph C5, which is a strongly regular graph +with parameters srg(5, 2, 0, 1). The second-largest and smallest eigenvalues of G are equal to +1 +2( +√ +5 − 1) and − 1 +2( +√ +5 + 1), respectively, and their multiplicities coincide, being both equal +to 2. In light of Item (b) in Corollary 3, the sequence {gℓ(G)}5 +ℓ=1 in (32) gets only two distinct +values: n − d − 2 = 1 at ℓ = 1, and −1 for 2 ≤ ℓ ≤ 5. +Remark 5: We discuss here an implication of the conditions for equalities in Proposition 1. +Let G be a d-regular graph. Inequality (30) holds with equality if and only if both inequalities +in (24) hold with equality. By Item (a) of Proposition 1, the leftmost inequality in (30) holds +with equality if G is both vertex-transitive and edge-transitive, and the rightmost inequality in +(30) holds with equality if G is edge-transitive. Combining both sufficient conditions for the +two inequalities in (30) to hold with equality, it follows that a sufficient condition for equality + +15 +in (30) is given by the requirement that G and G are both vertex-transitive and edge-transitive +(recall that G is vertex-transitive if and only if G is so). +By Corollary 2, inequality (30) holds with equality if and only if G is strongly regular. +By a comparison of the former (sufficient) condition with the latter (necessary and sufficient) +condition for equality to hold in (30), it follows that if G and G are both vertex-transitive and +edge-transitive, then G is strongly regular. +It should be noted, with gratitude to an anonymous reviewer, that a stronger result can be +obtained by replacing the requirement of the vertex-transitivity with the weaker condition of +regularity. Namely, if G is regular, and G and G are both edge-transitive, then G is strongly +regular. This can be shown as follows. +• +(G is edge transitive) ⇒ (every edge in G is contained in the same number of triangles) +⇔ (every pair of adjacent vertices in G has the same number of common neighbors); +• +(G is edge transitive) ⇒ (for every edge {u, v} ∈ E(G), the same number of vertices are +not adjacent in G to either u or v) ⇔ (every pair of non-adjacent vertices in G has the +same number of common neighbors); +• +G is regular (by assumption); +and these observations correspond to the conditions in the definition of a strongly regular graph. +Example 3: The Schl¨afli graph G1 is 16-regular on 27 vertices. Both G1 and its complement +are edge-transitive. By Remark 5, the graph G1 is indeed strongly regular; its parameters are +given by srg(27, 16, 10, 8) [19, Section 10.10]. +Consider, on the other hand, the Shrikhande graph G2 which is 6-regular on 16 vertices. It +is a strongly regular graph with parameters srg(16, 6, 2, 2) [19, Section 10.6]. The graph G2 is +edge-transitive, but its complement is not edge-transitive (this was verified by the SageMath +software [85]). In addition, the Cameron graph G3 is a strongly regular graph srg(231, 30, 9, 3) +[19, Section 10.54]. It can be verified that it is edge-transitive, and that its complement is not +edge-transitive. This shows that the family of regular graphs G with the property that G and its +complement G are both edge-transitive is a strict subset of the family of strongly regular graphs. +Remark 6: In continuation to Example 3, strongly-regular graphs G such that G and G are +both edge-transitive include, e.g., the Hall-Janko, Hoffman-Singleton, Mesner, Petersen, Schl¨afli, +Sims-Gewirtz, and Suzuki graphs (this has been verified by the SageMath software [85]; for the +introduction of these graphs, the reader is referred to [19]). It also includes the infinite families +of Paley and Peisert graphs ([72], [81]), which are self-complementary and arc-transitive graphs. +(All self-complementary and arc-transitive graphs are strongly regular). +Remark 7: In connection to Remark 5, two related statements have been proved by Neumaier: +(a) A connected, edge-transitive and strongly regular graph is vertex-transitive [74, Lemma 1.3]. +(b) A vertex-transitive and edge-transitive graph containing a regular clique is strongly regular +(see [75, Corollary 2.4]). (A clique C is called regular if every vertex not in C is adjacent +to the same positive number of vertices in C). +Strongly regular graphs that are both vertex- and edge-transitive are studied in [71]. +Corollary 4: Let {Gℓ}ℓ∈N be a sequence of graphs where Gℓ is dℓ-regular of order nℓ, and +lim +ℓ→∞nℓ = ∞, +lim +ℓ→∞ +dℓ +nℓ += 0. +(39) + +16 +Then, +lim sup +ℓ→∞ +ω(Gℓ) ≤ a, +(40) +lim inf +ℓ→∞ +χ(Gℓ) +nℓ +≥ 1 +a, +(41) +with +a ≜ 1 + lim sup +ℓ→∞ +⌊λ2(Gℓ)⌋. +(42) +Proof: See Section IV-B3. +Corollary 5: Let {Gℓ}ℓ∈N be a sequence of connected, Ramanujan, and d-regular graphs where +d ∈ N is fixed, Gℓ is a graph on nℓ vertices, and lim +ℓ→∞nℓ = ∞. Then, +lim sup +ℓ→∞ +ω(Gℓ) ≤ 1 + ⌊2 +√ +d − 1⌋, +(43) +lim inf +ℓ→∞ +θ(Gℓ) +nℓ +≥ +1 +1 + 2 +√ +d − 1, +(44) +lim inf +ℓ→∞ +χ(Gℓ) +nℓ +≥ +1 +1 + ⌊2 +√ +d − 1⌋. +(45) +Proof: See Section IV-B4. +In continuation to Corollary 5, the following result provides non-asymptotic bounds on some +graph parameters. +Corollary 6: Let G be a connected, Ramanujan, and d-regular graph with n vertices. Then, +ω(G) ≤ +� n +� +1 + 2 +√ +d − 1 +� +n − d + 2 +√ +d − 1 +� +, +(46) +θ(G) ≥ n − d + 2 +√ +d − 1 +1 + 2 +√ +d − 1 +, +(47) +χ(G) ≥ +�n − d + 2 +√ +d − 1 +1 + 2 +√ +d − 1 +� +. +(48) +Proof: See Section IV-B5. +Remark 8: Inequalities (40), (43), and (46) can be also obtained from [42, Theorem 2.1.3], +which states that for an arbitrary simple, finite, and undirected graph G on n vertices, whose +maximal degree is given by ∆(G), the following bound on the clique number of G holds: +ω(G) ≤ +n +� +d + λ1(G) λ2(G) +� +dn − ∆(G)2 + λ1(G) λ2(G). +(49) +For a d-regular graph G on n vertices, we have λ1(G) = d = ∆(G), which then specializes (49) +to (see [42, Theorem 2.1.4]) +ω(G) ≤ n +� +1 + λ2(G) +� +n − d + λ2(G). +(50) +Inequality (46) can be also obtained from (50), combined with the satisfiability of the inequality +��λ2(G) +�� ≤ 2 +√ +d − 1 if G is a connected, d-regular, and Ramanujan graph, together with the fact +that the right-hand side of (50) is monotonically increasing in the parameter λ2(G). + +17 +C. Bounds on eigenvalues of strong products of regular graphs +A small second-largest eigenvalue of the adjacency matrix of a regular graph implies that the +graph is a good expander (see, e.g., [25, Theorem 12.1.2], and [6], [28], [50], [53], [59], [68]). +The Alon–Boppana bound in (8) is a lower bound on the second-largest eigenvalue of a regular +graph (see Section II). By the Alon–Boppana bound, for every sequence {Gk}∞ +k=1 of d-regular +graphs, with a fixed integer d ≥ 3 and orders tending to infinity (i.e., lim +k→∞| V(Gk)| = ∞), the +second-largest eigenvalues of their adjacency matrices satisfy +lim inf +k→∞ λ2(Gk) ≥ 2 +√ +d − 1, +(51) +(see, e.g., [25, Theorem 12.1.2]). This lower bound is asymptotically tight, and its tightness can +be strengthened beyond the second-largest eigenvalue. More explicitly, by Serre’s theorem [61], +for every fixed integer d ≥ 3 and (an arbitrarily small) ε > 0, there exists a positive constant +c = c(ε, d) such that every d-regular graph with n vertices has at least cn eigenvalues that are +larger than or equal to 2 +√ +d − 1 − ε. In other words, Serre’s theorem states that a non-vanishing +fraction of the n eigenvalues of every d-regular graph has the property of satisfying the Alon– +Boppana lower bound within the desired accuracy (see a simplified proof in [25, Theorem 12.2.3] +or [21, Theorem 1]). Analogous theorems, concerning the least eigenvalues of d-regular graphs, +also hold under an additional hypothesis that the graphs do not have odd cycles below a certain +length (see [21, Section 4]). It overall justifies the definition of Ramanujan graphs (see (7)) as +d-regular graphs, with d ≥ 3, whose all non-trivial eigenvalues are (in absolute value) at most +2 +√ +d − 1. +For a k-fold strong power of a d-regular graph, the degree is increased exponentially in k, +being equal to dk = (1 + d)k − 1. It is therefore of interest to obtain an alternative lower bound +on the second-largest eigenvalue of strong products of regular graphs. Its derivation is motivated +by the significance of strong products, and in particular strong powers of a given graph: +(1) The graph capacity in Shannon’s problem of zero-error communication [86] is given in (22), +which is expressed in terms of the independence numbers of all k-fold strong powers of the +graph (with k ∈ N); +(2) The Witsenhausen rate [94] in the zero-error source coding problem, with perfect side +information at the receiver, is expressed in a dual form to (22), where the independence +numbers of k-fold strong powers of a graph (with k ∈ N) are replaced by their chromatic +numbers, and the supremum over k is replaced by an infimum (see [7, Section 3]); +(3) There exists a polynomial-time algorithm that finds the unique prime factorization of any +connected graph under the operation of strong graph multiplication [31]. +It is demonstrated in this section that, under a certain condition, the suggested lower bound +on the second-largest eigenvalue of the k-fold strong power of a regular graph offers a larger +exponential growth rate in k, as compared to the Alon–Boppana bound. +The following proposition provides a lower bound on the second-largest eigenvalue, and an +upper bound on the smallest eigenvalue of the adjacency matrices of strong products of regular +graphs. Their derivation relies on Proposition 1, jointly with the factorization property in (17). +Both bounds are expressed in terms of the Lov´asz θ-function of each factor. This enables to +obtain analytical bounds on the second-largest and smallest eigenvalues of a k-fold strong power + +18 +of a regular graph, with a low computational complexity that is not affected by k. This stays +in contrast to the computational complexity of these eigenvalues, which significantly increases +with k. +Proposition 2: Let G1, . . . , Gk be regular graphs such that, for all ℓ ∈ [k], the graph Gℓ is +dℓ-regular of order nℓ. The following bounds hold for their strong product: +(a) Unless all Gℓ (with ℓ ∈ [k]) are complete graphs, then +λ2(G1 ⊠ . . . ⊠ Gk) ≥ +k� +ℓ=1 +nℓ − +k� +ℓ=1 +(1 + dℓ) +k� +ℓ=1 +θ(Gℓ) − 1 +− 1 +(52) +≥ +k� +ℓ=1 +nℓ − +k� +ℓ=1 +(1 + dℓ) +k� +ℓ=1 +� +− nℓ λmin(Gℓ) +dℓ − λmin(Gℓ) +� +− 1 +− 1, +(53) +and inequality (53) holds with equality if, for all ℓ ∈ [k], the regular graph Gℓ is either +edge-transitive or strongly regular. +(b) Unless all Gℓ (with ℓ ∈ [k]) are empty graphs, then +λmin(G1 ⊠ . . . ⊠ Gk) ≤ − +k� +ℓ=1 +(1 + dℓ) − 1 +k� +ℓ=1 +� nℓ +θ(Gℓ) +� +− 1 +. +(54) +If each regular graph Gℓ (with ℓ ∈ [k]) is either edge-transitive or strongly regular, then (54) +can be expressed in an equivalent form as +λmin(G1 ⊠ . . . ⊠ Gk) ≤ − +k� +ℓ=1 +(1 + dℓ) − 1 +k� +ℓ=1 +� +1 − +dℓ +λmin(Gℓ) +� +− 1 +. +(55) +Proof: See Section IV-C1. +Remark 9: As a sanity check, it would be in place to verify that the lower bound on the +second-largest eigenvalue in the right-hand side of (52) is smaller than or equal to the largest +eigenvalue of the strong product: +λmax(G1 ⊠ . . . ⊠ Gk) = +k +� +ℓ=1 +(1 + dℓ) − 1. +(56) +First, equality (56) holds since G1 ⊠ . . . ⊠ Gk is d-regular with a value of d that is equal to the +right-hand side of (56). Straightforward algebra reveals that the required inequality we wish to +assert readily follows from the inequalities +θ(Gℓ) ≥ α(Gℓ) ≥ +nℓ +1 + dℓ +, +ℓ ∈ [k]. +(57) +Indeed, the first inequality in (57) holds since the Lov´asz θ-function of a graph is an upper +bound on its graph capacity (see [65, Theorem 1]), and (by definition) the graph capacity is + +19 +larger than or equal to the independence number of the graph. The second inequality in (57) +holds by Wei’s inequality [92], which provides a lower bound on the independence number or +the clique number of a finite simple graph as a function of the degrees of its vertices (see, e.g., +[5, p. 287] or [8, p. 100] for a nice probabilistic proof of these inequalities; for some further +such bounds, see [40]). For a simple graph G of order n, where vertex i ∈ [n] is of degree di, +Wei’s bound states that +α(G) ≥ +n +� +i=1 +1 +1 + di +, +ω(G) ≥ +n +� +i=1 +1 +n − di +, +(58) +so the first inequality in (58) is specialized to the second inequality in (57) for a dℓ-regular +graph Gℓ of order nℓ. It should be noted that the pair of inequalities in (57) also imply that, +unless not all graphs {Gℓ} are empty, the upper bound on λmin(G1 ⊠. . . ⊠Gk) in the right-hand +side of (54) is smaller than or equal to −1. It is a desired property of this upper bound since +the smallest eigenvalue of a non-empty and finite regular graph is smaller than or equal to −1, +while attaining this value if the graph is complete. +Corollary 7: Let G a d-regular graph of order n. Then, for all k ∈ N, the second-largest and +smallest eigenvalues of the k-fold strong power of G satisfy the inequalities +λ2(G⊠ k) ≥ nk − (1 + d)k +θ(G)k − 1 +− 1, +(59) +and +λmin(G⊠ k) ≤ −(1 + d)k − 1 +� +n +θ(G) +�k +− 1 +. +(60) +Proof: See Section IV-C2. +Example 4: By Corollary 7, the second-largest eigenvalue of the k-fold strong power of the +5-cycle (pentagon) graph C5 satisfies +λ2(C⊠ k +5 +) ≥ 5k − 3k +5 +k +2 − 1 +− 1, +k ∈ N, +(61) +which holds since θ(C5) = +√ +5 (it is also the Shannon capacity of the pentagon [65, Theorem 2]), +and C5 is 2-regular (d = 2). The lower bound on λ2(C⊠ k +5 +) in the right-hand side of (61) scales +asymptotically (for large k) like 5 +k +2 . It is next compared with the Alon–Boppana lower bound. +The k-fold strong power of C5 is a dk-regular graph with dk = (1+d)k −1 = 3k −1. The Alon– +Boppana lower bound in (8) is slightly smaller than 2√dk − 1, which scales asymptotically like +2 · 3 +k +2 . This exemplifies an improvement in the exponential growth rate of the lower bound in +(61), as compared to the Alon–Boppana lower bound. +We next compare numerically the exact values of λ2(C⊠ k +5 +), for 1 ≤ k ≤ 5, with their lower +bounds in the right-hand side of (61) (the exact values were calculated by the SageMath software +[85], and their numerical computation for k > 5 seem to be a difficult task). The exact values +for k ∈ [5] are equal to 0.6180, 3.8541, 13.5623, 42.6869, 130.0608, respectively, (with 4 digit +decimal precision), in comparison to the lower bound in the right-hand side of (61) which is +equal to 0.6180, 3.0000, 8.6264, 21.3333 and 51.4938, respectively. + +20 +The next result shows that every non-complete regular graph G, whose Lov´asz θ-function +is below a fixed value, has the property that almost all the strong powers of G are highly +non-Ramanujan. The derivation of this result relies on inequality (59). +Proposition 3: Let G be a non-empty and non-complete connected d-regular graph on n +vertices. If +θ(G) < +n +√ +d + 1, +(62) +then there exists k0 ∈ N such that, for all k ≥ k0, the k-fold strong power G⊠ k is (highly) +non-Ramanujan. An explicit closed-form expression for the value of k0 is given by +k0 = max + + +3, + + +log +� +2 + (d + 1)− 3 +2 � ++ log +� +n3 +n3−(d+1)3 +� +log +� +n +θ(G) +√ +d+1 +� + + + + + . +(63) +This holds, in particular, for all finite graphs that are self-complementary and vertex-transitive +(they all satisfy the condition in (62) if n > 1), with a value of k0 which, respectively, is equal +to 5, 4 or 3 if n = 5, n = 9 or n ≥ 13 with n ≡ 1 (mod 4). +Proof: See Section IV-C3. +Remark 10: A necessary and sufficient condition for the existence of a self-complementary +graph on n vertices is that n ≡ 0 (mod 4) or n ≡ 1 (mod 4) (see [20, pp. 16–17]). A self- +complementary and d-regular graph of order n satisfies d = n−1 +2 , which implies that n needs to +be odd. Since a vertex-transitive graph is regular, the option of n ≡ 0 (mod 4) is rejected for +graphs of order n that are self-complementary and vertex-transitive. This implies that the order +n of such graphs must satisfy n ≡ 1 (mod 4). For n = 1 and n = 5, there exist graphs of order +n that are self-complementary and vertex-transitive; they are, respectively, given by K1 and C5. +Graphs that are self-complementary and vertex-transitive, and approaches for their construction, +received attention in the literature (see, e.g., [60], [62], [63], [65], [81], [82], [83]). +Remark 11: Proposition 3 was inspired by Example 4, referring to the 5-cycle graph C5 that +is self-complementary and vertex-transitive. It can be numerically verified, with the SageMath +software [85], that C5 and C⊠ 2 +5 +are Ramanujan graphs, and then the higher strong powers C⊠ 3 +5 , +C⊠ 4 +5 , C⊠ 5 +5 +are non-Ramanujan graphs. Proposition 3 shows that all strong powers C⊠ k +5 +, with +k ≥ 5, are non-Ramanujan graphs. An expression for k0, as it is given in (63), was derived in +order to reduce the minimal analytical value of k0 for which the k-fold strong power of C5 is +asserted to be non-Ramanujan for all k ≥ k0. It started with an initial value of k0 = 8 for n = 5, +with a more simple initial expression for k0, and it was reduced to k0 = 5 with the closed-form +expression in (63). Proposition 3, and the numerical experimentation as above, gives that C⊠ k +5 +is a Ramanujan graph if and only if k = 1 or k = 2. +Example 5: Consider the connected Kneser graphs G = K(m, r) with m > 2r and m, r ∈ N. +By their construction and [65, Theorem 13], +n = +�m +r +� +, +d = +�m − r +r +� +, +θ(G) = +�m − 1 +r − 1 +� +. +(64) +The expression of θ(G) in (64) relies on the proof of [65, Theorem 13], which shows that +α(G) = Θ(G) = θ(G) (it holds by combining the Erd¨os-Ko-Rado theorem for finite sets, which + +21 +serves to determine the independence number of the graph, together with the upper bound +on θ(G) in [65, Theorem 9]). Straightforward algebra shows that the condition in (62) is not +necessarily satisfied for the set of parameters in (64). For example, by selecting m = 2r +1, the +condition in (62) is satisfied if and only if 1 ≤ r ≤ 3 (e.g., it is satisfied by the Petersen graph, +which corresponds to r = 2). The condition in (62) is violated for r ≥ 4 so, for these values of +the parameter r, the exponential growth rate of the Alon–Boppana lower bound on λ2(G⊠ k) is +larger than the exponent of the suggested lower bound in the right-hand side of (59). +Remark 12: Several methods for obtaining upper bounds on the smallest eigenvalue of a graph +G rely on the identity λmin(G) = min +x̸=0 +xT A x +xTx +(see, e.g., [27]). The dimensions of the adjacency +matrix of a k-fold strong power of a graph G grow exponentially in k, so the computational +complexity of such bounds is typically very high for strong graph powers. However, for strong +powers, the eigenvalue bounds in Corollary 7 are analytical, and their computational complexity +is not affected by k. The reader is also referred to recent works which derive lower bounds on +the smallest eigenvalue of a graph based on graph decompositions [23], and lower bounds on the +smallest eigenvalue of regular graphs containing many copies of a smaller fixed subgraph [24]. +Such lower bounds on the smallest eigenvalue of a regular graph transform to upper bounds +on the second-largest eigenvalue of regular graphs by using equality (5). Some upper bounds +on the second-largest eigenvalue of connected graphs, with conditions for their attainability, are +provided in [96]. The paper [49] surveys (less recent) bounds on the eigenvalues of simple graphs. +D. Lower bounds on the chromatic numbers of strong products +The problem of relating the chromatic number of a graph to its eigenvalues dates back to +Haemers [42, Section 2.2], Hoffman [48] and Wilf [93], obtaining upper and lower bounds on the +chromatic numbers of graphs in terms of their largest, second-largest and smallest eigenvalues. +This section presents lower bounds on the chromatic number of a strong product of graphs +(and its complement) in terms of the Lov´asz θ-functions (or smallest eigenvalues) of its factors. +Proposition 4: The following lower bounds on chromatic numbers of strong products hold: +(a) Let G1, . . . , Gk be k simple graphs, +��V(Gℓ) +�� = nℓ for ℓ ∈ [k], and G = G1 ⊠ . . . ⊠ Gk. Then, +χ(G) ≥ +� +k +� +ℓ=1 +nℓ +θ(Gℓ) +� +, +(65) +χ(G) ≥ +� +k +� +ℓ=1 +θ(Gℓ) +� +. +(66) +(b) Let G1, . . . , Gk be regular graphs, where Gℓ is dℓ-regular of order nℓ for all ℓ ∈ [k]. Then, +χ(G) ≥ +� +k +� +ℓ=1 +nℓ +θ(Gℓ) +� +(67) +≥ +� +k +� +ℓ=1 +� +1 − +dℓ +λmin(Gℓ) +�� +, +(68) +and inequality (68) holds with equality if each regular graph Gℓ is either edge-transitive or +strongly regular. + +22 +(c) If, for all ℓ ∈ [k], Gℓ is dℓ-regular, and it is either edge-transitive or strongly regular, then +k +� +ℓ=1 +� +1 − +dℓ +λmin(Gℓ) +� +≥ 1 − +d(G) +λmin(G), +(69) +where +d(G) = +k +� +ℓ=1 +(1 + dℓ) − 1 +(70) +is the valency of the regular graph G = G1⊠. . .⊠Gk, and λmin(G) is its smallest eigenvalue. +(d) Let G1, . . . , Gk be regular graphs, where Gℓ is dℓ-regular of order nℓ for all ℓ ∈ [k]. +(1) If, for all ℓ ∈ [k], the graph Gℓ is either vertex-transitive or strongly regular, then the +lower bound on χ(G) in the right-hand side of (65) is larger than or equal to the lower +bound +k� +ℓ=1 +ω(Gℓ). +(2) If, for all ℓ ∈ [k], the graph Gℓ is either (i) both vertex-transitive and edge-transitive, +or (ii) strongly regular, then the lower bound on χ(G) in the right-hand side of (68) is +larger than or equal to the lower bound +k� +ℓ=1 +ω(Gℓ). +(e) Let, for all ℓ ∈ [k], the graph Gℓ be dℓ-regular on nℓ vertices, and suppose that it is either +edge-transitive or strongly regular. Then, +χ(G) ≥ +� +k +� +ℓ=1 +� +− nℓ λmin(Gℓ) +dℓ − λmin(Gℓ) +�� +. +(71) +(f) Let, for all ℓ ∈ [k], Gℓ be a self-complementary graph on nℓ vertices that is either vertex- +transitive or strongly regular. Let n ≜ +k� +ℓ=1 +nℓ be the order of G = G1 ⊠ . . . ⊠ Gn. Then, +χ(G) ≥ +�√n +� +, +(72) +χ(G) ≥ +�√n +� +. +(73) +Proof: See Section IV-D1. +Remark 13: The following inequality is proved in [4, Theorem 11]: +χ(G1 ⊠ G2) ≥ θ(G1) θ(G2). +(74) +(Analogous inequalities to (74) were derived by Hales [45], and by McEliece and Posner [70]; +see [4, Theorem 10]). Inequality (65) can be obtained by combining (17), (18) and (74). This +can be done by first replacing G1 and G2 in (74) with G′ +1 ≜ G1 and G′ +2 ≜ G2 ⊠ . . . ⊠ Gk, +respectively, then relying on (18) to get the inequalities θ(G′ +1) ≥ +n1 +θ(G1) and θ(G′ +2) ≥ n2...nk +θ(G′ +2) , and +finally relying on (17) to get the equality θ(G′ +2) = θ(G2) . . . θ(Gk). Our two simple proofs of +(65) are, however, easier than this one, and they do not require to rely on (74). +Remark 14: Inequality (74) differs from (66), although the lower bounds in the right-hand +sides are similar for k = 2. The difference between (66) and (74) is that the left-hand side of +(66) refers to the chromatic number of G1 ⊠ G2, whereas the left-hand side of (74) refers to + +23 +G1 ⊠ G2. The result in (66) appears in Proposition 4 for completeness since it is an immediate +consequence of the sandwich theorem in (15), and the identity in (17). +Remark 15: Under the assumptions of Item (c) in Proposition 4, the lower bound on χ(G) +in the right-hand side of (68) is larger than or equal to Hoffman’s lower bound (see (69)). +This holds in addition to the high complexity in computing λmin(G) in the right-hand side of +(69), even for relatively small values of k. Hoffman’s bound was originally proved for regular +graphs [48], and it was later extended by Haemers to general simple and finite graphs (see, +[18, Proposition 3.5.3], [42, Theorem 2.1.3], [89, Corollary 8.10 and Theorem 8.11]), and to +hypergraphs [32]. A recent perspective on Hoffman’s bound appears in [44]. +Remark 16: We refer to Item (f) of Proposition 4. If G is self-complementary of order n, then +χ(G) ≥ +�√n +� +. +(75) +Indeed, for every graph G, +χ(G) χ(G)≥ θ(G) θ(G) +(76) +≥ n, +(77) +where inequality (76) holds since, by (15) and (16), χ(G) ≥ θ(G) and χ(G) ≥ θ(G), and +inequality (77) holds by [65, Corollary 2]. The chromatic number is invariant under isomorphism, +so χ(G) = χ(G) if G is self-complementary. This gives (75) from (77). It should be noted, +however, that the result in Item (f) (see (72) and (73)) is not implied by (75). This is because a +strong product of self-complementary graphs, where each factor is also either vertex-transitive or +strongly regular, does not necessarily give a self-complementary graph. As a counter example, +let G1 = G2 = C5. The pentagon C5 is a self-complementary, vertex-transitive and strongly +regular graph, whereas G = C5 ⊠ C5 is not a self-complementary graph by [62, Theorem 1.1]; +according to that theorem, G would have been self-complementary if the strong product of G1 +and G2 had been replaced by their lexicographic product (these last two observations were also +verified by the SageMath software). +The next result specializes Item (a) of Proposition 4 (see (65)) to strong products of strongly +regular graphs. This result is obtained by relying on Corollary 1 that provides a closed-form +expression for the Lov´asz θ-function of each factor. +Corollary 8: Let G1, . . . , Gk be strongly regular graphs with parameters srg(nℓ, dℓ, λℓ, µℓ) for +ℓ ∈ [k] (they need not be distinct). Then, the chromatic number of their strong product satisfies +� +k +� +ℓ=1 +� +1 + +2dℓ +tℓ + µℓ − λℓ +�� +≤ χ(G1 ⊠ . . . ⊠ Gk) ≤ +k +� +ℓ=1 +χ(Gk), +(78) +where {tℓ}k +ℓ=1 in the leftmost term of (78) is given by +tℓ ≜ +� +(λℓ − µℓ)2 + 4(dℓ − µℓ), +ℓ ∈ [k]. +(79) +The leftmost term in (78) is also larger than or equal to the product of the clique numbers of +the factors {Gℓ}k +ℓ=1. +Proof: See Section IV-D2. +The next two examples present strong products of strongly regular graphs, whose chromatic +numbers are exactly determined by Corollary 8. + +24 +Example 6: Let G1, G2, G3 be the Schl¨afli, Shrikhande, and Hall-Janko graphs, respectively. +These are strongly regular graphs whose parameters are srg(27, 16, 10, 8), srg(16, 6, 2, 2), and +srg(100, 36, 14, 12), respectively. Their chromatic numbers are equal to χ(G1) = 9, χ(G2) = 4, +and χ(G3) = 10. Consider the chromatic number of the strong product of arbitrary nonnegative +powers of G1, G2 and G3. It can be verified that, for all such strong products, the upper and +lower bounds in Corollary 8 coincide, so for all integers k1, k2, k3 ≥ 0, +χ(G⊠ k1 +1 +⊠ G⊠ k2 +2 +⊠ G⊠ k3 +3 +) = 9k1 4k2 10k3. +(80) +For comparison, the lower bound that is given by the product of the clique numbers of each +factor is equal to 6k13k24k3 (since ω(G1) = 6, ω(G2) = 3, and ω(G3) = 4). This shows that it +is significantly looser than the tight lower bound in the right-hand side of (80). +Example 7: Consider the three non-isomorphic Chang graphs. These are strongly regular +graphs with the same set of parameters srg(28, 12, 6, 4). The clique number of one of these +graphs is equal to 5, and the clique numbers of the other two graphs are equal to 6. Let us +denote these graphs by G1, G2 and G3, such that ω(G1) = 5, ω(G2) = 6, and ω(G3) = 6. +The chromatic numbers of all these three graphs are similar, and they are equal to 7, i.e., +χ(G1) = χ(G2) = χ(G3) = 7 (the clique and chromatic numbers of the three Chang graphs +are easy to verify with the SageMath software [85]). Let k1, k2 and k3 be arbitrary nonnegative +integers. By Corollary 8, +χ(G⊠ k1 +1 +⊠ G⊠ k2 +2 +⊠ G⊠ k3 +3 +) = 7k1+k2+k3 +(81) +due to the coincidence of the upper and lower bounds in (78). For comparison, the lower bound +on the chromatic number in the left-hand side of (81), which is given by the product of the +clique numbers of each factor, is equal to 5k16k2+k3. +The next example presents a power product of a vertex and edge-transitive regular graph, +whose chromatic number is exactly determined by Item (b) of Proposition 4. +Example 8: The present example provides the exact value of χ(G⊠ k) where G is the Perkel +graph, and k ∈ N. The Perkel graph is 6-regular on 57 vertices, and it is both vertex-transitive and +edge-transitive. The clique and chromatic numbers of G are equal to ω(G) = 2 and χ(G) = 3, +respectively, and the smallest eigenvalue of (the adjacency matrix of) G is equal to λmin(G) = −3. +By Item (b) of Proposition 4, for all k ∈ N, +χ(G⊠ k) ≥ +� +1 − +d +λmin(G) +�k += 3k, +(82) +which can be compared here to the simple upper and lower bounds on χ(G⊠ k) whose values +are given by χ(G)k = 3k and ω(G)k = 2k, respectively. The coincidence of the improved lower +bound in (82) and the upper bound gives that, for all k ∈ N, +χ(G⊠ k) = 3k. +(83) +The next two examples illustrate numerically Part 2 of Item (d) in Proposition 4. +Example 9: Let G be the Suzuki graph, which is a strongly regular graph with parameters +srg(1782, 416, 100, 96) (see [19, Section 10.83]). The lower bound in Corollary 8 (see the +leftmost term in (78)) gives that χ(G⊠ k) ≥ 27k for all k ∈ N. For comparison, since ω(G) = 6, +the lower bound that is based on the clique number of G gives χ(G⊠ k) ≥ 6k for all k ∈ N. The +exact value of χ(G) is not available for the Suzuki graph, so the upper bound in the rightmost + +25 +term of (78) is unknown. The improvement in the exponential lower bound on the chromatic +number of the strong powers G⊠ k is, however, significant since it is increased from 6k to 27k. +Example 10: Let G be the Gosset graph. We apply here Item (b) of Proposition 4 in order +to obtain an improved lower bound on χ(G⊠ k) for all k ∈ N. The graph G is 27-regular on +56 vertices (i.e., d = 27 and n = 56); it is both vertex-transitive and edge-transitive, and it is +also not strongly regular. The clique and chromatic numbers of G are equal to ω(G) = 7 and +χ(G) = 14, respectively, and the smallest eigenvalue of (the adjacency matrix of) G is equal to +λmin(G) = −3. By Item (b) of Proposition 4 (note that the edge-transitivity of G implies that +(68) holds with equality), it follows that for all k ∈ N, +χ(G⊠ k) ≥ +� +1 − +d +λmin(G) +�k +(84) += 10k, +(85) +which can be compared here to the simple upper and lower bounds on χ(G⊠ k) whose values +are given by χ(G)k = 14k and ω(G)k = 7k, respectively. +All the factors of the strong products in Examples 6–10 are either strongly regular or otherwise, +they are both vertex-transitive and edge-transitive graphs. Consequently, by Item (b) of Propo- +sition 4, the two lower bounds in the right-hand sides of (67) and (68) coincide. Furthermore, +by Part 2 of Item (d) in Proposition 4, they also offer an improvement over the simple lower +bound that is equal to the product of the clique numbers of each factor of the strong product. +The next example shows that such an improvement does not necessarily take place if the factors +of the strong product of the regular graphs are not vertex-transitive and edge-transitive, while +also not being strongly regular. +Example 11: Let G be the Frucht graph, which is 3-regular on 12 vertices (i.e., d = 3 and +n = 12). This graph is not strongly regular, and also not vertex-transitive or edge-transitive. The +clique and chromatic numbers of this graph are both equal to 3 (i.e., ω(G) = χ(G) = 3), so +χ(G⊠ k) = 3k, +k ∈ N. +(86) +The smallest eigenvalue of the adjacency matrix of G is equal to λmin(G) = −2.33866. By (68), +for all k ∈ N, +χ(G⊠ k) ≥ +� +1 − +d +λmin(G) +�k +(87) += 2.28278k. +(88) +This shows that the lower bound on the chromatic number in the right-hand side of (68) may be +looser than the simple lower bound that is equal to the product of the clique numbers of each +factor. This may happen if some of the factors, which are regular graphs, are not either strongly +regular, or both vertex-transitive and edge-transitive. +E. The Shannon capacity of strongly regular graphs +The Lov´asz θ-function of a strongly regular graph is expressed in closed-form in Corollary 1, +being also an upper bound on the Shannon capacity of such a graph [65, Theorem 1]. The +independence number of a graph is, on the other hand, a lower bound on the Shannon capacity. + +26 +The Shannon capacity of such a graph is therefore determined if these upper and lower bounds +coincide. The following examples show such cases of coincidence for strongly regular graphs. +The first one reproduces (in a different way) the Shannon capacity of the Petersen graph, a result +dating back to Lov´asz [65]. The rest of the examples provide new results that determine the +Shannon capacity of some strongly regular graphs. +Example 12: Let G be the Petersen graph, whose Shannon capacity is equal to 4. It is a +special case of [65, Theorem 13], which gives the Shannon capacity of Kneser graphs +Θ +� +K(m, r) +� += +�m − 1 +r − 1 +� +, +m ≥ 2r. +(89) +Then, the capacity of the Petersen graph is obtained in [65, Corollary 6] by viewing it as an +isomorphic graph to the Kneser graph K(5, 2). +As an alternative way to determine its Shannon capacity, the Petersen graph is a strongly +regular graph with parameters srg(10, 3, 0, 1) [19, Section 10.3]. By Corollary 1, it follows that +θ(G) = 4. This coincides with the independence number α(G) = 4, so Θ(G) = 4. +Example 13: Let G be the Shrikhande graph, which is strongly regular with the parameters +srg(16, 6, 2, 2) [19, Section 10.6]. By Corollary 1, it can be verified that θ(G) = 4. Its chromatic +number is χ(G) = 4, and its independence number is α(G) = 4 (so α(G) χ(G) = n, which +implies that the n = 16 vertices in G are partitioned into four color classes, where each color +class is a largest independent set in G of size 4). Hence, the graph capacity is equal to Θ(G) = 4. +The capacity of every graph G of order n that is self-complementary and vertex-transitive is +Θ(G) = √n (see [65, Theorem 12]). It should be noted, however, that although the capacity of +the Shrikhande graph is a case where Θ(G) = √n (with n = 16), it does not follow from [65, +Theorem 12] since the Shrikhande graph is vertex-transitive, but it is not self-complementary. +Example 14: Consider the Hall-Janko graph G, which is strongly regular with parameters +srg(100, 36, 14, 12) [19, Section 10.32]. By Corollary 1, θ(G) = 10. The independence number +of this graph is α(G) = 10 (this was obtained by the SageMath software [85]). Hence, the +Shannon capacity of the Hall-Janko graph is Θ(G) = 10. This graph is vertex-transitive but it is +not self-complementary so, similarly to Example 13, this result does not follow from the capacity +result in [65, Theorem 12] for finite graphs that are vertex-transitive and self-complementary. +Example 15: The Hoffman-Singleton graph G is a strongly regular graph with parameters +srg(50, 7, 0, 1) [19, Section 10.19]. By Corollary 1, it follows that θ(G) = 15. The independence +number of this graph is α(G) = 15, so the Shannon capacity of this graph is Θ(G) = 15. +Example 16: The Schl¨afli graph G is strongly regular with parameters srg(27, 16, 10, 8) [19, +Section 10.10]. By Corollary 1, it follows that θ(G) = 3. The independence number of this +graph is α(G) = 3, so the Shannon capacity of this graph is Θ(G) = 3. +Example 17: The Sims-Gewirtz graph G is strongly regular with parameters srg(56, 10, 0, 2) +(a.k.a. the Gewirtz graph) [19, Section 10.20]. By Corollary 1, θ(G) = 16, and also α(G) = 16 +(it can be verified by the SageMath software [85]). The Shannon capacity of the Sims-Gewirtz +graph is therefore equal to Θ(G) = 16. +Example 18: The M22 graph (a.k.a. Mesner graph) G is strongly regular with parameters +srg(77, 16, 0, 4) [19, Section 10.27]. Its independence number is α(G) = 21, and by Corollary 1 + +27 +also θ(G) = 21. Consequently, the Shannon capacity of the M22 graph is Θ(G) = 21. +Example 19: The Cameron graph G is strongly regular with parameters srg(231, 30, 9, 3) (see +[19, Section 10.54]). By Corollary 1, θ(G) = 21, which coincides with the independence number +of G, i.e., α(G) = 21. The Shannon capacity of the Cameron graph is, hence, equal to Θ(G) = 21. +Example 20: Consider the three non-isomorphic Chang graphs. These are strongly regular +graphs with the same set of parameters srg(28, 12, 6, 4) [19, Section 10.11]. By Corollary 1, +the Lov´asz θ-function of these graphs is equal to 4, which coincides with their independence +number. The Shannon capacity of each of these three graphs is therefore equal to 4. +Remark 17: Examples 12–20 provide several strongly regular graphs for which their Lov´asz +θ-function coincides with their independence number, thus determining the Shannon capacity of +these graphs. This, however, is not the case in general for strongly regular graphs. For example, +in continuation to Example 16, the Lov´asz θ-function of the complement of the Schl¨afli graph G +(it is a strongly regular graph srg(27, 10, 1, 5)) is equal to θ(G) = 9. Indeed, θ(G) θ(G) = 27 is +the order of G. It was proved, however, by the rank-bound of Haemers [43] that the capacity of +the complement of the Schl¨afli graph is at most 7. It can be also verified that the independence +number of G (i.e., the clique number of the Schl¨afli graph G) is equal to 6. This overall gives +that 6 = ω(G) ≤ Θ(G) ≤ 7 < 9 = θ(G). +We end this section by studying some capacity results of affine polar graphs. +Example 21: Affine polar graphs (see, e.g., [19, Section 3.3]) are constructed by considering +a vector space V of dimension d over a finite field Fq with q elements, equipped with a non- +degenerate quadratic form Q. The vertices in the graph are represented by the vectors in V, and +any two vectors u and v (i.e., a pair of vertices in the graph) are adjacent if Q(u − v) = 0. +The resulting graph is denoted by VO+(d, q), VO−(d, q), and VO(d, q) when the quadratic form +Q is hyperbolic, elliptic or parabolic, respectively. In the first two cases, d is even, and in the +third case, d is odd; in all cases, q is a positive integral power of a prime number (as it is the +cardinality of the finite field Fq). Under these conditions on d and q, the graphs VO+(d, q) and +VO−(d, q) are strongly regular. Let d = 2e with e ∈ N. The parameters of G+ = VO+(2e, q), +which is a strongly regular graph srg(n+, d+, λ+, µ+), are given by (see [19, Section 3.3]) +n+ = q2e, +(90) +d+ = (qe−1 + 1)(qe − 1), +(91) +λ+ = q(qe−2 + 1)(qe−1 − 1) + q − 2, +(92) +µ+ = qe−1(qe−1 + 1), +(93) +λ2(G+) = qe − qe−1 − 1, +(94) +λn+(G+) = −qe−1 − 1, +(95) +and G− = VO−(2e, q), a strongly regular graph srg(n−, d−, λ−, µ−), has the parameters +n− = q2e, +(96) +d− = (qe−1 − 1)(qe + 1), +(97) +λ− = q(qe−2 − 1)(qe−1 + 1) + q − 2, +(98) +µ− = qe−1(qe−1 − 1), +(99) +λ2(G−) = qe−1 − 1, +(100) +λn−(G−) = −qe + qe−1 − 1. +(101) + +28 +By Corollary 1, combined with the parameters (and eigenvalues) in (90)–(101), it follows that +θ(G+) = qe = +√ +n+, +(102) +θ(G+) = qe, +(103) +θ(G−) = q(qe − qe−1 + 1), +(104) +θ(G−) = +q2e−1 +qe − qe−1 + 1. +(105) +Consequently, it can be verified that for some of these affine polar graphs (with the free +parameters q and e as above), their Lov´asz θ-function and independence number coincide. (The +numerical computations of the independence numbers of these graphs were performed by the +SageMath software [85].) This gives, for example, that +Θ +� +VO+(4, 2) +� += 4, +Θ +� +VO+(6, 2) +� += 8, +(106) +Θ� +VO+(4, 3)� += 9, +Θ� +VO+(6, 3)� += 27, +(107) +as the exact values of the Shannon capacities of these strongly regular graphs, whose parameters +are, respectively, srg(16, 9, 4, 6), srg(64, 35, 18, 20), srg(81, 32, 13, 12), and srg(729, 260, 97, 90). +IV. +PROOFS +A. Proofs for Section III-A +1) Proof of Proposition 1: Let G be a d-regular graph with n vertices. The rightmost inequality +in (24) is provided in [65, Theorem 9], together with a sufficient condition that it holds with +equality if G is edge-transitive. Another sufficient condition for that inequality to hold with +equality is obtained later in this proof. +We first prove the leftmost inequality in (24), and also obtain sufficient conditions that it holds +with equality. By (18) (see [65, Corollary 2]), +θ(G) ≥ +n +θ(G) +(108) +and, by [65, Theorem 8], equality holds in (108) if G is vertex-transitive (or, equivalently, if G +is vertex-transitive). Since G is an (n − d − 1)-regular graph of order n, an application of (19) +(see [65, Theorem 9]) to G gives +θ(G) ≤ − +nλn(G) +(n − d − 1) − λn(G), +(109) +with equality in (109) if G is edge-transitive. Combining (108) and (109) gives +θ(G) ≥ 1 − n − d − 1 +λn(G) +, +(110) +with equality in (110) if G is vertex-transitive and edge-transitive (recall that a vertex-transitive +graph is regular). By the regularity of G, we have from (5), +λn(G) = −1 − λ2(G). +(111) +Combining (110) and (111) gives the leftmost inequality in (24), together with the conclusion +that it holds with equality if G is both vertex-transitive and edge-transitive. Next, (25) is obtained +from (24) by replacing G with G, and relying on the equality in (111). A sufficient condition + +29 +that the leftmost inequality in (25) holds with equality is therefore the same condition that +the leftmost inequality in (24) holds with equality, while replacing G with G; this means that +the leftmost inequality in (25) holds with equality if G is vertex-transitive and edge-transitive. +Likewise, the rightmost inequality in (25) holds with equality under the same condition that the +rightmost inequality in (24) holds with equality, while the graph G in (24) is replaced by its +complement G in (25). This means that the rightmost inequality in (24) holds with equality if +G is edge-transitive (recall that, unlike vertex-transitivity that is a property of G if and only if it +is a property of its complement G, it is not the case for edge-transitivity). +We finally prove that if G is a strongly regular graph, then the four inequalities in (24) and +(25) hold with equalities. Let G be a strongly regular graph srg(n, d, λ, µ). Then, the largest, +second-largest, and smallest eigenvalues of G are given by +λ1(G) = d, +(112) +λ2(G) = r ≜ 1 +2 +� +λ − µ + +� +(λ − µ)2 + 4(d − µ) +� +, +(113) +λn(G) = s ≜ 1 +2 +� +λ − µ − +� +(λ − µ)2 + 4(d − µ) +� +, +(114) +where p1,2 in (10) are replaced here with r and s in (113) and (114), respectively. (Recall that if +G is a disconnected strongly regular graph, then µ = 0 and λ = d−1 by (9), which then indeed +gives that λ2(G) = d = λ1(G) since the d-regular graph is disconnected). The substitution of +(113) and (114) into (24) and (25) gives +n − d + r +1 + r +≤ θ(G) ≤ − ns +d − s, +(115) +and +1 − d +s ≤ θ(G) ≤ n(1 + r) +n − d + r. +(116) +We show as follows that the rightmost and leftmost terms in (115) coincide, turning the two +inequalities in (115) into equalities. The coincidence of the rightmost and leftmost terms in (115) +is equivalent to proving the equality +(n − d + r) (d − s) + ns (1 + r) = 0. +(117) +Indeed, we get +(n − d + r) (d − s) + ns (1 + r) += nd − d2 + d (r + s) + (n − 1)rs +(118) += nd − d2 + d (λ − µ) + (n − 1) (µ − d) +(119) += −d (d − λ − 1) + µ (n − d − 1) +(120) += 0, +(121) +where (118) and (120) hold by straightforward algebra; (119) holds by (113) and (114); (121) +holds by the identity in (9). +The same conclusion also holds for the rightmost and leftmost terms in (116), which can be +shown to coincide if G is a strongly regular graph. This turns the two inequalities in (116) into +equalities. It holds in a similar way to the previous part since the complement of a strongly +regular graph is also strongly regular, and the second-largest and smallest eigenvalues of G in + +30 +(115) (i.e., r, s respectively) are replaced in (115) by the second-largest and smallest eigenvalues +of G (i.e., −1 − s and −1 − r respectively). This proves that each of the four inequalities in +(24) and (25) holds with equality if G is a strongly regular graph. +2) Proof of Corollary 1: Let G be a strongly regular graph srg(n, d, λ, µ). By Proposition 1, +θ(G) = n − d + r +1 + r +(122) += − ns +d − s, +(123) +and +θ(G) = 1 − d +s +(124) += n(1 + r) +n − d + r, +(125) +where r and s are given in (113) and (114), respectively. Their substitution into the right-hand +sides of (123) and (124) gives (26) and (28), respectively, with the auxiliary parameter t as +defined in (29). Equality (27) is trivial by (122)–(125). This verifies (26)–(28) for all strongly +regular graphs. +Since the multiplicities of the distinct eigenvalues of a strongly regular graph G are integers, +it follows from (11) that if 2d + (n − 1)(λ − µ) ̸= 0, then t in (29) should be an integer. This +implies that, under the latter condition, θ(G) and θ(G) are necessarily rational numbers. +B. Proofs for Section III-B +1) Proof of Corollary 2: Let G be a non-complete, d-regular graph of order n. Straightforward +algebra gives (30) by comparing the rightmost and leftmost terms in (24). Inequalities (30) +and (31) can be verified to be equivalent by relying on the inequality d+(n−1)λn(G) < 0 (this +holds since λn(G) ≤ −1, and d < n−1 for a non-complete, d-regular graph G with n vertices). +Equality holds in (30) if and only if the rightmost and leftmost terms in (24) are equal to θ(G) +(i.e., they are both equal to the middle term in (24)). According to Item (a) of Proposition 1, +equality holds in (30) and (31) if G is strongly regular. +We next show that the latter sufficient condition is also necessary. To that end, we provide +a second proof that relies on linear algebra. Let A = A(G) and A = A(G) be the adjacency +matrices of G and G, respectively. Let Jn and In denote the n-times-n all-ones and identity +matrices, respectively. Then, by (2), A = Jn − In − A. Let jn be the all-ones n-length column +vector, so jn jT +n = Jn, and +A + c1 In +c2 Jn = (c1 − 1) In +(c2 + 1) Jn − A, +(126) +for all c1, c2 ∈ R. Since G is a d-regular graph, the largest eigenvalue of A is equal to d with +jn as an eigenvector. The vector jn is also an eigenvector of Jn and In with eigenvalues n and +1, respectively. By (126), and since G is d-regular, +(A + c1 In +c2 Jn) jn = +� +c1 − 1 + n(c2 + 1) − d +� +jn, +(127) +which means that jn is an eigenvector of the matrix A+c1 In +c2 Jn. The other n−1 eigenvectors +of the symmetric matrix A + c1 In +c2 Jn can be made orthogonal to the eigenvector jn (since +eigenvectors of a symmetric matrix, which correspond to distinct eigenvalues, are orthogonal + +31 +with respect to the standard inner product in Rn; furthermore, eigenvectors that correspond to +the same eigenvalue can be made orthogonal by the Gram-Schmidt procedure). Let v be such +an eigenvector, different from jn. Then Jn v = jn(jT +nv) = 0, so by (126) +(A + c1 In +c2 Jn) v = (c1 − 1)v − A v, +(128) +which means that v is also an eigenvector of the adjacency matrix A (since, by assumption, v +is an eigenvector of the matrix A + c1 In +c2 Jn in the left-hand side of (128)). The equality +A v = λv holds, and λ ∈ {λ2(G), . . . , λn(G)} with λ2(G) being the largest among them. This +gives that λ ≤ λ2(G). The symmetric matrix A + c1 In +c2 Jn is positive semi-definite if and +only if all its eigenvalues are nonnegative, i.e. (see (127) and (128)), +c1 − 1 + n(c2 + 1) − d ≥ 0, +(129) +c1 − 1 − λ2(G) ≥ 0. +(130) +Setting the two conditions in (129) and (130) to be satisfied with equalities gives +c1 = 1 + λ2(G), +c2 = −n − d + λ2(G) +n +, +(131) +which implies that (see the left-hand side of (126)) +A + +� +1 + λ2(G) +� +In − +�n − d + λ2(G) +n +� +Jn ⪰ 0 +(132) +is a positive semi-definite matrix. Clearly, also +A −λn(G) In ⪰ 0 +(133) +is positive semi-definite. The trace of a product of two n-times-n positive semi-definite matrices +is nonnegative, so it follows from (132) and (133) that +tr +�� +A + +� +1 + λ2(G) +� +In − +�n − d + λ2(G) +n +� +Jn +� � +A −λn(G) In +� +� +≥ 0. +(134) +For an adjacency matrix A of a d-regular graph G, +tr(A A) = tr(A) = tr(A) = 0, +(135) +tr(Jn A) = nd, +(136) +and +tr(In) = tr(Jn) = n. +(137) +From (134)–(137), expanding the left-hand side in (134) gives the inequality +−d +� +n − d + λ2(G) +� +− λn(G) +� +d + (n − 1)λ2(G) +� +≥ 0, +(138) +which proves (30) (we have d + (n − 1)λ2(G) > 0, by the assumption that the d-regular graph +G is non-complete and non-empty, so λ2(G) ≥ 0). This alternative proof is next used to find the +necessary and sufficient condition for the satisfiability of (30) with equality. An equality in (30) +holds if and only if (134) holds with equality (i.e., the trace of the product of the two positive +semi-definite matrices in the left-hand sides of (132) and (133) is equal to zero). This holds if +and only if the column spaces of the two matrices in the left-hand sides of (132) and (133) are + +32 +orthogonal. [Clarification: if B, C ⪰ 0 are positive semi-definite matrices, then B = SST and +C = QQT for some matrices S and Q. Hence, under the assumption that tr(BC) = 0, +0 = tr(BC) = tr(SSTQQT) = tr(STQQTS) = tr(STQ (STQ)T) = +n +� +i=1 +n +� +j=1 +� +STQ +�2 +i,j (139) +implies that STQ = 0, so also BC = S(STQ)QT = 0. Since B and C are symmetric matrices +with BC = 0, it means that the column spaces of B and C are orthogonal]. These column +spaces, however, can be orthogonal only if A has no eigenvalues other than d, λ2(G) and +λn(G); otherwise, there is a joint eigenvector of A in the two column spaces of the matrices in +the left-hand sides of (132) and (133), so these two column spaces cannot be orthogonal in the +latter case. [Clarification: suppose that A has an eigenvector v that corresponds to an eigenvalue +λ∗ ̸∈ {d, λ2(G), λn(G)}. Then, by (128) and the value of c1 in (131), the right-hand side of (128) +is equal to (λ2(G) − λ∗)v, and (A −λn(G) In)v = (λ∗ − λn(G))v. Both coefficients of v in +these two expressions are nonzero, so v belongs to the two column spaces of the two matrices in +the left-hand sides of (132) and (133). These column spaces are therefore not orthogonal under +the assumption of the existence of a distinct eigenvalue λ∗ of A, as above]. We next distinguish +between two cases in regard to the connectivity of the graph G. +(a) If the d-regular graph G is connected, then λ1(G) = d > λ2(G). By assumption, G is also +non-complete and non-empty graph, so λ2(G) > λn(G). The connected regular graph G thus +has exactly three distinct eigenvalues, so it is strongly regular. +(b) If the d-regular graph G is disconnected, then λ1(G) = d = λ2(G). If, by assumption, +inequality (30) holds with equality, then λn(G) = −1. This means that G is a disjoint union +of equal-sized complete graphs Kd+1, so it is an imprimitive strongly regular graph (i.e., +there are no common neighbors of any pair of non-adjacent vertices in G). +We therefore conclude that, for a non-complete and non-empty d-regular graph on n vertices, +the condition that G is strongly regular is also necessary (and not only sufficient, as it is shown +in the first proof) for the inequality in (30) to hold with equality. Equivalently, G being strongly +regular is a necessary condition for the inequality in (31) to hold with equality. This completes +the proof of the necessity and sufficiency of the condition on the strong regularity of G. +2) Proof of Corollary 3: Let G be a d-regular graph of order n, which is non-complete. +Proof of Item (a): By the definition of g2(G) and gn(G) in (32), and in light of the equalities +λn(G) = −1−λ2(G) and λ2(G) = −1−λn(G), it can be readily verified that the rightmost and +leftmost inequalities in (33) are equivalent to (30) and (31), respectively. Furthermore, in light +of Corollary 2, each inequality in (33) holds with equality if and only if G is strongly regular. +Proof of Item (b): Let G be a strongly regular graph. We next prove the claim about the +dichotomy in the number of distinct values in the sequence {gℓ(G)}n +ℓ=1. +(1) For a d-regular graph on n vertices, λ1(G) = d, and λ1(G) = n − d − 1. Substituting these +eigenvalues into (32) gives that g1(G) = n − d − 2 ≥ 0 (G is non-complete, so d ≤ n − 2). +(2) The graph G and its complement G are both strongly regular, so each one of them has at +most three distinct eigenvalues. +(3) For a strongly regular graph, by Item (a), g2(G) = −1 = gn(G), and gℓ(G) = −1 if either +(i) λℓ(G) = λ2(G) and λℓ(G) = λ2(G), or (ii) λℓ(G) = λn(G) and λℓ(G) = λn(G). + +33 +(4) By assumption, G is a strongly regular graph, which implies that so is G. Due to their +regularity, λ1(G) ≥ λ2(G), and λ1(G) ≥ λ2(G) with equalities, respectively, if and only if +G or G are disconnected graphs. The sequence {gℓ(G)}n +ℓ=1 gets an additional (third) distinct +value if and only if the multiplicities of the smallest and the second-largest eigenvalues +of G in the subsequence (λ2(G), . . . , λn(G)) are distinct. Indeed, in the latter case, only +one of the following two options is possible: (iii) λℓ(G) = λ2(G) and λℓ(G) = λn(G), or +(iv) λℓ(G) = λn(G) and λℓ(G) = λ2(G). This holds since, by (4), the multiplicity of the +second-largest eigenvalue of G is equal to the multiplicity of the smallest eigenvalue of G, +and similarly, the multiplicity of the smallest eigenvalue of G is equal to the multiplicity of +the second-largest eigenvalue of G. It therefore follows that the third distinct value (as above) +is attained by the sequence {gℓ}n +ℓ=1 a number of times that is equal to the absolute value of +the difference between the multiplicities of the second-largest and the smallest eigenvalues +of G in the subsequence (λ2(G), . . . , λn(G)) (provided that the latter two multiplicities are +distinct). +Proof of Item (c): Let G be self-complementary and d-regular on n vertices. Then, +d = 1 +2(n − 1), +λ2(G) = λ2(G), +λn(G) = λn(G). +(140) +Combining the rightmost inequality in (33) and the equalities in (140) readily gives +2λ2(G) 2 − 1 +2(n + 1) +1 + 2λ2(G) +≥ −1, +(141) +Since G is non-complete and non-empty, we get λ2(G) > 0, which then gives from (141) the +quadratic inequality +2λ2(G) 2 + 2λ2(G) − 1 +2(n − 1) ≥ 0. +(142) +Its solution gives (34) (n > 1 as otherwise, G = K1, but G is by assumption non-complete). +Hence, it also follows that +λn(G) = λn(G) +(143) += −1 − λ2(G) +(144) +≤ −1 − 1 +2(√n − 1) +(145) += − 1 +2(√n + 1), +(146) +where (143) holds since G is (by assumption) self-complementary; (144) holds since G is regular, +and (145) holds by (34). +Proof of Item (d): If G is self-complementary and strongly regular, then in light of Items (b) +and (c) here, both inequalities in (34) and (35) hold with equality. +3) Proof of Corollary 4: Let {Gℓ}ℓ∈N be a sequence of regular graphs where Gℓ is dℓ-regular +of order nℓ, such that nℓ → ∞ and dℓ +nℓ → 0 as we let ℓ tend to infinity. Then, +lim sup +ℓ→∞ +ω(Gℓ) ≤ lim sup +ℓ→∞ +θ(Gℓ) +(147) +≤ lim sup +ℓ→∞ +nℓ +� +1 + λ2(Gℓ) +� +nℓ − dℓ + λ2(Gℓ) +(148) += 1 + lim sup +ℓ→∞ +λ2(Gℓ), +(149) + +34 +where (147) holds by the leftmost inequality in (16); (148) holds by the rightmost inequality in +(25); (149) holds by the assumption that nℓ → ∞, and since the eigenvalues of Gℓ are bounded +(in absolute value) by dℓ with lim +ℓ→∞ +dℓ +nℓ = 0. This leads to inequality (40), by a floor operation in +the right-hand side of (149), since clique numbers are integers. +We next prove inequality (41). For any graph G with n vertices, +α(G) χ(G) ≥ n. +(150) +(This well-known inequality holds since the independence number α(G) denotes the size of a +largest independent set in G, and in coloring the vertices in G with χ(G) colors, all color classes +are independent). Additionally, ω(G) = α(G), so +ω(G) χ(G) ≥ n. +(151) +This gives +lim inf +ℓ→∞ +χ(Gℓ) +nℓ +≥ lim inf +ℓ→∞ +1 +ω(Gℓ) +(152) += +1 +lim sup +ℓ→∞ +ω(Gℓ) +(153) +≥ +1 +1 + lim sup +ℓ→∞ +⌊λ2(Gℓ)⌋ , +(154) +where (152) holds by (151); (153) is trivial, and (154) holds by (40). +4) Proof of Corollary 5: Inequalities (43) and (45) readily follow from Corollary 4 since if +{Gℓ}ℓ∈N is a sequence of connected Ramanujan d-regular graphs (d is a fixed degree of the +vertices), then (by definition) +λ2(Gℓ) ≤ 2 +√ +d − 1, +(155) +for all ℓ ∈ N. By the assumption that the graph Gℓ has order nℓ with lim +ℓ→∞nℓ = ∞, inequalities +(43) and (45) are obtained by combining, respectively, (40) and (41) with (155). +Inequality (44) is obtained by combining the leftmost inequality in (24) with (155). Indeed, +since |λ2(Gℓ)| ≤ d (where the degree of the vertices of Gℓ is, by assumption, equal to a fixed +value d), it follows that +θ(Gℓ) ≥ +nℓ − 2d +1 + λ2(Gℓ) ≥ +nℓ − 2d +1 + 2 +√ +d − 1, +(156) +which then yields (44). +5) Proof of Corollary 6: Let G be a connected Ramanujan d-regular graph with n vertices. If +G = Kn is the complete graph, which is a Ramanujan (n−1)-regular graph, then inequality (46) +clearly holds (with ω(G) = n). Otherwise, if G is non-complete, then combining the leftmost +inequality in (16) and the rightmost inequality in (25) gives that +ω(G) ≤ +� n +� +1 + λ2(G) +� +n − d + λ2(G) +� +, +(157) +where the floor operation in the right-hand side of (157) is enabled because ω(G) is an integer. +Since 0 ≤ λ2(G) ≤ 2 +√ +d − 1 for a non-complete and d-regular connected Ramanujan graph G, +and since the function f1 : (−1, ∞) → (0, ∞) that is given by +f1(x) ≜ n(1 + x) +n − d + x, +x > −1, +(158) + +35 +is monotonically increasing, inequality (46) then follows from (157) and the monotonicity of the +function f1. Eq. (48) follows from (46), (151), and since the chromatic number is an integer. +Inequality (47) holds with equality if G = Kn (both sides are equal to 1). Otherwise, +inequality (47) holds by the leftmost inequality in (24), since 0 ≤ λ2(G) ≤ 2 +√ +d − 1 for a non- +complete and d-regular Ramanujan graph G, and also since the function f2 : (−1, ∞) → (0, ∞) +that is given by +f2(x) ≜ n − d + x +1 + x +, +x > −1, +(159) +is monotonically decreasing. +C. Proofs for Section III-C +1) Proof of Proposition 2: Let G1, . . . , Gk be regular graphs such that, for all ℓ ∈ [k], the +graph Gℓ is dℓ-regular with nℓ vertices. Let G = G1 ⊠ . . . ⊠ Gk be the strong product of these +k regular graphs. We next prove the two items of Proposition 2. +(a) By the leftmost inequality in (24), unless G = Kn, +θ(G) ≥ n(G) − d(G) + λ2(G) +1 + λ2(G) +, +(160) +where n(G) and d(G) denote, respectively, the order and valency of the strong product, which +is a regular graph (since, by assumption, each factor is regular). The following equalities +hold as a result of the strong product operation: +n(G) = +k +� +ℓ=1 +nℓ, +(161) +d(G) = +k +� +ℓ=1 +(1 + dℓ) − 1, +(162) +θ(G) = +k +� +ℓ=1 +θ(Gℓ). +(163) +Indeed, equality (161) holds since the cardinality of a Cartesian product of finite sets is +equal to the product of the cardinalities of each set; equality (162) can be justified by first +verifying the special case of a strong product of two regular graphs, and then proceeding by +a mathematical induction on k. Finally, equality (163) holds by (17) (see [65, Theorem 7]). +Combining the bound in (160) with equalities (161)–(163) gives +k +� +ℓ=1 +θ(Gℓ) ≥ 1 + +k� +ℓ=1 +nℓ − +k� +ℓ=1 +(1 + dℓ) +1 + λ2(G) +. +(164) +Unless all Gℓ (with ℓ ∈ [k]) are complete graphs, the left-hand side of (164) is strictly larger +than 1, and then rearrangement of the terms in (164) gives the lower bound on λ2(G) in (52). +Next, the possible loosening of the lower bound in the right-hand side of (52) to the lower +bound in the right-hand side of (53) holds by (19) (see [65, Theorem 9]). Inequality (53) +holds with equality if each regular factor Gℓ is either edge-transitive (by [65, Theorem 9]) +or strongly regular (by Item (a) of Proposition 1). + +36 +(b) Combining (19) with equalities (161)–(163) gives, with n = n(G) and d = d(G), +k +� +ℓ=1 +θ(Gℓ) = θ(G) +(165) +≤ − nλn(G) +d − λn(G) +(166) += − +k� +ℓ=1 +nℓ · λn(G) +k� +ℓ=1 +(1 + dℓ) − 1 − λn(G) +. +(167) +Unless all Gℓ (with ℓ ∈ [k]) are empty graphs, the denominator in the right-hand side of +(167) is strictly positive. This gives (54) after rearrangement of terms. Finally, the transition +from (54) to (55) is justified if +θ(Gℓ) = − nℓλmin(Gℓ) +dℓ − λmin(Gℓ), +∀ ℓ ∈ [k]. +(168) +As above (the end of the proof of Item (a)), the condition in (168) holds if the regular graph +Gℓ is either edge-transitive or strongly regular. +2) Proof of Corollary 7: Let G a d-regular graph of order n. The lower bound on the second- +largest eigenvalue λ2(G⊠ k) in the right-hand side of (59) follows from (52) by setting there +G1, . . . , Gk to be all identical to G. The upper bound on the smallest eigenvalue λmin(G⊠ k) in +the right-hand side of (60) follows in a similar way from (54). +3) Proof of Proposition 3: Let G be a non-empty and non-complete connected d-regular graph +on n vertices, and let k ∈ N. Then, G⊠ k is a connected regular graph, which is non-complete +and non-empty (so, its largest eigenvalue is of multiplicity 1). By (59), +λ2(G⊠ k) ≥ nk − (1 + d)k +θ(G)k − 1 +− 1. +(169) +In order to prove that the k-fold strong power G⊠ k is non-Ramanujan, it is sufficient to show that +the lower bound on its second-largest eigenvalue in the right-hand side of (169) is larger than +2√dk − 1; here, dk = (1+d)k −1 is the valency of the considered strong power (composed of k +factors, where each factor is the d-regular graph G). Since G is d-regular and non-complete (i.e., +d < n − 1 and θ(G) > 1), the right-hand side of (169) scales asymptotically like +� +n +θ(G) +�k (for +a sufficiently large k), whereas the expression 2√dk − 1 scales asymptotically like 2(1 + d) +k +2 . +Comparing these two exponents gives that if +θ(G) < +n +√ +1 + d, +(170) +then the exponential growth rate of the right-hand side of (169) is larger than that one of +2√dk − 1. Hence, for sufficiently large k, the strong power G⊠ k is a (highly) non-Ramanujan +graph under the condition in (170). This means that there exists k0 ∈ N such that the strong +power G⊠ k is non-Ramanujan for all k ≥ k0. We next obtain an explicit value of such k0, which +is not necessarily the smallest one, proving that such a valid value for k0 is given by (63). To +that end, based on the above explanation, one needs to deal with the inequality +nk − (1 + d)k +θ(G)k − 1 +− 1 > 2 +� +(1 + d)k − 2. +(171) + +37 +In order to obtain a closed-form solution, we strengthen the condition in (171) to +nk − (1 + d)k +θ(G)k +− 1 ≥ 2(1 + d) +k +2 . +(172) +Dividing both sides of (172) by (1 + d) +k +2 gives +� +n +√ +1 + d θ(G) +�k +− +�√ +1 + d +θ(G) +�k +− (1 + d)− k +2 ≥ 2. +(173) +Let k ≥ 3. The condition imposed in (173) can be further strengthened to +� +n +√ +1 + d θ(G) +�k +− +�√ +1 + d +θ(G) +�k +≥ 2 + (1 + d)− 3 +2 . +(174) +Since d < n − 1 for a non-complete d-regular graph of order n, for all k ≥ 3, +� +n +√ +1 + d θ(G) +�k +− +�√ +1 + d +θ(G) +�k += +� +n +√ +1 + d θ(G) +�k � +1 − +�1 + d +n +�k� +≥ +� +n +√ +1 + d θ(G) +�k � +1 − +�1 + d +n +�3� +> 0, +(175) +which, by combining (174) and (175), gives the stronger condition +n3 − (1 + d)3 +n3 +� +n +√ +1 + d θ(G) +�k +≥ 2 + (1 + d)− 3 +2 , +(176) +with k ≥ 3. Solving inequality (176) implies that inequality (171) is satisfied for all k ≥ k0, +with the closed-form expression of k0 in (63). It therefore gives that if G is a d-regular graph +on n vertices, which satisfies the condition in (170), then G⊠ k is non-Ramanujan for all k ≥ k0 +(it becomes, in fact, a highly non-Ramanujan graph since both sides of inequality (171) have +different exponential growth rates, so the condition for a Ramanujan graph is strongly violated +for the strong power G⊠ k when the value of k is increased). +We next specialize this result for graphs that are self-complementary and vertex-transitive. For +n = 1, the complete graph G = K1 is a self-complementary and vertex-transitive graph, whose +all strong powers are also isomorphic to K1, so they are therefore non-Ramanujan graphs. +Let G be a graph of order n > 1 that is self-complementary and vertex-transitive, so it is +d-regular with d = 1 +2(n − 1). Additionally, for such a graph G, the Lov´asz θ-function is equal +to θ(G) = √n, and it coincides with the Shannon capacity of G (see [65, Theorem 8] and [65, +Theorem 12]). Then, +n +√ +d + 1 = +� +2n2 +n + 1 > √n = θ(G), +(177) +so the required condition in Proposition 3 is fulfilled by graphs of order n that are self- +complementary and vertex-transitive. The value of k0 in (63) is specialized for such graphs to +k0 = max + + +3, + + +2 log +� +8n3 +8n3−(n+1)3 +� ++ 2 log +� +2 + +� +8 +(n+1)3 +� +log +� 2n +n+1 +� + + + + + +(178) += + + + + + +5 +if n = 5, +4 +if n = 9, +3 +if n ≥ 13 with n ≡ 1 (mod 4). +(179) + +38 +The constraint on n in (179) is the necessary condition on n in Remark 10. +D. Proofs for Section III-D +1) Proof of Proposition 4: +(a) Let G1, . . . , Gk be k simple, finite and undirected graphs, +��V(Gℓ) +�� = nℓ for ℓ ∈ [k], and let +G = G1 ⊠ . . . ⊠ Gk. We provide two alternative simple proofs of (65). +First proof: +χ(G) ≥ θ(G) +(180) +≥ | V(G)| +θ(G) +(181) += +k +� +ℓ=1 +| V(Gℓ)| +θ(Gℓ) , +(182) +where (180) holds by (16), (181) holds by (18), and equality (182) holds by (17) and since +V(G) = V(G1) × . . . × V(Gk). The ceiling operation can be add to the right-hand side of +(182) since a chromatic number is an integer. +Second proof: +χ(G) ≥ | V(G)| +α(G) +(183) +≥ | V(G)| +θ(G) +(184) += +k +� +ℓ=1 +| V(Gℓ)| +θ(Gℓ) , +(185) +where (183) holds by (150), (184) holds by (23), and (185) is (182). +We next prove (66). +χ(G) ≥ θ(G) +(186) += +k +� +ℓ=1 +θ(Gℓ), +(187) +where (186) holds by (15), and (187) holds by (17). +(b) Let G1, . . . , Gk be regular graphs, where Gℓ is dℓ-regular of order nℓ for all ℓ ∈ [k]. +Inequality (67) is (65). Inequality (68) follows from (19) and (67). Furthermore, by Item (a) +in Proposition 1, inequality (68) holds with equality if each regular graph Gℓ, for ℓ ∈ [k], is +either edge-transitive or strongly regular. +(c) By (182), with | V(Gℓ)| = nℓ, +| V(G)| +θ(G) += +k +� +ℓ=1 +nℓ +θ(Gℓ). +(188) +Suppose that, for all ℓ ∈ [k], Gℓ is dℓ-regular, and it is also either edge-transitive or strongly +regular. By Item (a) in Proposition 1, for all ℓ ∈ [k], +θ(Gℓ) = − nℓ λmin(Gℓ) +dℓ − λmin(Gℓ). +(189) + +39 +Combining (188) and (189) gives +| V(G)| +θ(G) += +k +� +ℓ=1 +� +1 − +dℓ +λmin(Gℓ) +� +. +(190) +On the other hand, since G = G1 ⊠ . . . ⊠ Gℓ is d-regular, with d ≜ d(G) as given in (70), it +follows from (19) that +θ(G) ≤ −| V(G)| λmin(G) +d(G) − λmin(G). +(191) +It should be noted, in regard to (191), that even if all Gℓ’s are regular and edge-transitive +graphs, their strong product G is not necessarily edge-transitive. In fact, G is not edge- +transitive, unless all the k factors {Gℓ}k +ℓ=1 are complete graphs (see [47, Theorem 3.1]). +For this reason, (191) does not hold in general with equality (see [65, Theorem 9]). Finally, +combing (190) and (191) gives inequality (69). +(d) Let G1, . . . , Gk be regular graphs, where Gℓ is dℓ-regular on nℓ vertices for all ℓ ∈ [k]. Then, +under the assumptions of Item (d), +(1) +k +� +ℓ=1 +| V(Gℓ)| +θ(Gℓ) += +k +� +ℓ=1 +θ(Gℓ) +(192) +≥ +k +� +ℓ=1 +ω(Gℓ) +(193) +where (192) holds since, by assumption, each of the graphs G1, . . . , Gk is vertex- +transitive or a strongly regular graph (this is because [65, Theorem 8] and (27) provide +different sufficient conditions for inequality (18) to hold with equality). Inequality (193) +holds by the leftmost inequality in (16). +(2) +k +� +ℓ=1 +� +1 − +dℓ +λmin(Gℓ) +� += +k +� +ℓ=1 +| V(Gℓ)| +θ(Gℓ) +(194) +≥ +k +� +ℓ=1 +ω(Gℓ) +(195) +where, by Item (a) of Proposition 1, equality (194) holds since (by assumption), for all +ℓ ∈ [k], the graph Gℓ is either regular and edge-transitive, or a strongly regular graph. +Inequality (195) holds under the same reasoning as of (192) and (193). +To summarize, it shows that under proper assumptions, the lower bound on the chromatic +number of G in the right-hand side of (65), or even its loosened bound in the right-hand +side of (68), are larger than or equal to the lower bound +k� +ℓ=1 +ω(Gℓ). +(e) By (66), +χ(G) ≥ +k +� +ℓ=1 +θ(Gℓ). +(196) + +40 +Let, for all ℓ ∈ [k], the graph Gℓ be dℓ-regular on nℓ vertices, and suppose that it is either +edge-transitive or strongly regular. Then, by Item (a) of Proposition 1, +θ(Gℓ) = − nℓ λmin(Gℓ) +dℓ − λmin(Gℓ), +∀ ℓ ∈ [k]. +(197) +Combining (196) and (197), followed by taking a ceiling operation on the lower bound on +the chromatic number χ(G), gives (71). +(f) By the assumption that G1, . . . , Gk are self-complementary, +θ(Gℓ) = θ(Gℓ), +∀ ℓ ∈ [k]. +(198) +Furthermore, by the assumption that for all ℓ ∈ [k], Gℓ is a graph on nℓ vertices that is +either vertex-transitive or strongly regular, +θ(Gℓ) θ(Gℓ) = nℓ, +∀ ℓ ∈ [k]. +(199) +Combining (198) and (199) gives +θ(Gℓ) = √nℓ, +∀ ℓ ∈ [k]. +(200) +Consequently, by (65) and (200), +χ(G) ≥ +� +k +� +ℓ=1 +nℓ +θ(Gℓ) +� +(201) += +� +k +� +ℓ=1 +√nℓ +� +(202) += +�√n +� +, +(203) +and, from (66) and (200), +χ(G) ≥ +� +k +� +ℓ=1 +θ(Gℓ) +� +(204) += +� +k +� +ℓ=1 +√nℓ +� +(205) += +�√n +� +. +(206) +This proves (72) and (73), and it completes the proof of Proposition 4. +2) Proof of Corollary 8: The rightmost inequality in (78) is a well-known upper bound on +the chromatic number of strong products (see [11], [15, Theorem 3]). The leftmost inequality +in (78) gives a lower bound on the chromatic number of a strong product of (not necessarily +distinct) non-complete, and strongly regular graphs. It readily follows by combining equality (26) +in Corollary 1, together with inequality (65) in Proposition 4. Finally, by Part 1 of Item (d) in +Proposition 4, the leftmost term in (78) is larger than or equal to the product of the clique +numbers of {Gℓ}k +ℓ=1. +ACKNOWLEDGMENTS +The author wishes to acknowledge the two anonymous reviewers for helpful and timely reports. + +41 +REFERENCES +[1] +E. Abbe, E. Boix-Adser´a, P. Ralli and C. Sandon, “Graph powering and spectral robustness,” SIAM +Journal on Mathematics of Data Science, vol. 2, no. 1, pp. 132–157, 2020. +[2] +E. Abbe and P. Ralli, “An Alon–Boppana theorem for powered graphs, and generalized Ramanujan +graphs,” preprint, June 2020. [Online] https://arxiv.org/abs/2006.11248v1. +[3] +E. Abbe and P. Ralli, “An Alon–Boppana theorem for powered graphs, generalized Ramanujan +graphs and robust community detection,” Proceedings of the 2022 International Zurich Seminar on +Information and Communication, pp. 19–23, Zurich, Switzerland, March 2–4, 2022. +[4] +A. Ac´ın, R. Duanc, D. E. Roberson, A. B. Sainz and A. Winter, “A new property of the Lov´asz +number and duality relations between graph parameters,” Discrete Applied Mathematics, vol. 216, +pp. 489–501, 2017. +[5] +M. Aigner and G. M. Ziegler, Proofs from THE BOOK, Sixth Edition, Springer, Berlin, 2018. +[6] +N. Alon, “Eigenvalues and expanders,” Combinatorica, vol. 6, no. 2, pp. 83–96, June 1986. +[7] +N. Alon, “Graph powers,” Contemporary Combinatorics, Bolyai Society Mathematical Studies and +Springer, vol. 10 (B´ela Bollob´as Editor), pp. 11–28, Budapest, Hungary, 2002. +[8] +N. Alon and J. H. Spencer, The Probabilistic Method, Fourth Edition, John Wiley & Sons, Hoboken, +New Jersey, USA, 2016. +[9] +N. Alon, “Lov´asz, vectors, graphs and codes,” Building Bridges II - Mathematics of L´aszl´o Lov´asz, +Bolyai Society Mathematical Studies and Springer, vol. 28 (I. B´ar´any, G. O. H. Katona and A. Sali +Editors), pp. 1–16, Budapest, Hungary, 2019. +[10] +N. Alon and E. Lubetzky, “The Shannon capacity of a graph and the independence numbers of its +powers,” IEEE Transactions on Information Theory, vol. 52, no. 5, pp. 2172–2176, May 2006. +[11] +C. Berge, Graphs and Hypergraphs, North-Holland Mathematical Library, vol. 6, 1973. +[12] +T. Bohman and R. Holzman, “A nontrivial lower bound on the Shannon capacities of the +complements of odd cycles,” IEEE Transactions on Information Theory, vol. 49, no. 3, pp. 721–722, +March 2003. +[13] +T. Bohman, R. Holzman, and V. Natarajan, “Maximum independent sets in certain powers of odd +cycles,” The Electronic Journal of Combinatorics, vol. 16, no. 1, paper 26, pp. 1–8, July 2009. +[14] +T. Bohman, R. Holzman, and V. Natarajan, “On the independence numbers of the cubes of odd +cycles,” The Electronic Journal of Combinatorics, vol. 20, no. 3, paper 10, pp. 1–19, August 2013. +[15] +M. Borowiecki, “On chromatic number of products of two graphs,” Colloquium Mathematicum, +vol. 25, no. 1, pp. 49–52, 1972. +[16] +V. E. Brimkov, B. Codenotti, V. Crespi, and M. Leoncini, “Efficient computation of the Lov´asz +number of certain circulant graphs,” in Graph-Theoretic Concepts in Computer Science, LNCS 3353, +J. Hrommkoviˇc, M. Nagl and B. Westfechtel, eds., Springer, pp. 285–295, Berlin, Germany, 2004. +[17] +V. Brimkov, “Algorithmic and explicit determination of the Lov´asz number for certain circulant +graphs,” Discrete Applied Mathematics, vol. 155, pp. 1812–1825, 2007. +[18] +A. E. Brouwer and W. H. Haemers, Spectra of Graphs, Springer, 2011. +[19] +A. E. Brouwer and H. Van Maldeghem, Strongly Regular Graphs, Cambridge University Press, +(Encyclopedia of Mathematics and its Applications, Series Number 182), 2022. +[20] +G. Chartrand, L. Lesniak, and P. Zhang, Graphs and Digraphs, Sixth Edition, CRC Press, 2015. +[21] +S. M. Cioabˇa, “On the extreme eigenvalues of regular graphs,” Journal of Combinatorial Theory, +Series B, vol. 96, no. 3, pp. 367–373, 2006. +[22] +S. M. Cioabˇa, M. R. Murty, “Expander graphs and gaps between primes,” Forum Mathematicum, +vol. 20, no. 4, pp. 745–756, 2008. +[23] +S. M. Cioabˇa, R. J. Elzinga and D. A. Gregory, “Some observations on the smallest adjacency +eigenvalue of a graph,” Discussiones Mathematicae Graph Theory, vol. 40, no. 2, pp. 467–493, +May 2020. + +42 +[24] +S. M. Cioabˇa and V. Gupta, “A lower bound for the smallest eigenvalue of a graph and an application +to the associahedron graph,” preprint, October 18, 2022. [Online] https://arxiv.org/abs/2210.08516. +[25] +S. M. Cioabˇa, M. R. Murty, A First Course in Graph Theory and Combinatorics, Second Edition, +Springer, 2022. +[26] +D. Cvetkovi´c, P. Rowlinson, S. Simi´c, An Introduction to the Theory of Graph Spectra, London +Mathematical Society Student Texts 75, Cambridge University Press, 2009. +[27] +M. Desai and V. Rao, “A characterization of the smallest eigenvalue of a graph,” Journal of Graph +Theory, vol. 18, no. 2, pp. 181–194, March 1994. +[28] +M. C. Dowling, Expander Graphs and Coding Theory, Ph.D. dissertation, Clemson University, +South Carolina, USA, September 2016. +[29] +P. Erd¨os, R. J. McEliece and H. Taylor, “Ramsey bounds for graph products,” Pacific Journal of +Mathematics, vol. 37, no. 1, pp. 45–46, January 1971. +[30] +L. Esperet and D. R. Wood, “Colouring strong products,” preprint, May 10, 2022. [Online] +https://arxiv.org/abs/2205.04953. +[31] +J. Feigenbaum and A. Sch¨affer, “Finding the prime factors of strong direct product graphs in +polynomial time,” Discrete Mathematics, vol. 109, no. 1–3, pp. 77–102, November 1992. +[32] +Y. Filmus, K. Golubev and N. Lifshitz, “High dimensional Hoffman bound and applications +in extremal combinatorics,” Journal of Algebraic Combinatorics, vol. 4, no. 6, pp. 1005–1026, +November 2021. +[33] +J. Friedman, “On the second eigenvalue and random walks in random d-regular graphs,” Combina- +torica, vol. 11, no. 4, pp. 331–362, December 1991. +[34] +J. Friedman, “Some geometric aspects of graphs and their eigenfunctions,” Duke Mathematical +Journal, vol. 69, no. 3, pp. 487–525, March 1993. +[35] +J. Friedman, A Proof of Alon’s Second Eigenvalue Conjecture and Related Problems, Memoirs of +the American Mathematical Society, vol. 195, no. 910, 2008. +[36] +J. Geetha and K. Somasundaram, “Total colorings of product graphs,” Graphs and Combinatorics, +vol. 34, pp. 339–347, February 2018. +[37] +C. Godsil and G. Royle, Algebraic Graph Theory, Springer, New York, 2001. +[38] +C. Godsil, D. E. Roberson, R. S´amal and S. Severini, “Sabidussi versus Hedetniemi for three +variations of the chromatic number,” Combinatorica, vol. 36, pp. 395–415, August 2016. +[39] +F. Guo and Y. Watanabe, “On graphs in which the Shannon capacity is unachievable by finite +product,” IEEE Transactions on Information Theory, vol. 36, no. 3, pp. 622–623, May 1990. +[40] +J. R. Griggs, “Lower bounds on the independence number in terms of the degrees,” Journal of +Combinatorial Theory, Series B, vol. 34, pp. 22–39, February 1983. +[41] +M. Gr¨otschel, L. Lov´asz, and A. Schrijver, “The ellipsoid method and its consequences in +combinatorial optimization,” Combinatorica, vol. 1, no. 2, pp. 168-197, June 1981. +[42] +W. Haemers, Eigenvalue Techniques in Design and Graph Theory, Ph.D. dissertation, Stichting +Mathematisch Centrum, Amsterdam, 1979. +[43] +W. Haemers, “On some problems of Lov´asz concerning the Shannon capacity of a graph,” IEEE +Transactions on Information Theory, vol. 25, no. 2, pp. 231–232, March 1979. +[44] +W. Haemers, “Hoffman’s ratio bound,” Linear Algebra and its Applications, vol. 617, pp. 215–219, +May 2021. +[45] +R. S. Hales, “Numerical invariants and the strong product of graphs,” Journal of Combinatorial +Theory, Series B, vol. 15, no. 2, pp. 146–155, October 1973. +[46] +R. Hammack, W. Imrich and S. Klavˇzar, Handbook of Product Graphs, Second Edition, CRC Press, +2011. +[47] +R. Hammack, W. Imrich and S. Klavˇzar, “Edge-transitive products,” Journal of Algebraic Combi- +natorics, vol. 43, no. 4, pp. 837–850, June 2016. + +43 +[48] +A. J. Hoffman, “On eigenvalues and colorings of graphs,” in Graph Theory and its Applications +(B. Harris, ed.), pp. 79–91, Academic Press, New York, 1970. +[49] +Y. Hong, “Bounds on eigenvalues of graphs,” Discrete Mathematics, vol. 123, pp. 65–74, 1993. +[50] +S. Hoory, N. Linial and A. Wigderson, “Expander graphs and their applications,” Bulletin of the +American Mathematical Society, vol. 43, no. 4, pp. 439–561, October 2006. +[51] +S. Hu, I. Tamo, and O. Shayevitz, “A bound on the Shannon capacity via a linear programming +variation,” SIAM Journal on Discrete Mathematics, vol. 32, no. 3, pp. 2229–2241, September 2018. +[52] +M. Jurkiewicz, “A survey on known values and bounds on the Shannon capacity,” Selected Topics in +Modern Mathematics, ed. G. Gancarzewicz and M. Skrzy´nski, Publishing House AKAPIT, Krak´ow, +Poland, pp. 115–128, 2014. +[53] +N. Kahale, “Eigenvalues and expansion of regular graphs,” Journal of the Association for Computing +Machinery, vol. 42, no. 5, pp. 1091–1106, September 1995. +[54] +S. Klavˇzar, “Strong products of χ-critical graphs,” Aequationes Mathematicae, vol. 45, pp. 153–162, +April 1993. +[55] +S. Klavˇzar, “Coloring graph products - a survey,” Discrete Mathematics, vol. 155, no. 1–3, pp. 135– +145, August 1996. +[56] +S. Klavˇzar and U. Mulitinovi´c, “Strong products of Kneser graphs,” Discrete Mathematics, vol. 133, +no. 1–3, pp. 287–300, 1994. +[57] +D. E. Knuth, “The sandwich theorem,” Electronic Journal of Combinatorics, vol. 1, pp. 1–48, 1994. +[58] +J. K¨orner and A. Orlitsky, “Zero-error information theory,” IEEE Transactions on Information +Theory, vol. 44, no. 6, pp. 2207–2229, October 1998. +[59] +M. Krivelevich, “Expanders – how to find them, and what to find in them,” Surveys in Combinatorics +2019, vol. 4, pp. 115–142, June 2019. +[60] +C. Li, “On finite graphs that are self-complementary and vertex-transitive,” Australasian Journal of +Combinatorics, vol. 18, pp.147-155, 1998. +[61] +W. C. W. Li (with an appendix by J. P. Serre), “On negative eigenvalues of regular graphs,” +Comptes Rendus de l’Acad´emie des Sciences - Series I - Mathematics, vol. 333, no. 10, pp. 907–912, +November 2001. +[62] +C. H. Li and G. Rao, “Self-complementary vertex-transitive graphs of order a product of two +primes,” Bulletin of the Australian Mathematical Society, vol. 89, no. 2, pp. 322–330, April 2014. +[63] +C. H. Li, G. Rao, and S. J. Song, “New constructions of self-complementary Cayley graphs,” Journal +of the Australian Mathematical Society, vol. 111, no. 3, pp. 372–385, December 2021. +[64] +D. Li-xin, L. Feng and Z. Hai-xing, “On the transitivity of the strong product of graphs,” Chinese +Quarterly Journal of Mathematics, vol. 30, no. 4, pp. 620–623, 2015. +[65] +L. Lov´asz, “On the Shannon capacity of a graph,” IEEE Transactions on Information Theory, vol. 25, +no. 1, pp. 1–7, January 1979. +[66] +L. Lov´asz, An Algorithmic Theory of Numbers, Graphs and Convexity, Philadelphia, PA, USA: +SIAM, 1986. +[67] +L. Lov´asz, Graphs and Geometry, American Mathematical Society, Colloquium Publications, +volume 65, 2019. +[68] +A. Lubotzky, “Expander graphs in pure and applied mathematics,” Bulletin of the American +Mathematical Society, vol. 49, no. 1, pp. 113–162, January 2012. +[69] +A. W. Marcus, D. A. Spielman, and N. Srivastava, “Interlacing families I: Bipartite Ramanujan +graphs of all degrees,” Annals of Mathematics, vol. 182, no. 1, pp. 307–325, July 2015. +[70] +R. J. McEliece and E. C. Posner, “Hide and seek, data storage, and entropy,” Annals of Mathematical +Statistics, vol. 42, no. 5, pp. 1706–1716, October 1971. +[71] +J. Morris, C. E. Praeger and P. Spiga, “Strongly regular edge-transitive graphs,” Ars Mathematica +Contemporanea, vol. 2, no. 2, pp. 137-155, September 2009. + +44 +[72] +N. E. Mullin, Self-Complementary Arc-Transitive Graphs and Their Imposters, Master’s thesis, +University of Waterloo, Canada, 2009. +[73] +M. R. Murty, “Ramanujan graphs: an introduction,” Indian Journal of Discrete Mathematics, vol. 6, +no. 2, pp. 91–127, 2020. +[74] +A. Neumaier, “Cliques and claws in edge-transitive strongly regular graphs,” Mathematische +Zeitschrift, vol. 174, no. 3, pp. 197–202, October 1980. +[75] +A. Neumaier, “Regular cliques in graphs and special 1 1 +2 designs,” Finite Geometries and Designs +(editors: P. J. Cameron, J. W. P. Hirschfeld, and D. R. Hughes), pp. 244–259, 1981. +[76] +V. Nikiforov, “Eigenvalue problems of Nordhaus–Gaddum type,” Discrete Mathematics, vol 307, +no. 6, pp. 774–780, March 2007. +[77] +V. Nikiforov and X. Yuan, “More eigenvalue problems of Nordhaus–Gaddum type,” Linear Algebra +and its Applications, vol. 451, pp. 231–245, June 2014. +[78] +A. Nilli, “On the second eigenvalue of a graph,” Discrete Mathematics, vol. 91, no. 2, pp. 207–210, +August 1991. +[79] +A. Nilli, “Tight estimates for eigenvalues of regular graphs,” Electronic Journal of Combinatorics, +vol. 11, no. 1, paper N9, pp. 1–4, May 2004. +[80] +E. A. Nordhaus and J. Gaddum, “On complementary graphs,” American Mathematical Monthly, +vol. 63, no. 3, pp. 175–177, March 1956. +[81] +W. Peisert, “All self-complementary symmetric graphs,” Journal of Algebra, vol. 240, no. 1, pp. 209– +229, June 2001. +[82] +G. Rao, “Self-complementary vertex-transitive graphs,” PhD dissertation, the University of Western +Australia, September 2014. +[83] +G. Rao, “Self-complementary vertex-transitive graphs,” Bulletin of the Australian Mathematical +Society, vol. 94, no. 1, pp. 165-–166, 2016. +[84] +G. Sabidussi, “Graph multiplication,” Mathematische Zeitschrift, vol. 72, no. 1, pp. 446–457, 1960. +[85] +The Sage Developers, SageMath, the Sage Mathematics Software System (Version 9.2), 2020. +[86] +C. E. Shannon, “The zero error capacity of a noisy channel,” IEEE Transactions on Information +Theory, vol. 2, no. 3, pp. 8–19, September 1956. +[87] +E. Sonnemann and O. Krafft, “Independence numbers of product graphs,” Journal of Combinatorial +Theory Series B, vol. 17, no. 2, pp. 133–142, October 1974. +[88] +D. Spielman, “Spectral graph theory,” Chapter 18 (pp. 495–524) in Combinatorial Scientific +Computing (U. Naumann and O. Schenk editors), Chapman & Hall CRC Press, 2012. +[89] +Z. Stani´c, Inequalities for Graph Eigenvalues, London Mathematical Society Lecture Note Series, +Series Number 423, Cambridge University Press, 2015. +[90] +K. Vesztergombi, “Some remarks on the chromatic number of the strong product of graphs,” Acta +Cybernetica, vol. 4, no. 2, pp. 207–212, January 1979. +[91] +K. Vesztergombi, “Chromatic number of strong product of graphs,” in Algebraic Methods in Graph +Theory (editors: V. T. S´os and L. Lov´asz), pp. 819–825, Elsevier, 1981. +[92] +V. K. Wei, “A lower bound on the stability number of a simple graph,” Bell Laboratories Technical +Memorandum, 81–11217–9, Murray Hill, New Jersey, 1981. +[93] +H. S. Wilf, “The eigenvalues of a graph and its chromatic number,” Journal of the London +Mathematical Society, vol. s1–42, no. 1, pp. 330–332, January 1967. +[94] +H. Witsenhausen, “The zero-error side information problem and chromatic numbers,” IEEE +Transactions on Information Theory, vol. 22, no. 5, pp. 592–593, September 1976. +[95] +X. Xu and S. P. Radziszowski, “Bounds on Shannon capacity and Ramsey numbers from product +of graphs,” IEEE Transactions on Information Theory, vol. 59, no. 8, pp. 4767–4770, August 2013. +[96] +M. Zhai, H. Lin and B. Wang, “Sharp upper bounds on the second largest eigenvalues of connected +graphs,” Linear Algebra and Its Applications, vol. 437, no. 1, pp. 236–241, July 2012. + diff --git a/y9E0T4oBgHgl3EQf_AIq/content/tmp_files/load_file.txt b/y9E0T4oBgHgl3EQf_AIq/content/tmp_files/load_file.txt new file mode 100644 index 0000000000000000000000000000000000000000..b0c85aa87c21955abbf8efa92514885831db37db --- /dev/null +++ b/y9E0T4oBgHgl3EQf_AIq/content/tmp_files/load_file.txt @@ -0,0 +1,1667 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf,len=1666 +page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='02820v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='CO] 7 Jan 2023 Observations on the Lov´asz θ-Function, Graph Capacity, Eigenvalues, and Strong Products Igal Sason Dedicated to my friend and former teacher, Professor Emeritus Abraham (Avi) Berman, in the occasion of his eightieth birthday Citation: I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sason, “Observations on the Lov´asz θ-function, graph capacity, eigenvalues, and strong products,” Entropy, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 25, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, paper 104, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–41, January 2023.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' DOI: https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3390/e25010104.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Abstract This paper provides new observations on the Lov´asz θ-function of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These include a simple closed-form expression of that function for all strongly regular graphs, together with upper and lower bounds on that function for all regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These bounds are expressed in terms of the second-largest and smallest eigenvalues of the adjacency matrix of the regular graph, together with sufficient conditions for equalities (the upper bound is due to Lov´asz, followed by a new sufficient condition for its tightness).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These results are shown to be useful in many ways, leading to the determination of the exact value of the Shannon capacity of various graphs, eigenvalue inequalities, and bounds on the clique and chromatic numbers of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since the Lov´asz θ-function factorizes for the strong product of graphs, the results are also particularly useful for parameters of strong products or strong powers of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Bounds on the smallest and second-largest eigenvalues of strong products of regular graphs are consequently derived, expressed as functions of the Lov´asz θ-function (or the smallest eigenvalue) of each factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The resulting lower bound on the second-largest eigenvalue of a k-fold strong power of a regular graph is compared to the Alon–Boppana bound;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' under a certain condition, the new bound is superior in its exponential growth rate (in k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lower bounds on the chromatic number of strong products of graphs are expressed in terms of the order and the Lov´asz θ-function of each factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The utility of these bounds is exemplified, leading in some cases to an exact determination of the chromatic numbers of strong products or strong powers of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The present research paper is aimed to have tutorial value as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Keywords: Lov´asz θ-function, Shannon capacity of a graph, strongly regular graph, strong product of graphs, vertex- and edge-transitivity, Alon–Boppana bound, Ramanujan graph, chromatic number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' INTRODUCTION The notion of the graph capacity in Shannon’s problem of zero-error communication [86] had a significant impact on the development of information theory and graph theory, including the introduction of perfect graphs by C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Berge [11], strong graph products (or powers) ([7], [84]), the introduction of the Lov´asz θ-function of a graph as a computable upper bound on its Shannon capacity ([57], [65]), the rank-bound by Haemers [43], and other important follow-up works that are surveyed, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', in [7], [9], [52], [58].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In graph theory, there are four central sorts of graph products, each with its own applications and theoretical interpretations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The reader is referred to the excellent handbook [46], which presents the rich and fertile field of graph products.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Strong product of graphs is one of the most extensively studied sorts of graph products, and there exists a polynomial-time algorithm that I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sason is with the Viterbi Faculty of Electrical and Computer Engineering, and the Department of Mathematics at the Technion - Israel Institute of Technology, Haifa 3200003, Israel (e-mail: eeigal@technion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='il).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2 finds the unique prime factorization of any connected graph with that type of multiplication [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Strong powers of graphs are also fundamental in information theory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Their information- theoretic significance stems from the notion of the Shannon capacity of graphs for error-free communication [86], and the Witsenhausen rate [94] in the zero-error source coding problem with perfect side information at the receiver.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Properties of strong products and strong powers of graphs, and bounds on their independence numbers and chromatic numbers have been extensively studied, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', in [4], [7], [9], [10], [12], [13], [14], [15], [16], [17], [29], [30], [31], [36], [38], [39], [43], [45], [46], [47], [51], [52], [54], [55], [56], [64], [65], [70], [84], [86], [87], [90], [91], [94], [95].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The present work continues the above paths of research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It provides some new observations on the Lov´asz θ-function of regular graphs, calculation of the Shannon capacity of some strongly regular graphs, bounds on eigenvalues of graphs (in particular, the second-largest and smallest eigenvalues of the adjacency matrix, which play a key role in spectral graph theory), bounds related to Ramanujan graphs, and strong products of graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The analysis in the present work mainly relies on the notion of the Lov´asz θ-function of graphs [65].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The paper includes a thorough review of the backgrounds relevant to this work with suitable references or explanations, which also serve to motivate the presentation of the results in this work and to put them into perspective.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The presentation in this research paper is consequently aimed to have tutorial value as well.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The results obtained in this work are outlined as follows: (1) A known upper bound on the Lov´asz θ-function of a regular graph is expressed in terms of the smallest eigenvalue of its adjacency matrix [65].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A key result in this work provides a lower bound on the Lov´asz θ-function of a regular graph, which is expressed in terms of the second-largest eigenvalue of its adjacency matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' New sufficient conditions for equalities in these bounds are also obtained (Proposition 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (2) A simple and closed-form expression of the Lov´asz θ-function is derived for all strongly regular graphs (Corollary 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (3) Eigenvalue inequalities are derived, which relate the smallest and second-largest eigenvalues of a regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' They hold with equality if and only if the graph is strongly regular (Corollaries 2 and 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (4) The Shannon capacity of several strongly regular graphs is determined (Section III-E).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (5) Bounds on parameters of regular graphs, and in particular of Ramanujan graphs, are derived (Corollaries 4–6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (6) Bounds on the smallest and the second-largest eigenvalues of strong products of regular graphs are derived, which are expressed in terms of calculable parameters of its factors (Proposition 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (7) A new lower bound on the second-largest eigenvalue of a k-fold strong power of a regular graph is compared to the Alon–Boppana bound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Under a certain condition, the former bound shows an improvement in its exponential growth rate as a function of k (Section III-C).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (8) Every non-complete and non-empty connected regular graph, whose Lov´asz θ-function is below a certain value, is proved to have the property that almost all its strong powers are highly non-Ramanujan (Proposition 3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3 (9) Lower bounds on the chromatic number of strong products of graphs are expressed in terms of the order and Lov´asz θ-function of each factor (Proposition 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Their utility is exemplified, while also leading to exact chromatic numbers in some cases.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The paper is structured as follows: Section II provides notation and a thorough review of the backgrounds relevant to this work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Section III provides the results of this work, followed by examples and discussions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is composed of five subsections that address issues related to the Lov´asz θ-function, Shannon capacity of graphs, Ramanujan graphs, the second-largest and smallest eigenvalues of strong products or strong powers of graphs, and the chromatic numbers of such graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Section IV proves the results in Section III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' PRELIMINARIES This section provides essential notation and preliminaries for this paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following standard notation in set theory is used: N = {1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='} is the set of natural numbers, R is the set of real numbers, and [n] ≜ {1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , n} with n ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The cardinality of a set A is a measure of its number of elements;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' it is denoted by |A|, and (by definition) it is equal to its number of elements if A is a finite set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a graph, and let V(G) and E(G) denote, respectively, the sets of vertices and edges in G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The order and size of a graph G are defined to be | V(G)| and | E(G)|, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph G is said to be finite if V(G) is a finite set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A pair of vertices are adjacent in a graph G if these two vertices are the endpoints of an edge e ∈ E(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph is called simple if it has no loops (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', it has no edge with identical endpoints), and if there are no multiple edges between any pair of adjacent vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph G is said to be undirected if its edges have no directions;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' otherwise, it is a directed graph, a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' a digraph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Throughout this paper, it is assumed that the graphs under consideration are finite, undirected, and simple.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An empty graph is an edgeless graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph G is said to be an r-partite graph if its vertex set V(G) is a disjoint union of r subsets such that every pair of vertices that are elements of an identical subset are non-adjacent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If r = 2, then it is a bipartite graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A walk in a graph G is a sequence of its vertices such that (by definition) every pair of consecutive vertices are adjacent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A path in a graph is a walk with no repeated vertices (in other words, a path is a walk along the vertices of G such that no vertex can be visited twice).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The length of a path is defined as its number of edges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, P = [v1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , vℓ] is a path in a graph G if {vi, vi+1} ∈ E(G) for all i ∈ [ℓ − 1], and all the vertices in the sequence {vi}ℓ i=1 are distinct;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' the endpoints of the path P are v1 and vℓ, and its length is equal to ℓ − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A cycle C in a graph G is obtained by adding an edge to a path P such that it gives a closed walk (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', a walk whose endpoints are identical).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The cycle C = [v1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , vℓ, v1] is of length ℓ, which is obtained by adding the edge e = {vℓ, v1} ∈ E(G) to the above (ℓ − 1)–length path P;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' the two identical endpoints of the cycle C are the vertex v1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph is said to be connected if every two distinct vertices in that graph are connected by a path (otherwise, it is a disconnected graph).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The distance between a pair of distinct and connected vertices in a graph is defined to be the length of the shortest path whose two endpoints are the given pair of vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The distance between two disconnected vertices in a graph is defined to 4 be infinite, and the distance between a vertex and itself is set to be zero.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The diameter of a graph is the maximum distance between any pair of vertices in that graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A connected and finite graph has a finite diameter, and the diameter of a disconnected graph is (by definition) infinite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following standard notation in graph theory is used: (a) Kn denotes the complete graph with n ∈ N vertices, where every pair of distinct vertices are adjacent;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' hence, K1 is an empty graph with a single vertex.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) Kn,m denotes the complete bipartite graph, which is a bipartite graph consisting of a vertex set that is a disjoint union of two finite sets V1 and V2 of cardinalities n and m, respectively, and a set of edges that are all the possible connections of a vertex in V1 and a vertex in V2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (c) Pn denotes an (n − 1)–length path with n ∈ N, which is a graph with n vertices that forms a path of length n − 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' in particular, P1 = K1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (d) Cn denotes an n-length cycle, which is a graph with n ≥ 3 vertices that forms a cycle of length n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (e) K(m, r) denotes the Kneser graph with integers 1 ≤ r ≤ m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It has n = �m r � vertices, represented by all r-subsets of [m].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Two vertices are adjacent in that graph if they are represented by disjoint r-subsets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The graph K(m, r), provided that it has more than one vertex, is a connected graph if and only if either m > 2r or (m, r) = (2, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A subgraph is a graph that exists within another graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' More formally, F is a subgraph of a graph G if V(F) ⊆ V(G) and E(F) ⊆ E(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If F is a subgraph of G, we write F ⊆ G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A spanning subgraph is obtained by edge deletions from the original graph, while its vertex set is left unchanged.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An induced subgraph is obtained by removing vertices from the original graph, followed by the deletion of the edges that are adjacent to these removed vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A clique in a graph G is a subset of pairwise adjacent vertices in G (in other words, the induced subgraph of G on that subset is a complete subgraph).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The maximum size of a clique in G is called the clique number of G, and it is denoted by ω(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Similarly, an independent set (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' coclique) in G is a subset of pairwise non-adjacent vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The maximum size of an independent set is called the independence number of G, and it is denoted by α(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A proper vertex coloring of G is a coloring of its vertices such that no pair of adjacent vertices are assigned the same color.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The smallest required number of colors for a proper coloring of the vertices in G is called the chromatic number of G, and it is denoted by χ(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G = (V(G), E(G)) be a finite, undirected, and simple graph of order | V(G)| = n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Define the adjacency matrix A = A(G) of the graph G to be an n-times-n symmetric matrix such that A = (ai,j)1≤i,j≤n with ai,j = 1 if {i, j} ∈ E(G), and ai,j = 0 otherwise (hence, the diagonal elements of A are in particular zeros).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let the eigenvalues of A (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' the eigenvalues of G) be given in decreasing order by λmax(G) = λ1(G) ≥ λ2(G) ≥ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ≥ λn(G) = λmin(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (1) The spectrum of G consists of the eigenvalues of A, including their multiplicities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The terms λ1(G), λ2(G), and λn(G) are referred to as the largest, second-largest, and smallest eigenvalues of the graph G, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The second-largest and smallest eigenvalues play a key role in spectral graph theory, and the interested reader is referred to, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [18], [25], [26], [88], [89].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 5 The number of edges that are incident to a vertex in a graph is called the degree of the vertex, and a graph is said to be regular if its vertices have an identical degree.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A regular graph whose all vertices have a fixed degree d is called a d-regular graph, and it is said to have valency d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A d-regular graph has a largest eigenvalue λ1(G) = d, with the all-ones column vector (of length n) as an eigenvector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a finite, simple and undirected graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The complement of G, denoted by G, is defined to have the same vertex set as G, and (by definition) any pair of distinct vertices in V(G) are adjacent in G if and only if they are non-adjacent in the graph G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, we have the equalities | E(G)| + | E(G)| = 1 2 n(n − 1), and V(G) = V(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Furthermore, let Jn and In denote the n-times-n all-ones and identity matrices, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, by definition, it follows that the adjacency matrices of the graph G and its complement G are related by the equality A(G) = Jn − In − A(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (2) Let G be a d-regular graph on n vertices, and let G be the complement graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, by (2), the spectra of G and G are related as follows [18, Section 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2]: λ1(G) = n − d − 1 = n − 1 − λ1(G), (3) λℓ(G) = −1 − λn+2−ℓ(G), ℓ = 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (4) Specifically, setting ℓ = n in (4) gives λmin(G) = −1 − λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (5) A graph is called acyclic if it has no cycles, and a connected acyclic graph is called a tree.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A tree with n vertices has n − 1 edges, and for every pair of distinct vertices in a tree, there is a unique path joining them (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A leaf in a tree is a vertex of degree 1, and every tree contains at least two leaves (a tree is therefore not a regular graph, unless it is the complete graph on two vertices K2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A disjoint union of trees is a forest (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', a forest is an acyclic graph whose connected components are trees).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In particular, a deletion of an edge from a tree gives a forest of two trees.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The interested reader is referred to, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Chapter 5] for further properties and analysis related to trees and forests.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Two simple graphs G = (V(G), E(G)) and H = (V(H), E(H)) are said to be isomorphic if there is a bijection (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', a one-to-one and onto mapping) f : V(G) → V(H) that preserves adjacency and non-adjacency, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', {u, v} ∈ E(G) if and only if {f(u), f(v)} ∈ E(H).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The notation G ∼= H denotes that G and H are isomorphic graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An isomorphism from a graph to itself is called an automorphism of the graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph is called self-complementary if G and G are isomorphic graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These include, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', the trivial complete graph on one vertex K1, the length-3 path P4, and the 5-cycle graph C5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If G is a self-complementary graph of order n, then the size of G is m = | E(G)| = 1 2 �n 2 � = 1 4 n(n−1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since only n or n − 1 can be even, either n ≡ 0 (mod 4) or n ≡ 1 (mod 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For every such n, there exists a recursive algorithm for constructing a self-complementary graph of order n (see [20, Exercise 29]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' More explicitly, if G is a self-complementary graph of order n, then its disjoint union with the length-3 path P4 = [v1, v2, v3, v4], where each of the vertices v2 and v3 in P4 is connected to all the vertices in G, gives a self-complementary graph of order 6 n + 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Starting with a graph G that is equal to P4 or C5 (for graph orders of n = 4 or n = 5, respectively) gives, by the above suggested recursive construction, a self-complementary graph of order n for all integers n > 1 such that n ≡ 0 (mod 4) or n ≡ 1 (mod 4), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph G is vertex-transitive if for every two vertices of G, there is an automorphism of G that maps one vertex to the other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Similarly, G is said to be an edge-transitive graph if for every two edges {u1, v1} and {u2, v2} of G, there is an automorphism f : V(G) → V(G) of the graph G, such that {u2, v2} = {f(u1), f(v1)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A vertex-transitive graph is necessarily regular, but the opposite does not hold in general.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Unlike a vertex-transitive graph, an edge-transitive graph is not necessarily regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Types of graph transitivity are studied, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', in [37, Chapters 3–4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' All the eigenvalues of a d-regular graph G are bounded in absolute value by d, and the largest eigenvalue is equal to d with an eigenvector that is equal to the all-ones column vector (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Proposition 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The multiplicity of the largest eigenvalue of a d-regular graph is 1 if and only if G is connected (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2 and Proposition 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A graph G has its eigenvalues symmetric around zero (including their multiplicities) if and only if G is bipartite (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, the smallest eigenvalue of a d-regular bipartite graph is equal to −d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Moreover, if G is d-regular and connected, then G is bipartite if and only if −d is an eigenvalue of its adjacency matrix (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Proposition 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For a d-regular graph G, let λ(G) ≜ max ℓ: λℓ(G)̸=±d ��λℓ(G) ��.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (6) A d-regular graph G is called Ramanujan if λ(G) ≤ 2 √ d − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (7) The reason for the expression in the right-hand side of (7) is related to the Alon–Boppana bound, which addresses the question of how small can the second-largest eigenvalue be for a connected d-regular graph or for a sequence of connected d-regular graphs whose number of vertices tends to infinity (the value of the parameter d is kept fixed).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It states that for every d-regular graph G on n vertices, with d ≥ 3, λ2(G) ≥ 2 √ d − 1 − O � (logd n)−1� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (8) A non-asymptotic version of (8) appears in [78] (see also [25, Theorem 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Alon– Boppana bound was first mentioned in [6, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 95], and it was analyzed, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', in [1], [2], [3], [21], [22], [33], [34], [35], [61], [78], [79].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Moreover, all eigenvalues of a tree, with maximum degree d ≥ 2, are (in absolute value) at most 2 √ d − 1 (see [73, Theorem 1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Examples of Ramanujan graphs include: (a) The complete d-regular graph Kd+1, with d ≥ 2, whose eigenvalues are equal to d with multiplicity 1, and −1 with multiplicity d;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) The complete bipartite graph Kd,d, with d ≥ 2, is a d-regular graph whose two nonzero eigenvalues are ±d (each of multiplicity 1), and its other 2d − 2 eigenvalues are zeros.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (c) The Petersen graph, which is isomorphic to the Kneser graph K(5, 2), is a Ramanujan graph since it is 3-regular with the distinct eigenvalues 3, −1, and −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The interested reader is referred to [73] for a recent survey paper on Ramanujan graphs, and to references therein (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [69] that proves the existence of infinite families of bipartite Ramanujan graphs of every degree greater than 2, followed by the extension of that result 7 to bi-regular bipartite graphs;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' the proof in [69] uses an original technique for controlling the eigenvalues of some random matrices).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a d-regular graph of order n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The graph G is said to be a strongly regular graph if there exist nonnegative integers λ and µ such that the following two conditions hold: Every pair of adjacent vertices have exactly λ common neighbors;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Every pair of distinct and non-adjacent vertices have exactly µ common neighbors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Such a strongly regular graph is denoted by srg(n, d, λ, µ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Some basic properties of strongly regular graphs are next introduced, which also serve in our analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (a) The complement of a strongly regular graph is also strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' More explicitly, the complement of srg(n, d, λ, µ) is given by srg(n, n − d − 1, n − 2d + µ − 2, n − 2d + λ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) The four parameters of a strongly regular graph srg(n, d, λ, µ) satisfy the relation (n − d − 1) µ = d (d − λ − 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (9) (c) A strongly regular graph srg(n, d, λ, µ) has at most three distinct eigenvalues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If it is connected, then λ1(G) = d (multiplicity 1), and the other two distinct eigenvalues are p1,2 = 1 2 � λ − µ ± � (λ − µ)2 + 4(d − µ) � , (10) whose respective multiplicities are given by m1,2 = 1 2 � n − 1 ∓ 2d + (n − 1)(λ − µ) � (λ − µ)2 + 4(d − µ) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (11) Since multiplicities of eigenvalues must be nonnegative integers, their expressions in (11) provide further constraints on the values of n, d, λ and µ (in addition to equality (9)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (d) A connected regular graph with exactly three distinct eigenvalues is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (e) A strongly regular graph srg(n, d, λ, µ), with µ > 0, is a connected graph whose diameter is equal to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This holds since two non-adjacent vertices have µ > 0 common neighbors, so the distance between any pair of non-adjacent vertices is equal to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This can be also explained by spectral graph theory since the diameter of a connected graph is strictly smaller than the number of its distinct eigenvalues (see [25, Theorem 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In light of that, the above claim about the diameter holds for all graphs that are connected and strongly regular since these graphs only have three distinct eigenvalues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (f) If µ = 0, the strongly regular graph is disconnected, and it is a disjoint union of equal- sized complete graphs (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', a disjoint union of cliques of the same size).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A disjoint union of an arbitrary number ℓ ≥ 2 of equal-sized complete graphs, Kd+1, has the parameters srg((d + 1)ℓ, d, d − 1, 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In that case, d = p1 (see (10)), so the largest and second-largest eigenvalues coincide (by (11), that common eigenvalue has multiplicity m1 + 1 = ℓ in the graph spectrum).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A strongly regular graph G is called primitive if both G and its complement G are connected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Otherwise, G is called imprimitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An imprimitive graph is either a disjoint union of equal-sized complete graphs or its complement, which is a complete multipartite graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A strongly regular graph G is imprimitive if and only if 0 or −1 is an eigenvalue of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 8 (g) Let G be a primitive strongly regular graph srg(n, d, λ, µ) with the largest eigenvalue d (multiplicity 1), second-largest eigenvalue r = p1 (multiplicity m1), and smallest eigenvalue s = p2 (multiplicity m2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By (3) and (4), the complement G is a primitive strongly regular graph, having the largest eigenvalue n − d − 1 (multiplicity 1), second-largest eigenvalue −1 − s (multiplicity m2), and smallest eigenvalue −1 − r (multiplicity m1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Each of these primitive strongly regular graphs has three distinct eigenvalues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The reader is referred to [19], which is focused on properties and constructions of strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Part of this work is focused on the strong product of graphs, which is defined as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Definition 1: Let G1 and G2 be two graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The strong product G = G1 ⊠ G2 is a graph whose vertex set is V(G) = V(G1) × V(G2) (a Cartesian product), and distinct vertices {u1, u2} and {v1, v2} in G are adjacent if one of the following three conditions is satisfied: (a) u1 = v1 and {u2, v2} ∈ E(G2), (b) {u1, v1} ∈ E(G1) and u2 = v2, (c) {u1, v1} ∈ E(G1) and {u2, v2} ∈ E(G2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A strong product of graphs is commutative in the sense that G1 ⊠ G2 ∼= G2 ⊠ G1, (12) and it is also associative in the sense that (G1 ⊠ G2) ⊠ G3 ∼= G1 ⊠ (G2 ⊠ G3), (13) for every three graphs G1, G2 and G3 (see [46, Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This paper relies on the Lov´asz θ-function of a graph [65], which is next introduced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Definition 2: Let G be a simple graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An orthogonal representation of G in the d-dimensional Euclidean space (Rd) assigns to each vertex i ∈ V(G) a vector ui ∈ Rd such that uT i uj = 0 if {i, j} /∈ E(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In other words, the vertices of a simple graph are assigned vectors in Rd such that the vectors that are assigned to any pair of distinct and non-adjacent vertices of that graph are orthogonal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An orthonormal representation of G is an orthogonal representation of that graph such that all representing vectors have unit length.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 1: In an orthogonal representation of a graph G, non-adjacent vertices are mapped to orthogonal vectors, although adjacent vertices are not necessarily mapped to non-orthogonal vectors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If the latter condition is satisfied, then the orthogonal representation is said to be faithful.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Definition 3: Let G be a finite, undirected and simple graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its Lov´asz θ-function is given by θ(G) ≜ min u,c max i∈V(G) 1 � cTui �2 , (14) where the minimum is taken over all orthonormal representations {ui : i ∈ V(G)} of G, and all unit vectors c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The unit vector c is called the handle of the orthonormal representation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By the Cauchy-Schwarz inequality, ��cTui �� ≤ ∥c∥ ∥ui∥ = 1, so θ(G) ≥ 1 with equality if and only if G is a complete graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Lov´asz θ-function of a graph can be written as a semidefinite program, which satisfies strong duality ([41], [57], [66] and [67, Section 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This enables to compute the value of 9 θ(G) in polynomial time [41].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' More precisely, there is an algorithm that computes, for every graph G and every ε > 0, a real number t such that ��θ(G) − t �� < ε, where the running time of the algorithm is polynomial in n ≜ | V(G)| and log � 1 ε � [67, Theorem 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following properties of the Lov´asz θ-function are used throughout this paper: (a) The sandwich theorem ([41], [57], [66, Lemma 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='4], [67, Theorem 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]) is stated in the two equivalent forms α(G) ≤ θ(G) ≤ χ(G), (15) ω(G) ≤ θ(G) ≤ χ(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (16) (b) [65, Theorem 7]: The Lov´asz θ-function factorizes for the strong product of graphs, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', θ(G1 ⊠ G2) = θ(G1) θ(G2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (17) (c) [65, Corollary 2] and [65, Theorem 8]: θ(G) θ(G) ≥ | V(G)|, (18) with equality in (18) if the graph G is vertex-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (d) [65, Theorem 9]: Let G be a d-regular graph of order n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, θ(G) ≤ − n λn(G) d − λn(G), (19) where λn(G) < 0, unless G is an empty graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Equality holds in (19) if G is edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (e) Two simple observations relating the Lov´asz θ-functions of a graph and its subgraphs: If F is a spanning subgraph of a graph G, then θ(F) ≥ θ(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If F is an induced subgraph of a graph G, then θ(F) ≤ θ(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (f) [4, Theorem 2]: Although unrelated to the analysis in this paper, another interesting property of the Lov´asz θ-function is given by the identity sup H α(G ⊠ H) θ(G ⊠ H) = 1, (20) which holds for every simple, finite, and undirected graph G, where the supremum is taken over all such graphs H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that the leftmost inequality in (15) can be made arbitrarily tight by looking at the strong product of the given graph G with a suitable graph H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Shannon capacity of a simple, finite and undirected graph G was introduced in [86] to determine the maximum information rate that enables error-free communication.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' To that end, a discrete memoryless communication channel is represented by a confusion graph G that is constructed as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The vertices in the graph are represented by the symbols of the input alphabet to that channel, and any two distinct vertices in that graph are adjacent if the corresponding two input symbols are not distinguishable by the channel (in the sense that there exists an output symbol such that the transition probabilities from each of these two input symbols to that output symbol are strictly positive).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This means that the exact values of the positive transition probabilities of the channel, as well as the output alphabet of the channel, are irrelevant to the construction of the confusion graph G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The rationality in doing so is the interest to pictorially represent (by a graph) all those pairs of input symbols that are not distinguishable 10 by the channel.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Consider a transmission of k-length strings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The k-th confusion graph of the channel is defined as G⊠ k ≜ G ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ G � �� � k−1 strong products , (21) which is the k-fold strong power of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This is because the independence number of G⊠ k is equal to the maximum number of k-length strings at the channel input that can be transmitted with error-free communication (indeed, a pair of non-adjacent vertices in G⊠ k represent k-length strings that can be distinguished by the channel, as a result of having a common position in these two input strings where the corresponding two symbols at that position are distinguishable by the channel).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Consequently, the maximum information rate per symbol that is achievable by using input strings of length k is equal to 1 k log α(G⊠ k) = log k� α(G⊠ k), for all k ∈ N (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', it is the logarithm of the maximum number of k-length input strings that are distinguishable by the channel, normalized by the length k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Shannon capacity of a graph G is defined to be the (exponent of the) maximum information rate per symbol that is achievable with error-free communication, where the transmission takes place over a discrete memoryless channel whose confusion graph is equal to G, and the length of the input strings to the channel is unlimited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is denoted by Θ(G) (recall that the Lov´asz θ-function is denoted by θ(G)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Taking the supremum over k, the Shannon capacity of G is given by (see [86], and [5, Chapter 42]) Θ(G) = sup k∈N k� α(G⊠ k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (22) The Shannon capacity can be rarely computed exactly (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [7], [9], [43], [51], [52], [58], [65], [86]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Analytical observations that also explain why it is, in general, even difficult to approximate it are addressed in [10] and [39].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Calculable upper bounds on Θ(G) were derived by Shannon [86], Lov´asz [65], Haemers [43], and more recently by Hu et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [51].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Lov´asz θ-function θ(G) is a calculable upper bound on the graph capacity Θ(G), i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', α(G) ≤ Θ(G) ≤ θ(G), (23) where the leftmost inequality in (23) follows from (22) (by setting k = 1), and the rightmost inequality in (23) is [65, Theorem 1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In regard to the rightmost inequality in (23), it is also shown in [9] that the Lov´asz θ-function of a graph, θ(G), cannot be upper bounded by any function of its Shannon capacity Θ(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' As mentioned above, the computational task of the Lov´asz θ-function, θ(G), is in general feasible by semidefinite programming.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Fundamental graph parameters such as its Shannon capacity Θ(G), independence number α(G), clique number ω(G), and chromatic number χ(G) are all NP-hard problems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The polynomial-time computability of the Lov´asz θ-function of a graph makes inequalities (15), (16), and (23) very useful in obtaining polynomial-time computable upper bounds on the independence number, clique number, and the Shannon capacity of a graph, as well as having such a computable lower bound on the chromatic number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 11 III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' THEOREMS, DISCUSSIONS AND EXAMPLES The present section provides the results of this work, followed by examples and discussions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is composed of five subsections that address issues related to the Lov´asz θ-function, Shannon capacity of graphs, Ramanujan graphs, eigenvalues, and chromatic numbers of strong graph products or strong graph powers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Bounds on Lov´asz θ-function, and an exact result for strongly regular graphs Let G be a d-regular graph with n vertices, and let G be the complement graph of G that is an (n−d−1)-regular graph of order n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An upper bound on θ(G) and a lower bound on θ(G) were obtained by Lov´asz, expressed in terms of the smallest eigenvalue of the adjacency matrix of G (see [65, Theorem 9] and [65, Corollary 3]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The novelties in the next result (Proposition 1) are as follows: (a) It forms a counterpart of a bound by Lov´asz [65, Theorem 9], providing a lower bound on θ(G) and an upper bound on θ(G) that are both expressed in terms of the second-largest eigenvalue of the adjacency matrix of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) It asserts that these two pairs of upper and lower bounds on θ(G) and θ(G) are tight for the family of strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This gives a simple closed-form expression of the Lov´asz θ-function of a strongly regular graph srg(n, d, λ, µ) (and the complement graph) as a function of its four parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (c) Further sufficient conditions for the tightness of these bounds are provided.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proposition 1: Let G be a d-regular graph of order n, which is a non-complete and non-empty graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, the following bounds hold for the Lov´asz θ-function of G and its complement G: (a) n − d + λ2(G) 1 + λ2(G) ≤ θ(G) ≤ − nλn(G) d − λn(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (24) Equality holds in the leftmost inequality of (24) if G is both vertex-transitive and edge-transitive, or if G is a strongly regular graph;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Equality holds in the rightmost inequality of (24) if G is edge-transitive, or if G is a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) 1 − d λn(G) ≤ θ(G) ≤ n� 1 + λ2(G)� n − d + λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (25) Equality holds in the leftmost inequality of (25) if G is both vertex-transitive and edge-transitive, or if G is a strongly regular graph;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Equality holds in the rightmost inequality of (25) if G is edge-transitive, or if G is a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof: See Section IV-A1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 2: In light of the sufficient conditions for each of the four inequalities in Proposition 1 to hold with equality, define the following subfamilies of regular graphs: Let G1 be the family of graphs G such that G is both vertex-transitive and edge-transitive;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 12 Let G2 be the family of regular and edge-transitive graphs;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G3 be the family of graphs G such that G is regular and edge-transitive;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G4 be the family of graphs that are both vertex-transitive and edge-transitive;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G5 be the family of the strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next show by explicit examples, obtained by using the SageMath software [85], that none of the families G1, G2 G3 and G4 is included in the family G5, and vice versa.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (a) The Cameron graph is a strongly regular graph srg(231, 30, 9, 3) (see [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='54]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its complement is vertex-transitive (hence, regular), but not edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that G5 ̸⊆ G3, so also G5 ̸⊆ G1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) The complement of the Cameron graph is a strongly regular graph srg(231, 200, 172, 180);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' it is vertex-transitive (hence, regular), but not edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that G5 ̸⊆ G2, so also G5 ̸⊆ G4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (c) The Foster graph is 3-regular on 90 vertices (see [19, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 305]), which is vertex-transitive and edge-transitive, but it is not strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that G4 ̸⊆ G5, so also G2 ̸⊆ G5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (d) The complement of the Foster graph is an 86-regular graph on 90 vertices, whose complement (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', the Foster graph) is vertex-transitive and edge-transitive, but it is not strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that G1 ̸⊆ G5, so also G3 ̸⊆ G5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The next result provides a closed-form expression of the Lov´asz θ-function for strongly regular graphs (and their strongly regular complements).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This result relies on Proposition 1 and the closed-form expressions of the distinct eigenvalues of a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 1: Let G be a strongly regular graph with parameters srg(n, d, λ, µ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, θ(G) = n (t + µ − λ) 2d + t + µ − λ, (26) and θ(G) = n θ(G) (27) = 1 + 2d t + µ − λ, (28) where t ≜ � (µ − λ)2 + 4(d − µ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (29) Furthermore, if 2d + (n − 1) (λ − µ) ̸= 0, then θ(G) and θ(G) are rational numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof: See Section IV-A2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 3: By (27), if G is a strongly regular graph on n vertices, then θ(G) θ(G) = n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This relation is also known to hold if the graph G is vertex-transitive [65, Theorem 8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It should be noted that not all the strongly regular graphs are necessarily vertex-transitive, so the observation here is not implied by [65, Theorem 8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' As a counter example for strongly regular graphs that are not vertex-transitive, consider the Chang graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These are three of the existing four non- isomorphic strongly regular graphs with parameters srg(28, 12, 6, 4) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='11] (the fourth such graph, denoted by T8, is the line graph of the complete graph on 8 vertices K8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The three Chang graphs are not vertex-transitive and also not edge-transitive (in contrast to T8 that is vertex-transitive and edge-transitive), as it can be verified by the SageMath software [85].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 13 Remark 4: The 5-cycle C5 is a strongly regular graph srg(5, 2, 0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its Lov´asz θ-function coincides with its Shannon capacity, being equal to √ 5 (see [65, Theorem 2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Although it is an irrational number, it is consistent with Corollary 1 since 2d+(n−1) (λ−µ) = 2·2+4(0−1) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Eigenvalue inequalities, strongly regular graphs, and Ramanujan graphs The present subsection relies on Proposition 1, with the following contributions: (a) Derivation of inequalities that relate the second-largest and smallest eigenvalues of a regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These inequalities hold with equality if and only if the graph is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) Derivation of bounds on parameters of Ramanujan graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (c) A more general result is presented for a sequence of regular graphs whose degrees scale sub-linearly with the orders of these graphs, and their orders tend to infinity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This subsection is composed of Corollaries 2–6, which all rely on Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It starts by providing eigenvalue inequalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 2: Let G be a d-regular graph of order n, which is non-complete and non-empty.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, λn(G) ≤ −d (n − d + λ2(G)) d + (n − 1) λ2(G), (30) or equivalently, λ2(G) ≥ −d (n − d + λn(G)) d + (n − 1) λn(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (31) Furthermore, (30) and (31) hold with equality if and only if G is a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof: See Section IV-B1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The next result introduces, in part, Nordhaus–Gaddum type inequalities for the second-largest and smallest eigenvalues of regular graphs, which are tight for all strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Regarding Nordhaus–Gaddum type inequalities, the interested reader is referred to [76], [77], and [80].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 3: Let G be a d-regular graph of order n, which is non-complete and non-empty, and let gℓ(G) ≜ λℓ(G) − d(n − d + λℓ(G)) d + (n − 1)λℓ(G), ∀ ℓ ∈ [n].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (32) The following holds: (a) gn(G) ≤ −1 ≤ g2(G), (33) and the two inequalities in (33) hold with equality if and only if G is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) If G is a strongly regular graph, then the number of distinct values in the sequence {gℓ(G)}n ℓ=1 is either 2 or 3, and it is equal to 2 if the multiplicities of the second-largest and smallest eigenvalues of G are identical in the subsequence (λ2(G), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , λn(G));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' it is otherwise equal to 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 14 (c) If G is self-complementary, then λ2(G) ≥ 1 2 �√n − 1 � , (34) λn(G) ≤ − 1 2 �√n + 1� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (35) (d) If G is self-complementary and strongly regular, then (34) and (35) hold with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof: See Section IV-B2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 1: Item (b) of Corollary 3 refers to the dichotomy in the number of distinct values in the sequence {gℓ(G)}n ℓ=1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This statement applies to all strongly regular graphs (either connected or disconnected).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We believe that the following example contributes to its clarity, in addition to its formal proof in Section IV-B2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a disjoint union of the three complete graphs K2, which gives a disconnected strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its complement is the complete 3-partite graph G = K2,2,2, so n = 6, and {λℓ(G)}6 ℓ=1 = (1, 1, 1, −1, −1, −1), (36) {λℓ(G)}6 ℓ=1 = (4, 0, 0, 0, −2, −2), (37) {gℓ(G)}6 ℓ=1 = (3, −1, −1, 1, −1, −1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (38) By (36), the multiplicities of λ2(G) and λn(G) in the spectrum of G are identical, but these multiplicities are distinct in the subsequence {λℓ(G)}6 ℓ=2 = (1, 1, −1, −1, −1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, the fact that the sequence {gℓ(G)}6 ℓ=1 gets three distinct values is indeed consistent with the claim in Item (b) of Corollary 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 2: Let G be the Hall-Janko graph, which is a strongly regular graph with parameters srg(100, 36, 14, 12) (see Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='32 of [19]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' As a numerical verification of Item (b) in Corollary 3 (and its proof in Section IV-B2), the sequence {gℓ(G)}100 ℓ=1 in (32) gets the three distinct values: n − d − 2 = 62 (at ℓ = 1), −1 (for 2 ≤ ℓ ≤ 37 or 65 ≤ ℓ ≤ 100), and 9 (for 38 ≤ ℓ ≤ 64).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The third value (9) is attained by the sequence {gℓ(G)} twenty-seven times.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By the proof of Item (b) in Corollary 3, the multiplicity of the third value (27) is equal to the absolute value of the difference between the multiplicities of the second-largest and smallest eigenvalues of the graph G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The spectrum of the graph G is given by 361636(−4)63 (this can be verified by (10) and (11)), and the above difference (in absolute value) is indeed equal to |63 − 36| = 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Next, let G be the 5-cycle graph C5, which is a strongly regular graph with parameters srg(5, 2, 0, 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The second-largest and smallest eigenvalues of G are equal to 1 2( √ 5 − 1) and − 1 2( √ 5 + 1), respectively, and their multiplicities coincide, being both equal to 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In light of Item (b) in Corollary 3, the sequence {gℓ(G)}5 ℓ=1 in (32) gets only two distinct values: n − d − 2 = 1 at ℓ = 1, and −1 for 2 ≤ ℓ ≤ 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 5: We discuss here an implication of the conditions for equalities in Proposition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a d-regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (30) holds with equality if and only if both inequalities in (24) hold with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Item (a) of Proposition 1, the leftmost inequality in (30) holds with equality if G is both vertex-transitive and edge-transitive, and the rightmost inequality in (30) holds with equality if G is edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Combining both sufficient conditions for the two inequalities in (30) to hold with equality, it follows that a sufficient condition for equality 15 in (30) is given by the requirement that G and G are both vertex-transitive and edge-transitive (recall that G is vertex-transitive if and only if G is so).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 2, inequality (30) holds with equality if and only if G is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By a comparison of the former (sufficient) condition with the latter (necessary and sufficient) condition for equality to hold in (30), it follows that if G and G are both vertex-transitive and edge-transitive, then G is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It should be noted, with gratitude to an anonymous reviewer, that a stronger result can be obtained by replacing the requirement of the vertex-transitivity with the weaker condition of regularity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Namely, if G is regular, and G and G are both edge-transitive, then G is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This can be shown as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (G is edge transitive) ⇒ (every edge in G is contained in the same number of triangles) ⇔ (every pair of adjacent vertices in G has the same number of common neighbors);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (G is edge transitive) ⇒ (for every edge {u, v} ∈ E(G), the same number of vertices are not adjacent in G to either u or v) ⇔ (every pair of non-adjacent vertices in G has the same number of common neighbors);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' G is regular (by assumption);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' and these observations correspond to the conditions in the definition of a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 3: The Schl¨afli graph G1 is 16-regular on 27 vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Both G1 and its complement are edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Remark 5, the graph G1 is indeed strongly regular;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' its parameters are given by srg(27, 16, 10, 8) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Consider, on the other hand, the Shrikhande graph G2 which is 6-regular on 16 vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is a strongly regular graph with parameters srg(16, 6, 2, 2) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The graph G2 is edge-transitive, but its complement is not edge-transitive (this was verified by the SageMath software [85]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In addition, the Cameron graph G3 is a strongly regular graph srg(231, 30, 9, 3) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='54].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It can be verified that it is edge-transitive, and that its complement is not edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that the family of regular graphs G with the property that G and its complement G are both edge-transitive is a strict subset of the family of strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 6: In continuation to Example 3, strongly-regular graphs G such that G and G are both edge-transitive include, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', the Hall-Janko, Hoffman-Singleton, Mesner, Petersen, Schl¨afli, Sims-Gewirtz, and Suzuki graphs (this has been verified by the SageMath software [85];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' for the introduction of these graphs, the reader is referred to [19]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It also includes the infinite families of Paley and Peisert graphs ([72], [81]), which are self-complementary and arc-transitive graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (All self-complementary and arc-transitive graphs are strongly regular).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 7: In connection to Remark 5, two related statements have been proved by Neumaier: (a) A connected, edge-transitive and strongly regular graph is vertex-transitive [74, Lemma 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) A vertex-transitive and edge-transitive graph containing a regular clique is strongly regular (see [75, Corollary 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='4]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (A clique C is called regular if every vertex not in C is adjacent to the same positive number of vertices in C).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Strongly regular graphs that are both vertex- and edge-transitive are studied in [71].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 4: Let {Gℓ}ℓ∈N be a sequence of graphs where Gℓ is dℓ-regular of order nℓ, and lim ℓ→∞nℓ = ∞, lim ℓ→∞ dℓ nℓ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (39) 16 Then, lim sup ℓ→∞ ω(Gℓ) ≤ a, (40) lim inf ℓ→∞ χ(Gℓ) nℓ ≥ 1 a, (41) with a ≜ 1 + lim sup ℓ→∞ ⌊λ2(Gℓ)⌋.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (42) Proof: See Section IV-B3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 5: Let {Gℓ}ℓ∈N be a sequence of connected, Ramanujan, and d-regular graphs where d ∈ N is fixed, Gℓ is a graph on nℓ vertices, and lim ℓ→∞nℓ = ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, lim sup ℓ→∞ ω(Gℓ) ≤ 1 + ⌊2 √ d − 1⌋, (43) lim inf ℓ→∞ θ(Gℓ) nℓ ≥ 1 1 + 2 √ d − 1, (44) lim inf ℓ→∞ χ(Gℓ) nℓ ≥ 1 1 + ⌊2 √ d − 1⌋.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (45) Proof: See Section IV-B4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In continuation to Corollary 5, the following result provides non-asymptotic bounds on some graph parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 6: Let G be a connected, Ramanujan, and d-regular graph with n vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, ω(G) ≤ � n � 1 + 2 √ d − 1 � n − d + 2 √ d − 1 � , (46) θ(G) ≥ n − d + 2 √ d − 1 1 + 2 √ d − 1 , (47) χ(G) ≥ �n − d + 2 √ d − 1 1 + 2 √ d − 1 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (48) Proof: See Section IV-B5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 8: Inequalities (40), (43), and (46) can be also obtained from [42, Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3], which states that for an arbitrary simple, finite, and undirected graph G on n vertices, whose maximal degree is given by ∆(G), the following bound on the clique number of G holds: ω(G) ≤ n � d + λ1(G) λ2(G) � dn − ∆(G)2 + λ1(G) λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (49) For a d-regular graph G on n vertices, we have λ1(G) = d = ∆(G), which then specializes (49) to (see [42, Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='4]) ω(G) ≤ n � 1 + λ2(G) � n − d + λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (50) Inequality (46) can be also obtained from (50), combined with the satisfiability of the inequality ��λ2(G) �� ≤ 2 √ d − 1 if G is a connected, d-regular, and Ramanujan graph, together with the fact that the right-hand side of (50) is monotonically increasing in the parameter λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 17 C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Bounds on eigenvalues of strong products of regular graphs A small second-largest eigenvalue of the adjacency matrix of a regular graph implies that the graph is a good expander (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Theorem 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2], and [6], [28], [50], [53], [59], [68]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Alon–Boppana bound in (8) is a lower bound on the second-largest eigenvalue of a regular graph (see Section II).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By the Alon–Boppana bound, for every sequence {Gk}∞ k=1 of d-regular graphs, with a fixed integer d ≥ 3 and orders tending to infinity (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', lim k→∞| V(Gk)| = ∞), the second-largest eigenvalues of their adjacency matrices satisfy lim inf k→∞ λ2(Gk) ≥ 2 √ d − 1, (51) (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [25, Theorem 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This lower bound is asymptotically tight, and its tightness can be strengthened beyond the second-largest eigenvalue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' More explicitly, by Serre’s theorem [61], for every fixed integer d ≥ 3 and (an arbitrarily small) ε > 0, there exists a positive constant c = c(ε, d) such that every d-regular graph with n vertices has at least cn eigenvalues that are larger than or equal to 2 √ d − 1 − ε.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In other words, Serre’s theorem states that a non-vanishing fraction of the n eigenvalues of every d-regular graph has the property of satisfying the Alon– Boppana lower bound within the desired accuracy (see a simplified proof in [25, Theorem 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3] or [21, Theorem 1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Analogous theorems, concerning the least eigenvalues of d-regular graphs, also hold under an additional hypothesis that the graphs do not have odd cycles below a certain length (see [21, Section 4]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It overall justifies the definition of Ramanujan graphs (see (7)) as d-regular graphs, with d ≥ 3, whose all non-trivial eigenvalues are (in absolute value) at most 2 √ d − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For a k-fold strong power of a d-regular graph, the degree is increased exponentially in k, being equal to dk = (1 + d)k − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is therefore of interest to obtain an alternative lower bound on the second-largest eigenvalue of strong products of regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its derivation is motivated by the significance of strong products, and in particular strong powers of a given graph: (1) The graph capacity in Shannon’s problem of zero-error communication [86] is given in (22), which is expressed in terms of the independence numbers of all k-fold strong powers of the graph (with k ∈ N);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (2) The Witsenhausen rate [94] in the zero-error source coding problem, with perfect side information at the receiver, is expressed in a dual form to (22), where the independence numbers of k-fold strong powers of a graph (with k ∈ N) are replaced by their chromatic numbers, and the supremum over k is replaced by an infimum (see [7, Section 3]);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (3) There exists a polynomial-time algorithm that finds the unique prime factorization of any connected graph under the operation of strong graph multiplication [31].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is demonstrated in this section that, under a certain condition, the suggested lower bound on the second-largest eigenvalue of the k-fold strong power of a regular graph offers a larger exponential growth rate in k, as compared to the Alon–Boppana bound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following proposition provides a lower bound on the second-largest eigenvalue, and an upper bound on the smallest eigenvalue of the adjacency matrices of strong products of regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Their derivation relies on Proposition 1, jointly with the factorization property in (17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Both bounds are expressed in terms of the Lov´asz θ-function of each factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This enables to obtain analytical bounds on the second-largest and smallest eigenvalues of a k-fold strong power 18 of a regular graph, with a low computational complexity that is not affected by k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This stays in contrast to the computational complexity of these eigenvalues, which significantly increases with k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proposition 2: Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be regular graphs such that, for all ℓ ∈ [k], the graph Gℓ is dℓ-regular of order nℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following bounds hold for their strong product: (a) Unless all Gℓ (with ℓ ∈ [k]) are complete graphs, then λ2(G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk) ≥ k� ℓ=1 nℓ − k� ℓ=1 (1 + dℓ) k� ℓ=1 θ(Gℓ) − 1 − 1 (52) ≥ k� ℓ=1 nℓ − k� ℓ=1 (1 + dℓ) k� ℓ=1 � − nℓ λmin(Gℓ) dℓ − λmin(Gℓ) � − 1 − 1, (53) and inequality (53) holds with equality if, for all ℓ ∈ [k], the regular graph Gℓ is either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) Unless all Gℓ (with ℓ ∈ [k]) are empty graphs, then λmin(G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk) ≤ − k� ℓ=1 (1 + dℓ) − 1 k� ℓ=1 � nℓ θ(Gℓ) � − 1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (54) If each regular graph Gℓ (with ℓ ∈ [k]) is either edge-transitive or strongly regular, then (54) can be expressed in an equivalent form as λmin(G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk) ≤ − k� ℓ=1 (1 + dℓ) − 1 k� ℓ=1 � 1 − dℓ λmin(Gℓ) � − 1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (55) Proof: See Section IV-C1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 9: As a sanity check, it would be in place to verify that the lower bound on the second-largest eigenvalue in the right-hand side of (52) is smaller than or equal to the largest eigenvalue of the strong product: λmax(G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk) = k � ℓ=1 (1 + dℓ) − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (56) First, equality (56) holds since G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk is d-regular with a value of d that is equal to the right-hand side of (56).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Straightforward algebra reveals that the required inequality we wish to assert readily follows from the inequalities θ(Gℓ) ≥ α(Gℓ) ≥ nℓ 1 + dℓ , ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (57) Indeed, the first inequality in (57) holds since the Lov´asz θ-function of a graph is an upper bound on its graph capacity (see [65, Theorem 1]), and (by definition) the graph capacity is 19 larger than or equal to the independence number of the graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The second inequality in (57) holds by Wei’s inequality [92], which provides a lower bound on the independence number or the clique number of a finite simple graph as a function of the degrees of its vertices (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [5, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 287] or [8, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 100] for a nice probabilistic proof of these inequalities;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' for some further such bounds, see [40]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For a simple graph G of order n, where vertex i ∈ [n] is of degree di, Wei’s bound states that α(G) ≥ n � i=1 1 1 + di , ω(G) ≥ n � i=1 1 n − di , (58) so the first inequality in (58) is specialized to the second inequality in (57) for a dℓ-regular graph Gℓ of order nℓ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It should be noted that the pair of inequalities in (57) also imply that, unless not all graphs {Gℓ} are empty, the upper bound on λmin(G1 ⊠.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠Gk) in the right-hand side of (54) is smaller than or equal to −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is a desired property of this upper bound since the smallest eigenvalue of a non-empty and finite regular graph is smaller than or equal to −1, while attaining this value if the graph is complete.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 7: Let G a d-regular graph of order n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, for all k ∈ N, the second-largest and smallest eigenvalues of the k-fold strong power of G satisfy the inequalities λ2(G⊠ k) ≥ nk − (1 + d)k θ(G)k − 1 − 1, (59) and λmin(G⊠ k) ≤ −(1 + d)k − 1 � n θ(G) �k − 1 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (60) Proof: See Section IV-C2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 4: By Corollary 7, the second-largest eigenvalue of the k-fold strong power of the 5-cycle (pentagon) graph C5 satisfies λ2(C⊠ k 5 ) ≥ 5k − 3k 5 k 2 − 1 − 1, k ∈ N, (61) which holds since θ(C5) = √ 5 (it is also the Shannon capacity of the pentagon [65, Theorem 2]), and C5 is 2-regular (d = 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The lower bound on λ2(C⊠ k 5 ) in the right-hand side of (61) scales asymptotically (for large k) like 5 k 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is next compared with the Alon–Boppana lower bound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The k-fold strong power of C5 is a dk-regular graph with dk = (1+d)k −1 = 3k −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Alon– Boppana lower bound in (8) is slightly smaller than 2√dk − 1, which scales asymptotically like 2 · 3 k 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This exemplifies an improvement in the exponential growth rate of the lower bound in (61), as compared to the Alon–Boppana lower bound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next compare numerically the exact values of λ2(C⊠ k 5 ), for 1 ≤ k ≤ 5, with their lower bounds in the right-hand side of (61) (the exact values were calculated by the SageMath software [85], and their numerical computation for k > 5 seem to be a difficult task).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The exact values for k ∈ [5] are equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='6180, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='8541, 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='5623, 42.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='6869, 130.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='0608, respectively, (with 4 digit decimal precision), in comparison to the lower bound in the right-hand side of (61) which is equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='6180, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='0000, 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='6264, 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3333 and 51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='4938, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 20 The next result shows that every non-complete regular graph G, whose Lov´asz θ-function is below a fixed value, has the property that almost all the strong powers of G are highly non-Ramanujan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The derivation of this result relies on inequality (59).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proposition 3: Let G be a non-empty and non-complete connected d-regular graph on n vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If θ(G) < n √ d + 1, (62) then there exists k0 ∈ N such that, for all k ≥ k0, the k-fold strong power G⊠ k is (highly) non-Ramanujan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An explicit closed-form expression for the value of k0 is given by k0 = max \uf8f1 \uf8f2 \uf8f33, \uf8ee \uf8ef\uf8ef\uf8ef log � 2 + (d + 1)− 3 2 � + log � n3 n3−(d+1)3 � log � n θ(G) √ d+1 � \uf8f9 \uf8fa\uf8fa\uf8fa \uf8fc \uf8fd \uf8fe .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (63) This holds, in particular, for all finite graphs that are self-complementary and vertex-transitive (they all satisfy the condition in (62) if n > 1), with a value of k0 which, respectively, is equal to 5, 4 or 3 if n = 5, n = 9 or n ≥ 13 with n ≡ 1 (mod 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof: See Section IV-C3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 10: A necessary and sufficient condition for the existence of a self-complementary graph on n vertices is that n ≡ 0 (mod 4) or n ≡ 1 (mod 4) (see [20, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 16–17]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A self- complementary and d-regular graph of order n satisfies d = n−1 2 , which implies that n needs to be odd.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since a vertex-transitive graph is regular, the option of n ≡ 0 (mod 4) is rejected for graphs of order n that are self-complementary and vertex-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This implies that the order n of such graphs must satisfy n ≡ 1 (mod 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For n = 1 and n = 5, there exist graphs of order n that are self-complementary and vertex-transitive;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' they are, respectively, given by K1 and C5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Graphs that are self-complementary and vertex-transitive, and approaches for their construction, received attention in the literature (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [60], [62], [63], [65], [81], [82], [83]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 11: Proposition 3 was inspired by Example 4, referring to the 5-cycle graph C5 that is self-complementary and vertex-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It can be numerically verified, with the SageMath software [85], that C5 and C⊠ 2 5 are Ramanujan graphs, and then the higher strong powers C⊠ 3 5 , C⊠ 4 5 , C⊠ 5 5 are non-Ramanujan graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proposition 3 shows that all strong powers C⊠ k 5 , with k ≥ 5, are non-Ramanujan graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An expression for k0, as it is given in (63), was derived in order to reduce the minimal analytical value of k0 for which the k-fold strong power of C5 is asserted to be non-Ramanujan for all k ≥ k0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It started with an initial value of k0 = 8 for n = 5, with a more simple initial expression for k0, and it was reduced to k0 = 5 with the closed-form expression in (63).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proposition 3, and the numerical experimentation as above, gives that C⊠ k 5 is a Ramanujan graph if and only if k = 1 or k = 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 5: Consider the connected Kneser graphs G = K(m, r) with m > 2r and m, r ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By their construction and [65, Theorem 13], n = �m r � , d = �m − r r � , θ(G) = �m − 1 r − 1 � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (64) The expression of θ(G) in (64) relies on the proof of [65, Theorem 13], which shows that α(G) = Θ(G) = θ(G) (it holds by combining the Erd¨os-Ko-Rado theorem for finite sets, which 21 serves to determine the independence number of the graph, together with the upper bound on θ(G) in [65, Theorem 9]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Straightforward algebra shows that the condition in (62) is not necessarily satisfied for the set of parameters in (64).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For example, by selecting m = 2r +1, the condition in (62) is satisfied if and only if 1 ≤ r ≤ 3 (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', it is satisfied by the Petersen graph, which corresponds to r = 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The condition in (62) is violated for r ≥ 4 so, for these values of the parameter r, the exponential growth rate of the Alon–Boppana lower bound on λ2(G⊠ k) is larger than the exponent of the suggested lower bound in the right-hand side of (59).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 12: Several methods for obtaining upper bounds on the smallest eigenvalue of a graph G rely on the identity λmin(G) = min x̸=0 xT A x xTx (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [27]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The dimensions of the adjacency matrix of a k-fold strong power of a graph G grow exponentially in k, so the computational complexity of such bounds is typically very high for strong graph powers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' However, for strong powers, the eigenvalue bounds in Corollary 7 are analytical, and their computational complexity is not affected by k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The reader is also referred to recent works which derive lower bounds on the smallest eigenvalue of a graph based on graph decompositions [23], and lower bounds on the smallest eigenvalue of regular graphs containing many copies of a smaller fixed subgraph [24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Such lower bounds on the smallest eigenvalue of a regular graph transform to upper bounds on the second-largest eigenvalue of regular graphs by using equality (5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Some upper bounds on the second-largest eigenvalue of connected graphs, with conditions for their attainability, are provided in [96].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The paper [49] surveys (less recent) bounds on the eigenvalues of simple graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lower bounds on the chromatic numbers of strong products The problem of relating the chromatic number of a graph to its eigenvalues dates back to Haemers [42, Section 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2], Hoffman [48] and Wilf [93], obtaining upper and lower bounds on the chromatic numbers of graphs in terms of their largest, second-largest and smallest eigenvalues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This section presents lower bounds on the chromatic number of a strong product of graphs (and its complement) in terms of the Lov´asz θ-functions (or smallest eigenvalues) of its factors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proposition 4: The following lower bounds on chromatic numbers of strong products hold: (a) Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be k simple graphs, ��V(Gℓ) �� = nℓ for ℓ ∈ [k], and G = G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, χ(G) ≥ � k � ℓ=1 nℓ θ(Gℓ) � , (65) χ(G) ≥ � k � ℓ=1 θ(Gℓ) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (66) (b) Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be regular graphs, where Gℓ is dℓ-regular of order nℓ for all ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, χ(G) ≥ � k � ℓ=1 nℓ θ(Gℓ) � (67) ≥ � k � ℓ=1 � 1 − dℓ λmin(Gℓ) �� , (68) and inequality (68) holds with equality if each regular graph Gℓ is either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 22 (c) If, for all ℓ ∈ [k], Gℓ is dℓ-regular, and it is either edge-transitive or strongly regular, then k � ℓ=1 � 1 − dℓ λmin(Gℓ) � ≥ 1 − d(G) λmin(G), (69) where d(G) = k � ℓ=1 (1 + dℓ) − 1 (70) is the valency of the regular graph G = G1⊠.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='⊠Gk, and λmin(G) is its smallest eigenvalue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (d) Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be regular graphs, where Gℓ is dℓ-regular of order nℓ for all ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (1) If, for all ℓ ∈ [k], the graph Gℓ is either vertex-transitive or strongly regular, then the lower bound on χ(G) in the right-hand side of (65) is larger than or equal to the lower bound k� ℓ=1 ω(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (2) If, for all ℓ ∈ [k], the graph Gℓ is either (i) both vertex-transitive and edge-transitive, or (ii) strongly regular, then the lower bound on χ(G) in the right-hand side of (68) is larger than or equal to the lower bound k� ℓ=1 ω(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (e) Let, for all ℓ ∈ [k], the graph Gℓ be dℓ-regular on nℓ vertices, and suppose that it is either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, χ(G) ≥ � k � ℓ=1 � − nℓ λmin(Gℓ) dℓ − λmin(Gℓ) �� .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (71) (f) Let, for all ℓ ∈ [k], Gℓ be a self-complementary graph on nℓ vertices that is either vertex- transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let n ≜ k� ℓ=1 nℓ be the order of G = G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, χ(G) ≥ �√n � , (72) χ(G) ≥ �√n � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (73) Proof: See Section IV-D1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 13: The following inequality is proved in [4, Theorem 11]: χ(G1 ⊠ G2) ≥ θ(G1) θ(G2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (74) (Analogous inequalities to (74) were derived by Hales [45], and by McEliece and Posner [70];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' see [4, Theorem 10]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (65) can be obtained by combining (17), (18) and (74).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This can be done by first replacing G1 and G2 in (74) with G′ 1 ≜ G1 and G′ 2 ≜ G2 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk, respectively, then relying on (18) to get the inequalities θ(G′ 1) ≥ n1 θ(G1) and θ(G′ 2) ≥ n2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='nk θ(G′ 2) , and finally relying on (17) to get the equality θ(G′ 2) = θ(G2) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' θ(Gk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Our two simple proofs of (65) are, however, easier than this one, and they do not require to rely on (74).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 14: Inequality (74) differs from (66), although the lower bounds in the right-hand sides are similar for k = 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The difference between (66) and (74) is that the left-hand side of (66) refers to the chromatic number of G1 ⊠ G2, whereas the left-hand side of (74) refers to 23 G1 ⊠ G2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The result in (66) appears in Proposition 4 for completeness since it is an immediate consequence of the sandwich theorem in (15), and the identity in (17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 15: Under the assumptions of Item (c) in Proposition 4, the lower bound on χ(G) in the right-hand side of (68) is larger than or equal to Hoffman’s lower bound (see (69)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This holds in addition to the high complexity in computing λmin(G) in the right-hand side of (69), even for relatively small values of k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hoffman’s bound was originally proved for regular graphs [48], and it was later extended by Haemers to general simple and finite graphs (see, [18, Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3], [42, Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3], [89, Corollary 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='10 and Theorem 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='11]), and to hypergraphs [32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A recent perspective on Hoffman’s bound appears in [44].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 16: We refer to Item (f) of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If G is self-complementary of order n, then χ(G) ≥ �√n � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (75) Indeed, for every graph G, χ(G) χ(G)≥ θ(G) θ(G) (76) ≥ n, (77) where inequality (76) holds since, by (15) and (16), χ(G) ≥ θ(G) and χ(G) ≥ θ(G), and inequality (77) holds by [65, Corollary 2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The chromatic number is invariant under isomorphism, so χ(G) = χ(G) if G is self-complementary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This gives (75) from (77).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It should be noted, however, that the result in Item (f) (see (72) and (73)) is not implied by (75).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This is because a strong product of self-complementary graphs, where each factor is also either vertex-transitive or strongly regular, does not necessarily give a self-complementary graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' As a counter example, let G1 = G2 = C5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The pentagon C5 is a self-complementary, vertex-transitive and strongly regular graph, whereas G = C5 ⊠ C5 is not a self-complementary graph by [62, Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' according to that theorem, G would have been self-complementary if the strong product of G1 and G2 had been replaced by their lexicographic product (these last two observations were also verified by the SageMath software).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The next result specializes Item (a) of Proposition 4 (see (65)) to strong products of strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This result is obtained by relying on Corollary 1 that provides a closed-form expression for the Lov´asz θ-function of each factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Corollary 8: Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be strongly regular graphs with parameters srg(nℓ, dℓ, λℓ, µℓ) for ℓ ∈ [k] (they need not be distinct).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, the chromatic number of their strong product satisfies � k � ℓ=1 � 1 + 2dℓ tℓ + µℓ − λℓ �� ≤ χ(G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk) ≤ k � ℓ=1 χ(Gk), (78) where {tℓ}k ℓ=1 in the leftmost term of (78) is given by tℓ ≜ � (λℓ − µℓ)2 + 4(dℓ − µℓ), ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (79) The leftmost term in (78) is also larger than or equal to the product of the clique numbers of the factors {Gℓ}k ℓ=1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof: See Section IV-D2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The next two examples present strong products of strongly regular graphs, whose chromatic numbers are exactly determined by Corollary 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 24 Example 6: Let G1, G2, G3 be the Schl¨afli, Shrikhande, and Hall-Janko graphs, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These are strongly regular graphs whose parameters are srg(27, 16, 10, 8), srg(16, 6, 2, 2), and srg(100, 36, 14, 12), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Their chromatic numbers are equal to χ(G1) = 9, χ(G2) = 4, and χ(G3) = 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Consider the chromatic number of the strong product of arbitrary nonnegative powers of G1, G2 and G3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It can be verified that, for all such strong products, the upper and lower bounds in Corollary 8 coincide, so for all integers k1, k2, k3 ≥ 0, χ(G⊠ k1 1 ⊠ G⊠ k2 2 ⊠ G⊠ k3 3 ) = 9k1 4k2 10k3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (80) For comparison, the lower bound that is given by the product of the clique numbers of each factor is equal to 6k13k24k3 (since ω(G1) = 6, ω(G2) = 3, and ω(G3) = 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This shows that it is significantly looser than the tight lower bound in the right-hand side of (80).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 7: Consider the three non-isomorphic Chang graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These are strongly regular graphs with the same set of parameters srg(28, 12, 6, 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The clique number of one of these graphs is equal to 5, and the clique numbers of the other two graphs are equal to 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let us denote these graphs by G1, G2 and G3, such that ω(G1) = 5, ω(G2) = 6, and ω(G3) = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The chromatic numbers of all these three graphs are similar, and they are equal to 7, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', χ(G1) = χ(G2) = χ(G3) = 7 (the clique and chromatic numbers of the three Chang graphs are easy to verify with the SageMath software [85]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let k1, k2 and k3 be arbitrary nonnegative integers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 8, χ(G⊠ k1 1 ⊠ G⊠ k2 2 ⊠ G⊠ k3 3 ) = 7k1+k2+k3 (81) due to the coincidence of the upper and lower bounds in (78).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For comparison, the lower bound on the chromatic number in the left-hand side of (81), which is given by the product of the clique numbers of each factor, is equal to 5k16k2+k3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The next example presents a power product of a vertex and edge-transitive regular graph, whose chromatic number is exactly determined by Item (b) of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 8: The present example provides the exact value of χ(G⊠ k) where G is the Perkel graph, and k ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Perkel graph is 6-regular on 57 vertices, and it is both vertex-transitive and edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The clique and chromatic numbers of G are equal to ω(G) = 2 and χ(G) = 3, respectively, and the smallest eigenvalue of (the adjacency matrix of) G is equal to λmin(G) = −3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Item (b) of Proposition 4, for all k ∈ N, χ(G⊠ k) ≥ � 1 − d λmin(G) �k = 3k, (82) which can be compared here to the simple upper and lower bounds on χ(G⊠ k) whose values are given by χ(G)k = 3k and ω(G)k = 2k, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The coincidence of the improved lower bound in (82) and the upper bound gives that, for all k ∈ N, χ(G⊠ k) = 3k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (83) The next two examples illustrate numerically Part 2 of Item (d) in Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 9: Let G be the Suzuki graph, which is a strongly regular graph with parameters srg(1782, 416, 100, 96) (see [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='83]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The lower bound in Corollary 8 (see the leftmost term in (78)) gives that χ(G⊠ k) ≥ 27k for all k ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For comparison, since ω(G) = 6, the lower bound that is based on the clique number of G gives χ(G⊠ k) ≥ 6k for all k ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The exact value of χ(G) is not available for the Suzuki graph, so the upper bound in the rightmost 25 term of (78) is unknown.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The improvement in the exponential lower bound on the chromatic number of the strong powers G⊠ k is, however, significant since it is increased from 6k to 27k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 10: Let G be the Gosset graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We apply here Item (b) of Proposition 4 in order to obtain an improved lower bound on χ(G⊠ k) for all k ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The graph G is 27-regular on 56 vertices (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', d = 27 and n = 56);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' it is both vertex-transitive and edge-transitive, and it is also not strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The clique and chromatic numbers of G are equal to ω(G) = 7 and χ(G) = 14, respectively, and the smallest eigenvalue of (the adjacency matrix of) G is equal to λmin(G) = −3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Item (b) of Proposition 4 (note that the edge-transitivity of G implies that (68) holds with equality), it follows that for all k ∈ N, χ(G⊠ k) ≥ � 1 − d λmin(G) �k (84) = 10k, (85) which can be compared here to the simple upper and lower bounds on χ(G⊠ k) whose values are given by χ(G)k = 14k and ω(G)k = 7k, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' All the factors of the strong products in Examples 6–10 are either strongly regular or otherwise, they are both vertex-transitive and edge-transitive graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Consequently, by Item (b) of Propo- sition 4, the two lower bounds in the right-hand sides of (67) and (68) coincide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Furthermore, by Part 2 of Item (d) in Proposition 4, they also offer an improvement over the simple lower bound that is equal to the product of the clique numbers of each factor of the strong product.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The next example shows that such an improvement does not necessarily take place if the factors of the strong product of the regular graphs are not vertex-transitive and edge-transitive, while also not being strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 11: Let G be the Frucht graph, which is 3-regular on 12 vertices (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', d = 3 and n = 12).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This graph is not strongly regular, and also not vertex-transitive or edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The clique and chromatic numbers of this graph are both equal to 3 (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', ω(G) = χ(G) = 3), so χ(G⊠ k) = 3k, k ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (86) The smallest eigenvalue of the adjacency matrix of G is equal to λmin(G) = −2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='33866.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By (68), for all k ∈ N, χ(G⊠ k) ≥ � 1 − d λmin(G) �k (87) = 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='28278k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (88) This shows that the lower bound on the chromatic number in the right-hand side of (68) may be looser than the simple lower bound that is equal to the product of the clique numbers of each factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This may happen if some of the factors, which are regular graphs, are not either strongly regular, or both vertex-transitive and edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Shannon capacity of strongly regular graphs The Lov´asz θ-function of a strongly regular graph is expressed in closed-form in Corollary 1, being also an upper bound on the Shannon capacity of such a graph [65, Theorem 1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The independence number of a graph is, on the other hand, a lower bound on the Shannon capacity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 26 The Shannon capacity of such a graph is therefore determined if these upper and lower bounds coincide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following examples show such cases of coincidence for strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The first one reproduces (in a different way) the Shannon capacity of the Petersen graph, a result dating back to Lov´asz [65].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The rest of the examples provide new results that determine the Shannon capacity of some strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 12: Let G be the Petersen graph, whose Shannon capacity is equal to 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It is a special case of [65, Theorem 13], which gives the Shannon capacity of Kneser graphs Θ � K(m, r) � = �m − 1 r − 1 � , m ≥ 2r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (89) Then, the capacity of the Petersen graph is obtained in [65, Corollary 6] by viewing it as an isomorphic graph to the Kneser graph K(5, 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' As an alternative way to determine its Shannon capacity, the Petersen graph is a strongly regular graph with parameters srg(10, 3, 0, 1) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, it follows that θ(G) = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This coincides with the independence number α(G) = 4, so Θ(G) = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 13: Let G be the Shrikhande graph, which is strongly regular with the parameters srg(16, 6, 2, 2) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, it can be verified that θ(G) = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its chromatic number is χ(G) = 4, and its independence number is α(G) = 4 (so α(G) χ(G) = n, which implies that the n = 16 vertices in G are partitioned into four color classes, where each color class is a largest independent set in G of size 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, the graph capacity is equal to Θ(G) = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The capacity of every graph G of order n that is self-complementary and vertex-transitive is Θ(G) = √n (see [65, Theorem 12]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It should be noted, however, that although the capacity of the Shrikhande graph is a case where Θ(G) = √n (with n = 16), it does not follow from [65, Theorem 12] since the Shrikhande graph is vertex-transitive, but it is not self-complementary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 14: Consider the Hall-Janko graph G, which is strongly regular with parameters srg(100, 36, 14, 12) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='32].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, θ(G) = 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The independence number of this graph is α(G) = 10 (this was obtained by the SageMath software [85]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, the Shannon capacity of the Hall-Janko graph is Θ(G) = 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This graph is vertex-transitive but it is not self-complementary so, similarly to Example 13, this result does not follow from the capacity result in [65, Theorem 12] for finite graphs that are vertex-transitive and self-complementary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 15: The Hoffman-Singleton graph G is a strongly regular graph with parameters srg(50, 7, 0, 1) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, it follows that θ(G) = 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The independence number of this graph is α(G) = 15, so the Shannon capacity of this graph is Θ(G) = 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 16: The Schl¨afli graph G is strongly regular with parameters srg(27, 16, 10, 8) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, it follows that θ(G) = 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The independence number of this graph is α(G) = 3, so the Shannon capacity of this graph is Θ(G) = 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 17: The Sims-Gewirtz graph G is strongly regular with parameters srg(56, 10, 0, 2) (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' the Gewirtz graph) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, θ(G) = 16, and also α(G) = 16 (it can be verified by the SageMath software [85]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Shannon capacity of the Sims-Gewirtz graph is therefore equal to Θ(G) = 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 18: The M22 graph (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Mesner graph) G is strongly regular with parameters srg(77, 16, 0, 4) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Its independence number is α(G) = 21, and by Corollary 1 27 also θ(G) = 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Consequently, the Shannon capacity of the M22 graph is Θ(G) = 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 19: The Cameron graph G is strongly regular with parameters srg(231, 30, 9, 3) (see [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='54]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, θ(G) = 21, which coincides with the independence number of G, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', α(G) = 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Shannon capacity of the Cameron graph is, hence, equal to Θ(G) = 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 20: Consider the three non-isomorphic Chang graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These are strongly regular graphs with the same set of parameters srg(28, 12, 6, 4) [19, Section 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Corollary 1, the Lov´asz θ-function of these graphs is equal to 4, which coincides with their independence number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The Shannon capacity of each of these three graphs is therefore equal to 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Remark 17: Examples 12–20 provide several strongly regular graphs for which their Lov´asz θ-function coincides with their independence number, thus determining the Shannon capacity of these graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This, however, is not the case in general for strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For example, in continuation to Example 16, the Lov´asz θ-function of the complement of the Schl¨afli graph G (it is a strongly regular graph srg(27, 10, 1, 5)) is equal to θ(G) = 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Indeed, θ(G) θ(G) = 27 is the order of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It was proved, however, by the rank-bound of Haemers [43] that the capacity of the complement of the Schl¨afli graph is at most 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It can be also verified that the independence number of G (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', the clique number of the Schl¨afli graph G) is equal to 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This overall gives that 6 = ω(G) ≤ Θ(G) ≤ 7 < 9 = θ(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We end this section by studying some capacity results of affine polar graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Example 21: Affine polar graphs (see, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', [19, Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3]) are constructed by considering a vector space V of dimension d over a finite field Fq with q elements, equipped with a non- degenerate quadratic form Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The vertices in the graph are represented by the vectors in V, and any two vectors u and v (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', a pair of vertices in the graph) are adjacent if Q(u − v) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The resulting graph is denoted by VO+(d, q), VO−(d, q), and VO(d, q) when the quadratic form Q is hyperbolic, elliptic or parabolic, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In the first two cases, d is even, and in the third case, d is odd;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' in all cases, q is a positive integral power of a prime number (as it is the cardinality of the finite field Fq).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Under these conditions on d and q, the graphs VO+(d, q) and VO−(d, q) are strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let d = 2e with e ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The parameters of G+ = VO+(2e, q), which is a strongly regular graph srg(n+, d+, λ+, µ+), are given by (see [19, Section 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='3]) n+ = q2e, (90) d+ = (qe−1 + 1)(qe − 1), (91) λ+ = q(qe−2 + 1)(qe−1 − 1) + q − 2, (92) µ+ = qe−1(qe−1 + 1), (93) λ2(G+) = qe − qe−1 − 1, (94) λn+(G+) = −qe−1 − 1, (95) and G− = VO−(2e, q), a strongly regular graph srg(n−, d−, λ−, µ−), has the parameters n− = q2e, (96) d− = (qe−1 − 1)(qe + 1), (97) λ− = q(qe−2 − 1)(qe−1 + 1) + q − 2, (98) µ− = qe−1(qe−1 − 1), (99) λ2(G−) = qe−1 − 1, (100) λn−(G−) = −qe + qe−1 − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (101) 28 By Corollary 1, combined with the parameters (and eigenvalues) in (90)–(101), it follows that θ(G+) = qe = √ n+, (102) θ(G+) = qe, (103) θ(G−) = q(qe − qe−1 + 1), (104) θ(G−) = q2e−1 qe − qe−1 + 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (105) Consequently, it can be verified that for some of these affine polar graphs (with the free parameters q and e as above), their Lov´asz θ-function and independence number coincide.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (The numerical computations of the independence numbers of these graphs were performed by the SageMath software [85].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=') This gives, for example, that Θ � VO+(4, 2) � = 4, Θ � VO+(6, 2) � = 8, (106) Θ� VO+(4, 3)� = 9, Θ� VO+(6, 3)� = 27, (107) as the exact values of the Shannon capacities of these strongly regular graphs, whose parameters are, respectively, srg(16, 9, 4, 6), srg(64, 35, 18, 20), srg(81, 32, 13, 12), and srg(729, 260, 97, 90).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' PROOFS A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proofs for Section III-A 1) Proof of Proposition 1: Let G be a d-regular graph with n vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The rightmost inequality in (24) is provided in [65, Theorem 9], together with a sufficient condition that it holds with equality if G is edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Another sufficient condition for that inequality to hold with equality is obtained later in this proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We first prove the leftmost inequality in (24), and also obtain sufficient conditions that it holds with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By (18) (see [65, Corollary 2]), θ(G) ≥ n θ(G) (108) and, by [65, Theorem 8], equality holds in (108) if G is vertex-transitive (or, equivalently, if G is vertex-transitive).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since G is an (n − d − 1)-regular graph of order n, an application of (19) (see [65, Theorem 9]) to G gives θ(G) ≤ − nλn(G) (n − d − 1) − λn(G), (109) with equality in (109) if G is edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Combining (108) and (109) gives θ(G) ≥ 1 − n − d − 1 λn(G) , (110) with equality in (110) if G is vertex-transitive and edge-transitive (recall that a vertex-transitive graph is regular).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By the regularity of G, we have from (5), λn(G) = −1 − λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (111) Combining (110) and (111) gives the leftmost inequality in (24), together with the conclusion that it holds with equality if G is both vertex-transitive and edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Next, (25) is obtained from (24) by replacing G with G, and relying on the equality in (111).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A sufficient condition 29 that the leftmost inequality in (25) holds with equality is therefore the same condition that the leftmost inequality in (24) holds with equality, while replacing G with G;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' this means that the leftmost inequality in (25) holds with equality if G is vertex-transitive and edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Likewise, the rightmost inequality in (25) holds with equality under the same condition that the rightmost inequality in (24) holds with equality, while the graph G in (24) is replaced by its complement G in (25).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This means that the rightmost inequality in (24) holds with equality if G is edge-transitive (recall that, unlike vertex-transitivity that is a property of G if and only if it is a property of its complement G, it is not the case for edge-transitivity).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We finally prove that if G is a strongly regular graph, then the four inequalities in (24) and (25) hold with equalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a strongly regular graph srg(n, d, λ, µ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, the largest, second-largest, and smallest eigenvalues of G are given by λ1(G) = d, (112) λ2(G) = r ≜ 1 2 � λ − µ + � (λ − µ)2 + 4(d − µ) � , (113) λn(G) = s ≜ 1 2 � λ − µ − � (λ − µ)2 + 4(d − µ) � , (114) where p1,2 in (10) are replaced here with r and s in (113) and (114), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (Recall that if G is a disconnected strongly regular graph, then µ = 0 and λ = d−1 by (9), which then indeed gives that λ2(G) = d = λ1(G) since the d-regular graph is disconnected).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The substitution of (113) and (114) into (24) and (25) gives n − d + r 1 + r ≤ θ(G) ≤ − ns d − s, (115) and 1 − d s ≤ θ(G) ≤ n(1 + r) n − d + r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (116) We show as follows that the rightmost and leftmost terms in (115) coincide, turning the two inequalities in (115) into equalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The coincidence of the rightmost and leftmost terms in (115) is equivalent to proving the equality (n − d + r) (d − s) + ns (1 + r) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (117) Indeed, we get (n − d + r) (d − s) + ns (1 + r) = nd − d2 + d (r + s) + (n − 1)rs (118) = nd − d2 + d (λ − µ) + (n − 1) (µ − d) (119) = −d (d − λ − 1) + µ (n − d − 1) (120) = 0, (121) where (118) and (120) hold by straightforward algebra;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (119) holds by (113) and (114);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (121) holds by the identity in (9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The same conclusion also holds for the rightmost and leftmost terms in (116), which can be shown to coincide if G is a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This turns the two inequalities in (116) into equalities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It holds in a similar way to the previous part since the complement of a strongly regular graph is also strongly regular, and the second-largest and smallest eigenvalues of G in 30 (115) (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', r, s respectively) are replaced in (115) by the second-largest and smallest eigenvalues of G (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', −1 − s and −1 − r respectively).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This proves that each of the four inequalities in (24) and (25) holds with equality if G is a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2) Proof of Corollary 1: Let G be a strongly regular graph srg(n, d, λ, µ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Proposition 1, θ(G) = n − d + r 1 + r (122) = − ns d − s, (123) and θ(G) = 1 − d s (124) = n(1 + r) n − d + r, (125) where r and s are given in (113) and (114), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Their substitution into the right-hand sides of (123) and (124) gives (26) and (28), respectively, with the auxiliary parameter t as defined in (29).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Equality (27) is trivial by (122)–(125).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This verifies (26)–(28) for all strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since the multiplicities of the distinct eigenvalues of a strongly regular graph G are integers, it follows from (11) that if 2d + (n − 1)(λ − µ) ̸= 0, then t in (29) should be an integer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This implies that, under the latter condition, θ(G) and θ(G) are necessarily rational numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proofs for Section III-B 1) Proof of Corollary 2: Let G be a non-complete, d-regular graph of order n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Straightforward algebra gives (30) by comparing the rightmost and leftmost terms in (24).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequalities (30) and (31) can be verified to be equivalent by relying on the inequality d+(n−1)λn(G) < 0 (this holds since λn(G) ≤ −1, and d < n−1 for a non-complete, d-regular graph G with n vertices).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Equality holds in (30) if and only if the rightmost and leftmost terms in (24) are equal to θ(G) (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', they are both equal to the middle term in (24)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' According to Item (a) of Proposition 1, equality holds in (30) and (31) if G is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next show that the latter sufficient condition is also necessary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' To that end, we provide a second proof that relies on linear algebra.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let A = A(G) and A = A(G) be the adjacency matrices of G and G, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let Jn and In denote the n-times-n all-ones and identity matrices, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, by (2), A = Jn − In − A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let jn be the all-ones n-length column vector, so jn jT n = Jn, and A + c1 In +c2 Jn = (c1 − 1) In +(c2 + 1) Jn − A, (126) for all c1, c2 ∈ R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since G is a d-regular graph, the largest eigenvalue of A is equal to d with jn as an eigenvector.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The vector jn is also an eigenvector of Jn and In with eigenvalues n and 1, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By (126), and since G is d-regular, (A + c1 In +c2 Jn) jn = � c1 − 1 + n(c2 + 1) − d � jn, (127) which means that jn is an eigenvector of the matrix A+c1 In +c2 Jn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The other n−1 eigenvectors of the symmetric matrix A + c1 In +c2 Jn can be made orthogonal to the eigenvector jn (since eigenvectors of a symmetric matrix, which correspond to distinct eigenvalues, are orthogonal 31 with respect to the standard inner product in Rn;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' furthermore, eigenvectors that correspond to the same eigenvalue can be made orthogonal by the Gram-Schmidt procedure).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let v be such an eigenvector, different from jn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then Jn v = jn(jT nv) = 0, so by (126) (A + c1 In +c2 Jn) v = (c1 − 1)v − A v, (128) which means that v is also an eigenvector of the adjacency matrix A (since, by assumption, v is an eigenvector of the matrix A + c1 In +c2 Jn in the left-hand side of (128)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The equality A v = λv holds, and λ ∈ {λ2(G), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , λn(G)} with λ2(G) being the largest among them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This gives that λ ≤ λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The symmetric matrix A + c1 In +c2 Jn is positive semi-definite if and only if all its eigenvalues are nonnegative, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (see (127) and (128)), c1 − 1 + n(c2 + 1) − d ≥ 0, (129) c1 − 1 − λ2(G) ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (130) Setting the two conditions in (129) and (130) to be satisfied with equalities gives c1 = 1 + λ2(G), c2 = −n − d + λ2(G) n , (131) which implies that (see the left-hand side of (126)) A + � 1 + λ2(G) � In − �n − d + λ2(G) n � Jn ⪰ 0 (132) is a positive semi-definite matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Clearly, also A −λn(G) In ⪰ 0 (133) is positive semi-definite.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The trace of a product of two n-times-n positive semi-definite matrices is nonnegative, so it follows from (132) and (133) that tr �� A + � 1 + λ2(G) � In − �n − d + λ2(G) n � Jn � � A −λn(G) In � � ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (134) For an adjacency matrix A of a d-regular graph G, tr(A A) = tr(A) = tr(A) = 0, (135) tr(Jn A) = nd, (136) and tr(In) = tr(Jn) = n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (137) From (134)–(137), expanding the left-hand side in (134) gives the inequality −d � n − d + λ2(G) � − λn(G) � d + (n − 1)λ2(G) � ≥ 0, (138) which proves (30) (we have d + (n − 1)λ2(G) > 0, by the assumption that the d-regular graph G is non-complete and non-empty, so λ2(G) ≥ 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This alternative proof is next used to find the necessary and sufficient condition for the satisfiability of (30) with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' An equality in (30) holds if and only if (134) holds with equality (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', the trace of the product of the two positive semi-definite matrices in the left-hand sides of (132) and (133) is equal to zero).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This holds if and only if the column spaces of the two matrices in the left-hand sides of (132) and (133) are 32 orthogonal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [Clarification: if B, C ⪰ 0 are positive semi-definite matrices, then B = SST and C = QQT for some matrices S and Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, under the assumption that tr(BC) = 0, 0 = tr(BC) = tr(SSTQQT) = tr(STQQTS) = tr(STQ (STQ)T) = n � i=1 n � j=1 � STQ �2 i,j (139) implies that STQ = 0, so also BC = S(STQ)QT = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since B and C are symmetric matrices with BC = 0, it means that the column spaces of B and C are orthogonal].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These column spaces, however, can be orthogonal only if A has no eigenvalues other than d, λ2(G) and λn(G);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' otherwise, there is a joint eigenvector of A in the two column spaces of the matrices in the left-hand sides of (132) and (133), so these two column spaces cannot be orthogonal in the latter case.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [Clarification: suppose that A has an eigenvector v that corresponds to an eigenvalue λ∗ ̸∈ {d, λ2(G), λn(G)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, by (128) and the value of c1 in (131), the right-hand side of (128) is equal to (λ2(G) − λ∗)v, and (A −λn(G) In)v = (λ∗ − λn(G))v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Both coefficients of v in these two expressions are nonzero, so v belongs to the two column spaces of the two matrices in the left-hand sides of (132) and (133).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' These column spaces are therefore not orthogonal under the assumption of the existence of a distinct eigenvalue λ∗ of A, as above].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next distinguish between two cases in regard to the connectivity of the graph G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (a) If the d-regular graph G is connected, then λ1(G) = d > λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By assumption, G is also non-complete and non-empty graph, so λ2(G) > λn(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The connected regular graph G thus has exactly three distinct eigenvalues, so it is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) If the d-regular graph G is disconnected, then λ1(G) = d = λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If, by assumption, inequality (30) holds with equality, then λn(G) = −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This means that G is a disjoint union of equal-sized complete graphs Kd+1, so it is an imprimitive strongly regular graph (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', there are no common neighbors of any pair of non-adjacent vertices in G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We therefore conclude that, for a non-complete and non-empty d-regular graph on n vertices, the condition that G is strongly regular is also necessary (and not only sufficient, as it is shown in the first proof) for the inequality in (30) to hold with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Equivalently, G being strongly regular is a necessary condition for the inequality in (31) to hold with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This completes the proof of the necessity and sufficiency of the condition on the strong regularity of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2) Proof of Corollary 3: Let G be a d-regular graph of order n, which is non-complete.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof of Item (a): By the definition of g2(G) and gn(G) in (32), and in light of the equalities λn(G) = −1−λ2(G) and λ2(G) = −1−λn(G), it can be readily verified that the rightmost and leftmost inequalities in (33) are equivalent to (30) and (31), respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Furthermore, in light of Corollary 2, each inequality in (33) holds with equality if and only if G is strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof of Item (b): Let G be a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next prove the claim about the dichotomy in the number of distinct values in the sequence {gℓ(G)}n ℓ=1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (1) For a d-regular graph on n vertices, λ1(G) = d, and λ1(G) = n − d − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Substituting these eigenvalues into (32) gives that g1(G) = n − d − 2 ≥ 0 (G is non-complete, so d ≤ n − 2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (2) The graph G and its complement G are both strongly regular, so each one of them has at most three distinct eigenvalues.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (3) For a strongly regular graph, by Item (a), g2(G) = −1 = gn(G), and gℓ(G) = −1 if either (i) λℓ(G) = λ2(G) and λℓ(G) = λ2(G), or (ii) λℓ(G) = λn(G) and λℓ(G) = λn(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 33 (4) By assumption, G is a strongly regular graph, which implies that so is G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Due to their regularity, λ1(G) ≥ λ2(G), and λ1(G) ≥ λ2(G) with equalities, respectively, if and only if G or G are disconnected graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The sequence {gℓ(G)}n ℓ=1 gets an additional (third) distinct value if and only if the multiplicities of the smallest and the second-largest eigenvalues of G in the subsequence (λ2(G), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , λn(G)) are distinct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Indeed, in the latter case, only one of the following two options is possible: (iii) λℓ(G) = λ2(G) and λℓ(G) = λn(G), or (iv) λℓ(G) = λn(G) and λℓ(G) = λ2(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This holds since, by (4), the multiplicity of the second-largest eigenvalue of G is equal to the multiplicity of the smallest eigenvalue of G, and similarly, the multiplicity of the smallest eigenvalue of G is equal to the multiplicity of the second-largest eigenvalue of G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It therefore follows that the third distinct value (as above) is attained by the sequence {gℓ}n ℓ=1 a number of times that is equal to the absolute value of the difference between the multiplicities of the second-largest and the smallest eigenvalues of G in the subsequence (λ2(G), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , λn(G)) (provided that the latter two multiplicities are distinct).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof of Item (c): Let G be self-complementary and d-regular on n vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, d = 1 2(n − 1), λ2(G) = λ2(G), λn(G) = λn(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (140) Combining the rightmost inequality in (33) and the equalities in (140) readily gives 2λ2(G) 2 − 1 2(n + 1) 1 + 2λ2(G) ≥ −1, (141) Since G is non-complete and non-empty, we get λ2(G) > 0, which then gives from (141) the quadratic inequality 2λ2(G) 2 + 2λ2(G) − 1 2(n − 1) ≥ 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (142) Its solution gives (34) (n > 1 as otherwise, G = K1, but G is by assumption non-complete).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, it also follows that λn(G) = λn(G) (143) = −1 − λ2(G) (144) ≤ −1 − 1 2(√n − 1) (145) = − 1 2(√n + 1), (146) where (143) holds since G is (by assumption) self-complementary;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (144) holds since G is regular, and (145) holds by (34).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proof of Item (d): If G is self-complementary and strongly regular, then in light of Items (b) and (c) here, both inequalities in (34) and (35) hold with equality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3) Proof of Corollary 4: Let {Gℓ}ℓ∈N be a sequence of regular graphs where Gℓ is dℓ-regular of order nℓ, such that nℓ → ∞ and dℓ nℓ → 0 as we let ℓ tend to infinity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, lim sup ℓ→∞ ω(Gℓ) ≤ lim sup ℓ→∞ θ(Gℓ) (147) ≤ lim sup ℓ→∞ nℓ � 1 + λ2(Gℓ) � nℓ − dℓ + λ2(Gℓ) (148) = 1 + lim sup ℓ→∞ λ2(Gℓ), (149) 34 where (147) holds by the leftmost inequality in (16);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (148) holds by the rightmost inequality in (25);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (149) holds by the assumption that nℓ → ∞, and since the eigenvalues of Gℓ are bounded (in absolute value) by dℓ with lim ℓ→∞ dℓ nℓ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This leads to inequality (40), by a floor operation in the right-hand side of (149), since clique numbers are integers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next prove inequality (41).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For any graph G with n vertices, α(G) χ(G) ≥ n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (150) (This well-known inequality holds since the independence number α(G) denotes the size of a largest independent set in G, and in coloring the vertices in G with χ(G) colors, all color classes are independent).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Additionally, ω(G) = α(G), so ω(G) χ(G) ≥ n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (151) This gives lim inf ℓ→∞ χ(Gℓ) nℓ ≥ lim inf ℓ→∞ 1 ω(Gℓ) (152) = 1 lim sup ℓ→∞ ω(Gℓ) (153) ≥ 1 1 + lim sup ℓ→∞ ⌊λ2(Gℓ)⌋ , (154) where (152) holds by (151);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (153) is trivial, and (154) holds by (40).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4) Proof of Corollary 5: Inequalities (43) and (45) readily follow from Corollary 4 since if {Gℓ}ℓ∈N is a sequence of connected Ramanujan d-regular graphs (d is a fixed degree of the vertices), then (by definition) λ2(Gℓ) ≤ 2 √ d − 1, (155) for all ℓ ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By the assumption that the graph Gℓ has order nℓ with lim ℓ→∞nℓ = ∞, inequalities (43) and (45) are obtained by combining, respectively, (40) and (41) with (155).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (44) is obtained by combining the leftmost inequality in (24) with (155).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Indeed, since |λ2(Gℓ)| ≤ d (where the degree of the vertices of Gℓ is, by assumption, equal to a fixed value d), it follows that θ(Gℓ) ≥ nℓ − 2d 1 + λ2(Gℓ) ≥ nℓ − 2d 1 + 2 √ d − 1, (156) which then yields (44).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 5) Proof of Corollary 6: Let G be a connected Ramanujan d-regular graph with n vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' If G = Kn is the complete graph, which is a Ramanujan (n−1)-regular graph, then inequality (46) clearly holds (with ω(G) = n).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Otherwise, if G is non-complete, then combining the leftmost inequality in (16) and the rightmost inequality in (25) gives that ω(G) ≤ � n � 1 + λ2(G) � n − d + λ2(G) � , (157) where the floor operation in the right-hand side of (157) is enabled because ω(G) is an integer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since 0 ≤ λ2(G) ≤ 2 √ d − 1 for a non-complete and d-regular connected Ramanujan graph G, and since the function f1 : (−1, ∞) → (0, ∞) that is given by f1(x) ≜ n(1 + x) n − d + x, x > −1, (158) 35 is monotonically increasing, inequality (46) then follows from (157) and the monotonicity of the function f1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (48) follows from (46), (151), and since the chromatic number is an integer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (47) holds with equality if G = Kn (both sides are equal to 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Otherwise, inequality (47) holds by the leftmost inequality in (24), since 0 ≤ λ2(G) ≤ 2 √ d − 1 for a non- complete and d-regular Ramanujan graph G, and also since the function f2 : (−1, ∞) → (0, ∞) that is given by f2(x) ≜ n − d + x 1 + x , x > −1, (159) is monotonically decreasing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proofs for Section III-C 1) Proof of Proposition 2: Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be regular graphs such that, for all ℓ ∈ [k], the graph Gℓ is dℓ-regular with nℓ vertices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G = G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk be the strong product of these k regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next prove the two items of Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (a) By the leftmost inequality in (24), unless G = Kn, θ(G) ≥ n(G) − d(G) + λ2(G) 1 + λ2(G) , (160) where n(G) and d(G) denote, respectively, the order and valency of the strong product, which is a regular graph (since, by assumption, each factor is regular).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The following equalities hold as a result of the strong product operation: n(G) = k � ℓ=1 nℓ, (161) d(G) = k � ℓ=1 (1 + dℓ) − 1, (162) θ(G) = k � ℓ=1 θ(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (163) Indeed, equality (161) holds since the cardinality of a Cartesian product of finite sets is equal to the product of the cardinalities of each set;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' equality (162) can be justified by first verifying the special case of a strong product of two regular graphs, and then proceeding by a mathematical induction on k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Finally, equality (163) holds by (17) (see [65, Theorem 7]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Combining the bound in (160) with equalities (161)–(163) gives k � ℓ=1 θ(Gℓ) ≥ 1 + k� ℓ=1 nℓ − k� ℓ=1 (1 + dℓ) 1 + λ2(G) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (164) Unless all Gℓ (with ℓ ∈ [k]) are complete graphs, the left-hand side of (164) is strictly larger than 1, and then rearrangement of the terms in (164) gives the lower bound on λ2(G) in (52).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Next, the possible loosening of the lower bound in the right-hand side of (52) to the lower bound in the right-hand side of (53) holds by (19) (see [65, Theorem 9]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (53) holds with equality if each regular factor Gℓ is either edge-transitive (by [65, Theorem 9]) or strongly regular (by Item (a) of Proposition 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 36 (b) Combining (19) with equalities (161)–(163) gives, with n = n(G) and d = d(G), k � ℓ=1 θ(Gℓ) = θ(G) (165) ≤ − nλn(G) d − λn(G) (166) = − k� ℓ=1 nℓ · λn(G) k� ℓ=1 (1 + dℓ) − 1 − λn(G) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (167) Unless all Gℓ (with ℓ ∈ [k]) are empty graphs, the denominator in the right-hand side of (167) is strictly positive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This gives (54) after rearrangement of terms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Finally, the transition from (54) to (55) is justified if θ(Gℓ) = − nℓλmin(Gℓ) dℓ − λmin(Gℓ), ∀ ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (168) As above (the end of the proof of Item (a)), the condition in (168) holds if the regular graph Gℓ is either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2) Proof of Corollary 7: Let G a d-regular graph of order n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The lower bound on the second- largest eigenvalue λ2(G⊠ k) in the right-hand side of (59) follows from (52) by setting there G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk to be all identical to G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The upper bound on the smallest eigenvalue λmin(G⊠ k) in the right-hand side of (60) follows in a similar way from (54).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3) Proof of Proposition 3: Let G be a non-empty and non-complete connected d-regular graph on n vertices, and let k ∈ N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, G⊠ k is a connected regular graph, which is non-complete and non-empty (so, its largest eigenvalue is of multiplicity 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By (59), λ2(G⊠ k) ≥ nk − (1 + d)k θ(G)k − 1 − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (169) In order to prove that the k-fold strong power G⊠ k is non-Ramanujan, it is sufficient to show that the lower bound on its second-largest eigenvalue in the right-hand side of (169) is larger than 2√dk − 1;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' here, dk = (1+d)k −1 is the valency of the considered strong power (composed of k factors, where each factor is the d-regular graph G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Since G is d-regular and non-complete (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', d < n − 1 and θ(G) > 1), the right-hand side of (169) scales asymptotically like � n θ(G) �k (for a sufficiently large k), whereas the expression 2√dk − 1 scales asymptotically like 2(1 + d) k 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Comparing these two exponents gives that if θ(G) < n √ 1 + d, (170) then the exponential growth rate of the right-hand side of (169) is larger than that one of 2√dk − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hence, for sufficiently large k, the strong power G⊠ k is a (highly) non-Ramanujan graph under the condition in (170).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' This means that there exists k0 ∈ N such that the strong power G⊠ k is non-Ramanujan for all k ≥ k0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next obtain an explicit value of such k0, which is not necessarily the smallest one, proving that such a valid value for k0 is given by (63).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' To that end, based on the above explanation, one needs to deal with the inequality nk − (1 + d)k θ(G)k − 1 − 1 > 2 � (1 + d)k − 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (171) 37 In order to obtain a closed-form solution, we strengthen the condition in (171) to nk − (1 + d)k θ(G)k − 1 ≥ 2(1 + d) k 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (172) Dividing both sides of (172) by (1 + d) k 2 gives � n √ 1 + d θ(G) �k − �√ 1 + d θ(G) �k − (1 + d)− k 2 ≥ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (173) Let k ≥ 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The condition imposed in (173) can be further strengthened to � n √ 1 + d θ(G) �k − �√ 1 + d θ(G) �k ≥ 2 + (1 + d)− 3 2 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (174) Since d < n − 1 for a non-complete d-regular graph of order n, for all k ≥ 3, � n √ 1 + d θ(G) �k − �√ 1 + d θ(G) �k = � n √ 1 + d θ(G) �k � 1 − �1 + d n �k� ≥ � n √ 1 + d θ(G) �k � 1 − �1 + d n �3� > 0, (175) which, by combining (174) and (175), gives the stronger condition n3 − (1 + d)3 n3 � n √ 1 + d θ(G) �k ≥ 2 + (1 + d)− 3 2 , (176) with k ≥ 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Solving inequality (176) implies that inequality (171) is satisfied for all k ≥ k0, with the closed-form expression of k0 in (63).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It therefore gives that if G is a d-regular graph on n vertices, which satisfies the condition in (170), then G⊠ k is non-Ramanujan for all k ≥ k0 (it becomes, in fact, a highly non-Ramanujan graph since both sides of inequality (171) have different exponential growth rates, so the condition for a Ramanujan graph is strongly violated for the strong power G⊠ k when the value of k is increased).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next specialize this result for graphs that are self-complementary and vertex-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For n = 1, the complete graph G = K1 is a self-complementary and vertex-transitive graph, whose all strong powers are also isomorphic to K1, so they are therefore non-Ramanujan graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Let G be a graph of order n > 1 that is self-complementary and vertex-transitive, so it is d-regular with d = 1 2(n − 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Additionally, for such a graph G, the Lov´asz θ-function is equal to θ(G) = √n, and it coincides with the Shannon capacity of G (see [65, Theorem 8] and [65, Theorem 12]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, n √ d + 1 = � 2n2 n + 1 > √n = θ(G), (177) so the required condition in Proposition 3 is fulfilled by graphs of order n that are self- complementary and vertex-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The value of k0 in (63) is specialized for such graphs to k0 = max \uf8f1 \uf8f2 \uf8f33, \uf8ee \uf8ef\uf8ef\uf8ef 2 log � 8n3 8n3−(n+1)3 � + 2 log � 2 + � 8 (n+1)3 � log � 2n n+1 � \uf8f9 \uf8fa\uf8fa\uf8fa \uf8fc \uf8fd \uf8fe (178) = \uf8f1 \uf8f4 \uf8f2 \uf8f4 \uf8f3 5 if n = 5, 4 if n = 9, 3 if n ≥ 13 with n ≡ 1 (mod 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (179) 38 The constraint on n in (179) is the necessary condition on n in Remark 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Proofs for Section III-D 1) Proof of Proposition 4: (a) Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be k simple, finite and undirected graphs, ��V(Gℓ) �� = nℓ for ℓ ∈ [k], and let G = G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We provide two alternative simple proofs of (65).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' First proof: χ(G) ≥ θ(G) (180) ≥ | V(G)| θ(G) (181) = k � ℓ=1 | V(Gℓ)| θ(Gℓ) , (182) where (180) holds by (16), (181) holds by (18), and equality (182) holds by (17) and since V(G) = V(G1) × .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' × V(Gk).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The ceiling operation can be add to the right-hand side of (182) since a chromatic number is an integer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Second proof: χ(G) ≥ | V(G)| α(G) (183) ≥ | V(G)| θ(G) (184) = k � ℓ=1 | V(Gℓ)| θ(Gℓ) , (185) where (183) holds by (150), (184) holds by (23), and (185) is (182).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' We next prove (66).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' χ(G) ≥ θ(G) (186) = k � ℓ=1 θ(Gℓ), (187) where (186) holds by (15), and (187) holds by (17).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (b) Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be regular graphs, where Gℓ is dℓ-regular of order nℓ for all ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (67) is (65).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (68) follows from (19) and (67).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Furthermore, by Item (a) in Proposition 1, inequality (68) holds with equality if each regular graph Gℓ, for ℓ ∈ [k], is either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (c) By (182), with | V(Gℓ)| = nℓ, | V(G)| θ(G) = k � ℓ=1 nℓ θ(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (188) Suppose that, for all ℓ ∈ [k], Gℓ is dℓ-regular, and it is also either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' By Item (a) in Proposition 1, for all ℓ ∈ [k], θ(Gℓ) = − nℓ λmin(Gℓ) dℓ − λmin(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (189) 39 Combining (188) and (189) gives | V(G)| θ(G) = k � ℓ=1 � 1 − dℓ λmin(Gℓ) � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (190) On the other hand, since G = G1 ⊠ .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ⊠ Gℓ is d-regular, with d ≜ d(G) as given in (70), it follows from (19) that θ(G) ≤ −| V(G)| λmin(G) d(G) − λmin(G).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (191) It should be noted, in regard to (191), that even if all Gℓ’s are regular and edge-transitive graphs, their strong product G is not necessarily edge-transitive.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' In fact, G is not edge- transitive, unless all the k factors {Gℓ}k ℓ=1 are complete graphs (see [47, Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' For this reason, (191) does not hold in general with equality (see [65, Theorem 9]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Finally, combing (190) and (191) gives inequality (69).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (d) Let G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk be regular graphs, where Gℓ is dℓ-regular on nℓ vertices for all ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, under the assumptions of Item (d), (1) k � ℓ=1 | V(Gℓ)| θ(Gℓ) = k � ℓ=1 θ(Gℓ) (192) ≥ k � ℓ=1 ω(Gℓ) (193) where (192) holds since, by assumption, each of the graphs G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk is vertex- transitive or a strongly regular graph (this is because [65, Theorem 8] and (27) provide different sufficient conditions for inequality (18) to hold with equality).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (193) holds by the leftmost inequality in (16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (2) k � ℓ=1 � 1 − dℓ λmin(Gℓ) � = k � ℓ=1 | V(Gℓ)| θ(Gℓ) (194) ≥ k � ℓ=1 ω(Gℓ) (195) where, by Item (a) of Proposition 1, equality (194) holds since (by assumption), for all ℓ ∈ [k], the graph Gℓ is either regular and edge-transitive, or a strongly regular graph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Inequality (195) holds under the same reasoning as of (192) and (193).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' To summarize, it shows that under proper assumptions, the lower bound on the chromatic number of G in the right-hand side of (65), or even its loosened bound in the right-hand side of (68), are larger than or equal to the lower bound k� ℓ=1 ω(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (e) By (66), χ(G) ≥ k � ℓ=1 θ(Gℓ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (196) 40 Let, for all ℓ ∈ [k], the graph Gℓ be dℓ-regular on nℓ vertices, and suppose that it is either edge-transitive or strongly regular.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Then, by Item (a) of Proposition 1, θ(Gℓ) = − nℓ λmin(Gℓ) dℓ − λmin(Gℓ), ∀ ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (197) Combining (196) and (197), followed by taking a ceiling operation on the lower bound on the chromatic number χ(G), gives (71).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (f) By the assumption that G1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' , Gk are self-complementary, θ(Gℓ) = θ(Gℓ), ∀ ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (198) Furthermore, by the assumption that for all ℓ ∈ [k], Gℓ is a graph on nℓ vertices that is either vertex-transitive or strongly regular, θ(Gℓ) θ(Gℓ) = nℓ, ∀ ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (199) Combining (198) and (199) gives θ(Gℓ) = √nℓ, ∀ ℓ ∈ [k].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (200) Consequently, by (65) and (200), χ(G) ≥ � k � ℓ=1 nℓ θ(Gℓ) � (201) = � k � ℓ=1 √nℓ � (202) = �√n � , (203) and, from (66) and (200), χ(G) ≥ � k � ℓ=1 θ(Gℓ) � (204) = � k � ℓ=1 √nℓ � (205) = �√n � .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' (206) This proves (72) and (73), and it completes the proof of Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2) Proof of Corollary 8: The rightmost inequality in (78) is a well-known upper bound on the chromatic number of strong products (see [11], [15, Theorem 3]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' The leftmost inequality in (78) gives a lower bound on the chromatic number of a strong product of (not necessarily distinct) non-complete, and strongly regular graphs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' It readily follows by combining equality (26) in Corollary 1, together with inequality (65) in Proposition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Finally, by Part 1 of Item (d) in Proposition 4, the leftmost term in (78) is larger than or equal to the product of the clique numbers of {Gℓ}k ℓ=1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ACKNOWLEDGMENTS The author wishes to acknowledge the two anonymous reviewers for helpful and timely reports.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 41 REFERENCES [1] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Abbe, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Boix-Adser´a, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Ralli and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sandon, “Graph powering and spectral robustness,” SIAM Journal on Mathematics of Data Science, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 132–157, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [2] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Abbe and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Ralli, “An Alon–Boppana theorem for powered graphs, and generalized Ramanujan graphs,” preprint, June 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [Online] https://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='org/abs/2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='11248v1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [3] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Abbe and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Ralli, “An Alon–Boppana theorem for powered graphs, generalized Ramanujan graphs and robust community detection,” Proceedings of the 2022 International Zurich Seminar on Information and Communication, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 19–23, Zurich, Switzerland, March 2–4, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [4] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Ac´ın, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Duanc, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Roberson, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sainz and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Winter, “A new property of the Lov´asz number and duality relations between graph parameters,” Discrete Applied Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 216, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 489–501, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [5] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Aigner and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Ziegler, Proofs from THE BOOK, Sixth Edition, Springer, Berlin, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [6] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Alon, “Eigenvalues and expanders,” Combinatorica, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 6, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 83–96, June 1986.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [7] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Alon, “Graph powers,” Contemporary Combinatorics, Bolyai Society Mathematical Studies and Springer, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 10 (B´ela Bollob´as Editor), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 11–28, Budapest, Hungary, 2002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [8] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Alon and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Spencer, The Probabilistic Method, Fourth Edition, John Wiley & Sons, Hoboken, New Jersey, USA, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [9] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Alon, “Lov´asz, vectors, graphs and codes,” Building Bridges II - Mathematics of L´aszl´o Lov´asz, Bolyai Society Mathematical Studies and Springer, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 28 (I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' B´ar´any, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Katona and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sali Editors), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–16, Budapest, Hungary, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [10] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Alon and E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lubetzky, “The Shannon capacity of a graph and the independence numbers of its powers,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 52, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2172–2176, May 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [11] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Berge, Graphs and Hypergraphs, North-Holland Mathematical Library, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 6, 1973.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [12] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Bohman and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Holzman, “A nontrivial lower bound on the Shannon capacities of the complements of odd cycles,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 49, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 721–722, March 2003.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [13] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Bohman, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Holzman, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Natarajan, “Maximum independent sets in certain powers of odd cycles,” The Electronic Journal of Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 16, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, paper 26, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–8, July 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [14] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Bohman, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Holzman, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Natarajan, “On the independence numbers of the cubes of odd cycles,” The Electronic Journal of Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 20, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, paper 10, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–19, August 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [15] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Borowiecki, “On chromatic number of products of two graphs,” Colloquium Mathematicum, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 25, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 49–52, 1972.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [16] V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Brimkov, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Codenotti, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Crespi, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Leoncini, “Efficient computation of the Lov´asz number of certain circulant graphs,” in Graph-Theoretic Concepts in Computer Science, LNCS 3353, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hrommkoviˇc, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Nagl and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Westfechtel, eds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=', Springer, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 285–295, Berlin, Germany, 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [17] V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Brimkov, “Algorithmic and explicit determination of the Lov´asz number for certain circulant graphs,” Discrete Applied Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 155, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1812–1825, 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [18] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Brouwer and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Haemers, Spectra of Graphs, Springer, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [19] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Brouwer and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Van Maldeghem, Strongly Regular Graphs, Cambridge University Press, (Encyclopedia of Mathematics and its Applications, Series Number 182), 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [20] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Chartrand, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lesniak, and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Zhang, Graphs and Digraphs, Sixth Edition, CRC Press, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [21] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cioabˇa, “On the extreme eigenvalues of regular graphs,” Journal of Combinatorial Theory, Series B, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 96, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 367–373, 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [22] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cioabˇa, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Murty, “Expander graphs and gaps between primes,” Forum Mathematicum, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 20, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 745–756, 2008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [23] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cioabˇa, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Elzinga and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Gregory, “Some observations on the smallest adjacency eigenvalue of a graph,” Discussiones Mathematicae Graph Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 40, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 467–493, May 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 42 [24] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cioabˇa and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Gupta, “A lower bound for the smallest eigenvalue of a graph and an application to the associahedron graph,” preprint, October 18, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [Online] https://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='org/abs/2210.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='08516.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [25] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cioabˇa, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Murty, A First Course in Graph Theory and Combinatorics, Second Edition, Springer, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [26] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cvetkovi´c, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Rowlinson, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Simi´c, An Introduction to the Theory of Graph Spectra, London Mathematical Society Student Texts 75, Cambridge University Press, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [27] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Desai and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Rao, “A characterization of the smallest eigenvalue of a graph,” Journal of Graph Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 18, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 181–194, March 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [28] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Dowling, Expander Graphs and Coding Theory, Ph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' dissertation, Clemson University, South Carolina, USA, September 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [29] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Erd¨os, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' McEliece and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Taylor, “Ramsey bounds for graph products,” Pacific Journal of Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 37, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 45–46, January 1971.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [30] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Esperet and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Wood, “Colouring strong products,” preprint, May 10, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [Online] https://arxiv.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='org/abs/2205.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='04953.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [31] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Feigenbaum and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sch¨affer, “Finding the prime factors of strong direct product graphs in polynomial time,” Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 109, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 77–102, November 1992.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [32] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Filmus, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Golubev and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lifshitz, “High dimensional Hoffman bound and applications in extremal combinatorics,” Journal of Algebraic Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 6, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1005–1026, November 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [33] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Friedman, “On the second eigenvalue and random walks in random d-regular graphs,” Combina- torica, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 11, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 331–362, December 1991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [34] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Friedman, “Some geometric aspects of graphs and their eigenfunctions,” Duke Mathematical Journal, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 69, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 487–525, March 1993.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [35] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Friedman, A Proof of Alon’s Second Eigenvalue Conjecture and Related Problems, Memoirs of the American Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 195, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 910, 2008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [36] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Geetha and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Somasundaram, “Total colorings of product graphs,” Graphs and Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 34, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 339–347, February 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [37] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Godsil and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Royle, Algebraic Graph Theory, Springer, New York, 2001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [38] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Godsil, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Roberson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' S´amal and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Severini, “Sabidussi versus Hedetniemi for three variations of the chromatic number,” Combinatorica, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 36, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 395–415, August 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [39] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Guo and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Watanabe, “On graphs in which the Shannon capacity is unachievable by finite product,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 36, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 622–623, May 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [40] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Griggs, “Lower bounds on the independence number in terms of the degrees,” Journal of Combinatorial Theory, Series B, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 34, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 22–39, February 1983.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [41] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Gr¨otschel, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lov´asz, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Schrijver, “The ellipsoid method and its consequences in combinatorial optimization,” Combinatorica, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 168-197, June 1981.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [42] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Haemers, Eigenvalue Techniques in Design and Graph Theory, Ph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' dissertation, Stichting Mathematisch Centrum, Amsterdam, 1979.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [43] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Haemers, “On some problems of Lov´asz concerning the Shannon capacity of a graph,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 25, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 231–232, March 1979.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [44] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Haemers, “Hoffman’s ratio bound,” Linear Algebra and its Applications, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 617, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 215–219, May 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [45] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hales, “Numerical invariants and the strong product of graphs,” Journal of Combinatorial Theory, Series B, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 15, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 146–155, October 1973.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [46] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hammack, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Imrich and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Klavˇzar, Handbook of Product Graphs, Second Edition, CRC Press, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [47] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hammack, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Imrich and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Klavˇzar, “Edge-transitive products,” Journal of Algebraic Combi- natorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 43, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 837–850, June 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 43 [48] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hoffman, “On eigenvalues and colorings of graphs,” in Graph Theory and its Applications (B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Harris, ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' ), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 79–91, Academic Press, New York, 1970.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [49] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hong, “Bounds on eigenvalues of graphs,” Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 123, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 65–74, 1993.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [50] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hoory, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Linial and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Wigderson, “Expander graphs and their applications,” Bulletin of the American Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 43, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 439–561, October 2006.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [51] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hu, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Tamo, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Shayevitz, “A bound on the Shannon capacity via a linear programming variation,” SIAM Journal on Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 32, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2229–2241, September 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [52] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Jurkiewicz, “A survey on known values and bounds on the Shannon capacity,” Selected Topics in Modern Mathematics, ed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Gancarzewicz and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Skrzy´nski, Publishing House AKAPIT, Krak´ow, Poland, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 115–128, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [53] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Kahale, “Eigenvalues and expansion of regular graphs,” Journal of the Association for Computing Machinery, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 42, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1091–1106, September 1995.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [54] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Klavˇzar, “Strong products of χ-critical graphs,” Aequationes Mathematicae, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 45, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 153–162, April 1993.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [55] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Klavˇzar, “Coloring graph products - a survey,” Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 155, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 135– 145, August 1996.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [56] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Klavˇzar and U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Mulitinovi´c, “Strong products of Kneser graphs,” Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 133, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 287–300, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [57] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Knuth, “The sandwich theorem,” Electronic Journal of Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–48, 1994.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [58] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' K¨orner and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Orlitsky, “Zero-error information theory,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 44, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 6, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2207–2229, October 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [59] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Krivelevich, “Expanders – how to find them, and what to find in them,” Surveys in Combinatorics 2019, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 115–142, June 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [60] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Li, “On finite graphs that are self-complementary and vertex-transitive,” Australasian Journal of Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 18, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='147-155, 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [61] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Li (with an appendix by J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Serre), “On negative eigenvalues of regular graphs,” Comptes Rendus de l’Acad´emie des Sciences - Series I - Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 333, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 10, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 907–912, November 2001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [62] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Li and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Rao, “Self-complementary vertex-transitive graphs of order a product of two primes,” Bulletin of the Australian Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 89, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 322–330, April 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [63] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Li, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Rao, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Song, “New constructions of self-complementary Cayley graphs,” Journal of the Australian Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 111, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 372–385, December 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [64] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Li-xin, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Feng and Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hai-xing, “On the transitivity of the strong product of graphs,” Chinese Quarterly Journal of Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 30, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 620–623, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [65] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lov´asz, “On the Shannon capacity of a graph,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 25, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–7, January 1979.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [66] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lov´asz, An Algorithmic Theory of Numbers, Graphs and Convexity, Philadelphia, PA, USA: SIAM, 1986.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [67] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lov´asz, Graphs and Geometry, American Mathematical Society, Colloquium Publications, volume 65, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [68] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lubotzky, “Expander graphs in pure and applied mathematics,” Bulletin of the American Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 49, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 113–162, January 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [69] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Marcus, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Spielman, and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Srivastava, “Interlacing families I: Bipartite Ramanujan graphs of all degrees,” Annals of Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 182, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 307–325, July 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [70] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' McEliece and E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Posner, “Hide and seek, data storage, and entropy,” Annals of Mathematical Statistics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 42, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1706–1716, October 1971.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [71] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Morris, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Praeger and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Spiga, “Strongly regular edge-transitive graphs,” Ars Mathematica Contemporanea, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 137-155, September 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 44 [72] N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Mullin, Self-Complementary Arc-Transitive Graphs and Their Imposters, Master’s thesis, University of Waterloo, Canada, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [73] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Murty, “Ramanujan graphs: an introduction,” Indian Journal of Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 6, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 91–127, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [74] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Neumaier, “Cliques and claws in edge-transitive strongly regular graphs,” Mathematische Zeitschrift, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 174, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 197–202, October 1980.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [75] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Neumaier, “Regular cliques in graphs and special 1 1 2 designs,” Finite Geometries and Designs (editors: P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Cameron, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hirschfeld, and D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Hughes), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 244–259, 1981.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [76] V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Nikiforov, “Eigenvalue problems of Nordhaus–Gaddum type,” Discrete Mathematics, vol 307, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 6, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 774–780, March 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [77] V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Nikiforov and X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Yuan, “More eigenvalue problems of Nordhaus–Gaddum type,” Linear Algebra and its Applications, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 451, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 231–245, June 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [78] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Nilli, “On the second eigenvalue of a graph,” Discrete Mathematics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 91, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 207–210, August 1991.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [79] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Nilli, “Tight estimates for eigenvalues of regular graphs,” Electronic Journal of Combinatorics, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 11, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, paper N9, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1–4, May 2004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [80] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Nordhaus and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Gaddum, “On complementary graphs,” American Mathematical Monthly, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 63, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 175–177, March 1956.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [81] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Peisert, “All self-complementary symmetric graphs,” Journal of Algebra, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 240, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 209– 229, June 2001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [82] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Rao, “Self-complementary vertex-transitive graphs,” PhD dissertation, the University of Western Australia, September 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [83] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Rao, “Self-complementary vertex-transitive graphs,” Bulletin of the Australian Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 94, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 165-–166, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [84] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sabidussi, “Graph multiplication,” Mathematische Zeitschrift, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 72, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 446–457, 1960.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [85] The Sage Developers, SageMath, the Sage Mathematics Software System (Version 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content='2), 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [86] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Shannon, “The zero error capacity of a noisy channel,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 8–19, September 1956.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [87] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Sonnemann and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Krafft, “Independence numbers of product graphs,” Journal of Combinatorial Theory Series B, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 17, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 133–142, October 1974.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [88] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Spielman, “Spectral graph theory,” Chapter 18 (pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 495–524) in Combinatorial Scientific Computing (U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Naumann and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Schenk editors), Chapman & Hall CRC Press, 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [89] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Stani´c, Inequalities for Graph Eigenvalues, London Mathematical Society Lecture Note Series, Series Number 423, Cambridge University Press, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [90] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Vesztergombi, “Some remarks on the chromatic number of the strong product of graphs,” Acta Cybernetica, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 207–212, January 1979.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [91] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Vesztergombi, “Chromatic number of strong product of graphs,” in Algebraic Methods in Graph Theory (editors: V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' S´os and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lov´asz), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 819–825, Elsevier, 1981.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [92] V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Wei, “A lower bound on the stability number of a simple graph,” Bell Laboratories Technical Memorandum, 81–11217–9, Murray Hill, New Jersey, 1981.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [93] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Wilf, “The eigenvalues of a graph and its chromatic number,” Journal of the London Mathematical Society, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' s1–42, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 330–332, January 1967.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [94] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Witsenhausen, “The zero-error side information problem and chromatic numbers,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 22, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 5, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 592–593, September 1976.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [95] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Xu and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Radziszowski, “Bounds on Shannon capacity and Ramsey numbers from product of graphs,” IEEE Transactions on Information Theory, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 59, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 8, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 4767–4770, August 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' [96] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Zhai, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Lin and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' Wang, “Sharp upper bounds on the second largest eigenvalues of connected graphs,” Linear Algebra and Its Applications, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 437, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'} +page_content=' 236–241, July 2012.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/y9E0T4oBgHgl3EQf_AIq/content/2301.02820v1.pdf'}